Paper - 11

Renewable Energy 145 (2020) 1543e1571
Contents lists available at ScienceDirect
Renewable Energy
journal homepage: www.elsevier.com/locate/renene
Review
Uncertainty models for stochastic optimization in renewable energy

applications
A. Zakaria a, Firas B. Ismail a, M.S. Hossain Lipu b, M.A. Hannan c, *
a
Power Generation Unit, Institute of Power Engineering, Universiti Tenaga Nasional, 43000 Kajang, Malaysia
b
Centre for Integrated Systems Engineering and Advanced Technologies (Integra), Faculty of Engineering and Built Environment, Universiti Kebangsaan
Malaysia, 43600 Bangi, Malaysia
c
Department of Electrical Power Engineering, College of Engineering, Universiti Tenaga Nasional, 43000 Kajang, Malaysia
a r t i c l e i n f o a b s t r a c t
Article history: With the rapid surge of renewable energy integrations into the electrical grid, the main questions
Received 8 March 2019 remain; how do we manage and operate optimally these surges of fluctuating resources? However, vast
Received in revised form optimization approaches in renewable energy applications have been widely used hitherto to aid
7 July 2019
decision-makings in mitigating the limitations of computations. This paper comprehensively reviews the
Accepted 15 July 2019
Available online 16 July 2019
generic steps of stochastic optimizations in renewable energy applications, from the modelling of the
uncertainties and sampling of relevant information, respectively. Furthermore, the benefits and draw-
backs of the stochastic optimization methods are highlighted. Moreover, notable optimization methods
Keywords:
Stochastic optimizations
pertaining to the steps of stochastic optimizations are highlighted. The aim of the paper is to introduce
Uncertainty model the recent advancements and notable stochastic methods and trending of the methods going into the
Scenario generations future of renewable energy applications. Relevant future research areas are identified to support the
Renewable energy applications transition of stochastic optimizations from the traditional deterministic approaches. We concluded based
on the surveyed literatures that the stochastic optimization methods almost always outperform the
deterministic optimization methods in terms of social, technical, and economic aspects of renewable
energy systems. Thus, this review will catalyse the effort in advancing the research of stochastic opti-
mization methods within the scopes of renewable energy applications.
© 2019 Elsevier Ltd. All rights reserved.
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1544
2. Overview of stochastic optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1544
3. Uncertainty modelling in stochastic optimization approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1545
3.1. Monte Carlo Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1545
3.1.1. Types of Monte Carlo Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1545
3.1.2. Recent MCS applications in renewable energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1546
3.2. Notable uncertainty modelling method: Generative Adversarial Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1549
4. Sampling methods in scenario generations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1550
4.1. Importance Sampling (IS) method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1551
4.1.1. Type of IS method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1551
4.1.2. IS method implementations in renewable energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1551
4.2. Notable sampling method: Markov Chain Monte Carlo method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1554
4.2.1. Overview of MCMC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1554
4.2.2. MCMC sampling procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1554
5. Stochastic optimization methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1556
5.1. Stochastic programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1556
* Corresponding author.
E-mail address: [email protected] (M.A. Hannan).
https://doi.org/10.1016/j.renene.2019.07.081
0960-1481/© 2019 Elsevier Ltd. All rights reserved.
1544 A. Zakaria et al. / Renewable Energy 145 (2020) 1543e1571
5.1.1. Two e stage models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1556

5.1.2. Multi e stage models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1558
5.2. Approximate stochastic dynamic programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1560
5.2.1. Model predictive control (MPC) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1561
5.2.2. Notable overview of MPC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1562
5.2.3. Stochastic MPC implementations in renewable energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1562
5.2.4. MPC's comparison and future trending . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1567
6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1567
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1568
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1568
1. Introduction from huge computational expenses due to large number of sce-

narios that needs to be considered in its calculations [10].
Renewable energy sectors have seen tremendous growth in the Numerous techniques have been developed by many authors in
last decade throughout the world especially in Northern America, increasing the stochastic optimizations' efficacy to reduce compu-
Western Europe, and China accounting for almost half of the tational expenses [11]. Nonetheless, it is still computationally
expansion [1]. The recent rapid energy shift in these parts of the demanding and suffers from the ‘curse of dimensionality’ in cases
world are mainly due to the reduction of production costs of the of assessing the problems over multivariate and multiple periods of
renewable energy generators, the drive to reduce carbon emissions, time intervals. Despite the advantages of the stochastic optimiza-
and attractive tariffs offered [2]. Wind energy and solar energy tions, its implementations in renewable energy applications are
accounts for the most rapid growth in renewable energy genera- still relatively new. The problems in its transparencies, computa-
tions with an approximate 77% of new capacity, with hydropower tional efficacies, and their full practical implementations are still
dominating the rest [3]. Despite being a clean and abundantly being addressed by system operators and other interested parties.
available energy (in some parts of the world), renewable energy Based on vast relevant surveys conducted, the paper's motiva-
resources still suffer from its lack of energy density and its inter- tion is to analyse recent and notable stochastic optimization
mittency [4]. The latter presents the most challenge to researchers methods in the lights of renewable energy applications while
in terms of successfully predicting and utilizing the usage and identifying its current and future research directions. We also
control of renewable energy resources. In contrast to the conven- pointed out various advantages and disadvantages of highlighted
tional generators (i.e. coal or steam turbine power plants), renew- stochastic optimization methods. Given the vastness of stochastic
able energy generators can only generate energy, when there are optimization methods that exist hitherto, the focus of the paper is
renewable resources available. Therefore, appropriate prediction, to provide a basic introduction to the highlighted methods while
control, and precise representation of renewable energy systems directing interested readers towards notable works of other au-
play an important role to ensure stable and uninterrupted energy thors mainly in the field of renewable energy applications.
supply. Optimization of renewable energy systems can be accu-
rately solved if uncertainties, probabilities, and fluctuating behav- 2. Overview of stochastic optimization
iours of renewable energy systems are being properly represented
[5]. The general stochastic optimization in renewable energy ap-
The current wave of optimization approach in renewable energy plications is broken down into several steps as summarized in Fig. 1.
applications are shifting. The first wave was in the form of deter- The paper highlights these steps and focuses on the notable sto-
ministic approaches [6]. During this wave, mixed integer pro- chastic methods in recent renewable energy applications. As stated,
gramming has stood out from the earlier modelling approaches the idea of the paper is to provide new researchers as well as
namely; dynamic programming, priority list, Lagrangian Relaxa- advanced readers in the optimization field with insights on the
tion, etc. However, deterministic methods with an assumption of recent and notable stochastic optimization methods in renewable
perfect information produced idealistic results which contradicted energy applications. The paper is focused on the intuitive part of
with the core value of renewable energy systems. With the fluc- the stochastic optimization methods rather than the mathematical
tuations of renewable resources, varying demands, and intermit- discourses of the field. Readers are also exposed to the recent
tent economic parameters, deterministic approaches alone could trending of the stochastic optimizations in renewable energy ap-
not fully capture the dynamics of the whole renewable energy plications as well as future works and relevant research themes in
systems [7]. these areas. From these recent works, gaps and future works from
Studies are now moving towards stochastic optimization in the literatures are analysed. The trending from the surveyed recent
which the optimization considers uncertainties and probabilities as literatures is highlighted and the efficacy of the stochastic optimi-
inputs, then evaluate its influence on the output of the system [8]. A zation approaches is presented from the main results of the
stochastic optimization then utilizes these scenario uncertainties in literatures.
its objective function's formulations. Hitherto, vast amount of lit- We consider only the most recent literatures in stochastic
eratures has been found regarding the stochastic optimization optimization methods in the field of renewable energy applica-
techniques [9]. Stochastic optimization provides a range of possible tions. Key aspects pertaining to the stochastic optimizations are
solutions which models closer to real e world situations that would featured such as its main results and contributions, its research
benefit operators/consumers in assessing the risks involved in the gaps, and its uncertainty parameters. The notable mentioned
uncertainties of renewable energy generations. Therefore, the methods are chosen based on its contributions in the field as well as
characteristics of stochastic optimization methods are more suit- its future implementation prospects. Past authoritative works are
able in handling renewable energy system's fluctuating and inter- also highlighted for interested readers to research further. In this
mittent nature. Stochastic optimization however, generally suffers paper, within the scope of renewable energy applications,
A. Zakaria et al. / Renewable Energy 145 (2020) 1543e1571 1545
No
Optimization/
Stochastic Decomposition of
Optimization samples generated Max iteration?
using stochastic
models
Yes
Parallel Scenario / Scenario
Tree generations / Chance
Sampling of
Constraints Scenario / Output of possible
relevant scenario
Forecasts of renewable data best solutions
generations
with varying PDF
(Uncertainty Modelling)
Fig. 1. Stochastic optimization flowchart.
uncertainty modelling/scenario generation approaches are initially or variables involved. Fig. 2 shows an overview of the uncertainties’
addressed, followed by notable sampling methods to capture modelling approaches. The scope of the paper is only limited to the
relevant scenarios in stochastic optimizations. Next, the stochastic numerical method of the uncertainty modelling approaches,
optimization approach is highlighted mainly in the lights of mainly in the recent Monte Carlo Simulation (MCS) approach in
approximate stochastic dynamic programming. The paper con- renewable energy applications. Interested readers are encouraged
cludes with main issues and challenges of stochastic optimization to read the works made by Refs. [11e13] for further information
approaches in renewable energy applications, followed by its crit- regarding other uncertainty modelling approaches.
ical remarks and future relevant themes in renewable energy
applications. 3.1. Monte Carlo Simulation
3. Uncertainty modelling in stochastic optimization approach MCS is one of the most used methods in the probabilistic un-
certainty modelling approach [15,16]. Historical probability distri-
In stochastic optimizations, representing the correct un- bution function, forecasting errors, and market variability are the
certainties are critical. Each uncertainty modelling technique parameters that can be utilized by the MCS method to learn and
would yield a different representation of the systems. Therefore, populate the scenarios’ generations. The MCS method is favoured
appropriate selection of uncertainty modelling methods is crucial. due to its ability to systematically sample from random processes
Uncertainty modelling is a typical way to represent the stochas- [16]. Furthermore, a transfer function is not necessarily needed in
ticity of renewables' systems. Instead of assuming perfect knowl- MCS. The problem can be treated as a black box system which can
edge of the parameters (i.e. wind speed, solar irradiation and load yield related output with given samples of inputs. MCS is also
demand) as opposed to a deterministic approach, random distri- intuitive and relatively easy to implement. MCS can also be
butions are added as inputs to a stochastic optimization approach implemented in non e differentiable as well as non e convex
to mimic the probabilistic characteristics of a renewable energy problems. Apart from that, it supports all probabilistic distribution
system. In representing the uncertainties, it is critical that the function (PDF) types. Regardless, MCS has some deficiencies issue
distribution dynamics of the scenarios are well captured. One of the such as expensive computation due to its iterative behaviour,
ways to do that is by generating large number of scenarios, where especially when the degrees of freedom and the space search ex-
each scenario would capture the possible realization of the un- pands [17]. The general MCS method in renewable energy appli-
derlying uncertainties. The idea is to find the close approximation cations can be described in Fig. 3.
of the uncertainties' true distributions. In other words, the main
goal is to infer a probability distribution(s) of an output(s) based on 3.1.1. Types of Monte Carlo Simulation
a given probability distribution function(s) (PDF) of an assumed According to Ref. [18], MCS method is typically divided in three
known input(s). PDF distributions of inputs varies from parameters different types. The first one is called the Sequential e MCS method.
Fig. 2. Uncertainty modelling overview.

determine the optimal amount of power generation to be

Start
committed by incorporating renewable power forecasting errors
and system reliability. A study by Ref. [22] proposed a sliding time
Initialize MCS Input PDF model of window optimization approach to find an optimal design and
parameters renewable and load dispatch scheduling strategy in a hybrid renewable energy system
consisting of biomass, wind, solar, gas e fired boiler, battery, and
thermal energy storage. Different scenarios were created to find out
Output PDF which combinations of renewable generators would yield the op-
Generate statistic data
parameters timum design and dispatch strategy. MCS was then tasked to
using MCS
calculations generate the cost of energy (COE) distribution as an output to
provide a risk indication of the chosen design. MCS has been uti-
lized by Ref. [23] to sample different system states from the self e
Constraints
No satisfied?
No adapted evolutionary strategy (SAES) combined with Fischer-
Burmeister algorithm in optimizing the one e time investment
and the operational costs of hybrid wind e energy storage power
Yes Record results system. Lopes et al. (2015) have addressed the impact and system
reliability on the combination of wind generation and small hydro
Max
iteration?
power plants [24].
The uncertainties of power generations are modelled using the
Obtain PDF of output
Yes MCS. A novel risk management method was investigated in the
parameters’ results
work of [25] based on managed charging of plug e in hybrid electric
vehicle (PHEV) and vehicle e to e grid (V2G) using MCS. MCS
analysis of wind farm lightning surge transients aided by a light-
Stop ning detection network data is implemented by Ref. [26] to produce
a statistical depiction of over e voltages distribution within the
Fig. 3. MCS Flowchart in renewable energy applications. wind farm electrical network. The statistic depiction can be used to
assist wind farm lightning risk management and surge protection
optimization. MCS has been implemented by Ref. [27] to consider
This method represents the uncertainties in chronological order the uncertainties of load and irradiation in the economic optimi-
and is valued for its flexibility for assessing the reliability of the zation of energy supply at off e grid healthcare facilities. In the
system's posterior distribution. This method is suitable in the ap- work of [28], the authors have utilized MCS by performing Tem-
plications of time series of variable energy sources and variable perature e Augmented Probabilistic Load Flow (TPLF) to charac-
load. Nonetheless, it requires a significant amount of computational terize the aspects of over e limit probabilities of events such as over
effort as the dimensions of the uncertainties grow due to sequential and under e loading of loads and voltages in a 39 e bus test system.
and iterative characteristics of this MCS. This method also may be
infeasible for some applications which are non e sequential. The 3.1.2.2. Pseudo e sequential MCS applications in renewable energy.
second type of the MCS is the Pseudo e Sequential MCS. It is named The MCS method handled the uncertainties which are; the solar
due to its ability of non e sequential sampling of system states and irradiance which is modelled using kernel density estimation, the
chronological simulation of only the sub e sequences tied with the load demand using a Gaussian distribution, the wind speed using a
failed states. This method has a faster convergence rate than the Weibull distribution. In the work of [29], the author coupled MCS
sequential e MCS method. However, this method is still demanding with quantile estimation techniques, and an efficient stochastic
in terms of computational expenses as the degree of the problem optimizer, Adaptive Global Local Search (AGLS) in sizing hybrid
increases. The third and the last type is called the NoneSequential renewable energy systems while considering the renewables un-
MCS method. It is known for its high computational efficiency but certainty as inputs to MCS. Authors found that the approach has
lacks the ability to simulate the chronological aspects of a renew- enabled the control of the upside risk, consequently enhancing the
able system operation. The summary of the three types of the MCS decision quality regarding the hybrid renewable energy systems.
method mentioned with respect to the literatures mentioned is Implementation of MCS in Ref. [30] has been represented to show
shown in Tables 1e3. the possible distribution of thermal energy collected at a solar
thermal power plant. Applying a pseudo e MCS reduce the search
3.1.2. Recent MCS applications in renewable energy space of the non e convex stochastic optimizer which is the PSO, to
3.1.2.1. Sequential MCS applications in renewable energy. The find an optimal design that leads to an improvement of yearly
sequential MCS method has been implemented as follows in thermal energy collected between 3.34% and 23.5%. MCS has been
renewable energy applications [19]. implemented the MCS in sce- applied in Ref. [31] to consider the intrinsic variability of electric
nario generations of irradiation, wind speed, load, and temperature power consumption in the probabilistic assessment simulation of
as inputs to optimize for the control of PV e wind e diesel e battery DG penetration in medium voltage distribution networks. The
stand e alone systems. Authors suggested that a reduction in fluctuations and uncertainties of load demands and generations of
search space is recommended to optimize the problem within solar PVs are represented using MCS in the work of [32]. A multi e
acceptable time frames. Genetic Algorithm (GA) has been used linear MCS method is proposed by Ref. [17] to analyse the steady
initially to reduce the possible scenario generations’ space of the state operating conditions of an active electrical distribution sys-
MCS. A novel hybrid GA and MCS approach was proposed by tem with Wind and PV generation plants. The uncertainties of
Ref. [20] to predict hourly energy consumption and generation by a power load demand and power production from renewable gen-
cluster of Net Zero Energy Buildings. The MCS aspect considers the erations are considered using the MCS combined with multi e
variability and the modelling aspects of the random energy con- linearized power flow equations. In the work of [33] MCS is
sumption in a building at a given specific hour. An analytical implemented to model the uncertainties of energy demand, solar
convolution process combined with MCS in the works of [21] to energy availability, and electricity prices followed by a space search
Table 1
Sequential MCS in renewable energy applications.
References Method Objective function Type of MCS Uncertain input Parameters Main results/contributions Future work/Gaps
[19] MCS e GA Minimize investment and Sequential Solar irradiation, More information available Optimize MCS samples and
operational costs Temperature, Wind speed, on expected performance computational time
Annual fuel price interest and costs of the system
rate, Average daily load with respect to the
deterministic optimization
[20] GA e MCS Minimize instantaneous Sequential Buildings' energy Reduction in net energy Extending the method's
and cumulative energy consumptions balance in buildings period to more than one
balance year
[21] MCS e Analytical Optimize cost/benefit Sequential Wind speed forecast error, Considerable improvement Increment of MCS
Convolutional Process relationship of RE generation unit reliability of computational efficiency computational efficiency
generations with reasonable cost/
benefit
[22] Receding Horizon Minimize Cost of Energy, Sequential Wind power, Solar power, Lowest cost option may Comparison of short-term
Optimization e MCS minimize risks Battery storage, Biomass have a higher risk of failing. performance with while
combined Heat Power, The model provides ranges considering demand side
thermal energy storage, gas of possible microgrid uncertainties
producer designs to determine major
risk factors
[23] SAES e ARM eMCS Minimize investment and Sequential Load demand, Wind speed Reduces iteration in a Investigate impact of
operational costs complex search space; energy management on
Investigate discharge cycle planning decisions
efficiency of different
energy storage on the
system
[24] Risks based e MCS Minimize loss load Sequential Wind speed, River inflows Precise estimation of Integration of various
probability (LOLP), EPNS, energy delivered at a given intermittent RESs
LOLD and LOLF time and reducing load
shedding risks
[25] LOEE e MCS Minimize loss of energy Sequential PHEV owner's behaviour, LOEE in novel charging N/A
expected and expected Solar and Wind power applications reduced by
energy not supplied 75% in comparison to
unmanaged charging
[26] LINET - MCS Mitigation of lightning risks Sequential Lightning transients/ Cost e effective N/A
activity at wind turbines overvoltage protection
selection
[27] NPC e MCS Minimize net present cost Sequential Load, Solar irradiation Realistic stochastic battery Analysing cost reduction
(NPC) lifetime prediction using and fossil fuel consumption,
weighted Ah Schiffer Improving supply reliability
method
[28] TPLF - MCS Minimize risk of system Sequential Load, Solar irradiation, Accurate uncertainty Considering multi e time
over e voltage and risk of Solar PV output, modelling of Solar PV spatial and temporal
system over e load Temperature output, load, and correlations in power
temperature at multi e generation dispatch
time instants strategy
Table 2
Pseudo - Sequential MCS in renewable energy applications.
References Method Objective function Type of MCS Uncertain input Main results/ Future work/Gaps
Parameters contributions
[29] MCS e AGLS Minimize risk Pseudo - Sequential Possible sizing of HRES Sizing of HRES with An efficient quantile
minimal risk estimation method to solve
largeescale problems
[30] Ray Tracing MCS e PSO Maximize yearly Pseudo e Sequential Sun ray's position, days Increment in yearly Integration of electrical
thermal energy of the year thermal energy output in the system,
collection collected optimization of levelized
cost of energy
[33] Multi objective e Roulette Minimize energy costs Pseudo e Sequential Supply side; Demand Models and proposed N/A
Wheel ee MCS and environmental Side, domestic hot methods provided
impacts water, space heating accurate optimization
and cooling) results in identifying
the economic/
environmental pareto
fronts
[36] Cholesky Decomposition e Minimize economic Pseudo e Sequential water inflow, wind Main characteristics of Test proposed method with
MCS risks and maximize speed, solar irradiance, the random variables plant's installation site data
financial returns temperature of PV are accurately modelled (real data); Adaptation of
panels, and average for energy applications the method in other
generation capacity markets
[34] Various techniques e MCS Minimize LOLP, Pseudo - Sequential Load demands, Quantify the impacts of Integration of other
Expected Unserved Conventional integrated renewable stochastic parameters such
Energy (EUE) generation resources, resources on reliability, as solar, demands, and
Wind resources power economics, and storages
emissions
reduction technique (Roulette e Wheel Mechanism) to reduce the
geothermal plants LCA with

method where correlations computational expenses. Pinheiro et al. (2017) has implemented
Generalization of analysis
Comparison of enhanced
between parameters are MCS associated with Cholesky Decomposition as inputs to
generate synthetic time series of water inflow, wind speed, solar
Future work/Gaps
traditional plants
irradiance, temperature of PV panels, and average generation
capacity [34]. The MCS's goal is to perform risk analysis with 2000
permissible
scenarios that spans over a period of 300 months. Authors in

Ref. [34] deployed the MCS to systematically sample random
N/A
processes of intermittent renewable resources and simulated the

power system and transmission constrained day e ahead market
operations. MCS approach to investigate the economic risk anal-
Optimization of conceptual
Main results/contributions
ysis of decentralized renewable energy infrastructures has been

investments, security, and
outperforms sequential
used in the work of [35]. The MCS method considers the net
impacts of geothermal
design with respect to
MCS in terms of risks

Main environmental
present value (NPV) estimation and its ranges for each scenario
Proposed method
plants are shown
involved.
analysis
3.1.2.3. Non e sequential MCS applications in renewable energy.

returns
Few recent works that have implemented the non e sequential

MCS in renewable energy applications are mentioned as follows.
The MCS approach to investigate the economic risk analysis of
decentralized renewable energy infrastructures has been imple-
Outage per year, duration of
Uncertain input Parameters
Drilling time, geothermal
mented in the work of [35]. The MCS method considers the NPV
parameters (refer to the
Well fluid composition,
estimation and its ranges for each scenario involved. Hanbury and
Risks and finances
Vasquez, 2018 employed the usage of MCS in geothermal plant's

paper for further
construction to stochastically capture the environmental impact

clarifications)
in terms of complete Life Cycle Analysis (LCA) relative to other

well life
methods of energy production [37]. A systematic approach of MCS

outage
to address the system distribution reliability considering inten-

tional islanding was implemented in the work of [38].
3.1.2.4. Trending of MCS applications in renewable energy.

Trending of MCS applications in renewable energy applications is
hybridized with either a (meta) e heuristic method, strategic
sampling methods, or other optimization methods. The (meta) e
Non - Sequential
Non - Sequential
Non/Sequential
heuristic method typically acted as a space search reducer for the

Type of MCS
MCS method in performing the stochastic optimization as shown

in Ref. [19], thus decreasing the overall computational expenses.
Other method such as sampling methods (Typically related to
Evaluate random variables
and SAIDI (refer to paper)

interruption; SAIFI, CAIDI
environmental impact
Generator DNNs
Objective function
Minimize system
Optimize NPV
Noise Input
Non - Sequential MCS in renewable energy applications.
Random MCS (RMCS) with
Historical Samples Generated Samples

annual branch fault &
Sequential MCS
NPV e MCS
LCA e MCS
Discriminator
Method
DNNs
References
Real Output Prediction Fake

Table 3
[35]
[37]
[38]
Fig. 4. Example of GANs structure for wind scenario generation.

Pseudo e Sequential MCS) has strengthened the weakness of MCS day might not be the same as another due to erratic weather
method that requires large amount of sampling data to be accurate changes and global warming, a new method can't only rely on
[38,39]. With strategic sampling of the scenario generations only generating/projecting scenarios based on historical data but must
the most important components are considered that contributes also correctly capture the rapid variations and strong diurnal cycles
the most to the objective function's stochastic optimization. Due to of renewable resources in generating authentic new scenarios.
the time dependent nature of renewable energy problems, most of Numerous amounts of literatures exist in scenario generations of
the applications of MCS methods in this field is either Sequential or renewable resources such as wind and solar as well as demands
Pseudo e Sequential. Section 4 describes further methods involved [42]. However, most of them were model e driven and it is
in stochastic optimization sampling processes. cumbersome to pin point the most efficient usage of an exact model
to an exact situation. A recent study derived from Artificial Neural
Networks (ANNs), namely Generative Adversarial Networks (GANs)
3.2. Notable uncertainty modelling method: Generative Adversarial
by Ref. [43] has been gaining a lot of attention due to its ability to
Networks
synthesize artificial images from trainings of real ones. Only few
works have been identified in literatures that implemented data e
It is important to note that as computational advancements
driven GANs in renewable energy applications. The method suc-
have been growing rapidly throughout the years, model e driven
cessfully synthesizes renewable system's scenarios in Ref. [41] us-
uncertainty modelling or scenario generations' methods have been
ing Wasserstein GANs. The generated scenarios are successful in
becoming less viable, difficult to apply, and hard to scale [15,41].
synthesizing new and distinct scenarios by capturing the intrinsic
These are caused by the complex dynamics of renewable energy
features of the historical data.
systems, time e varying nature of weather, and complex temporal
Fig. 4 depicts an overview of GANs system. The intuition behind
and spatial connections. Studies are now converging towards the
GANs is to exploit the capacity of Deep Neural Networks (DNNs) in
data e driven methods in generating new sets of unique and
both classifying complex signals (Discriminator) and expressing
distinctive scenarios in renewable energy application [41]. As one
Table 4
Characteristic and benefits of different types of GANs.
References Type of GANs Characteristic Main advantage(s) Future work(s)
[45] Wasserstein GANs Using the Earth e Mover Stable training of GANs Developing new algorithms
distance to evaluate the improves the learning for calculating Wasserstein
distribution gap between real parameter and optimization distance between different
and generated data method of conventional distributions
GANs
[46] Loss e Sensitive GANs Limiting the modelling ability Reduces over fitting of N/A
of the discriminator to better generated samples
distinguish the real and improves the learning
generated samples regardless parameter and optimization
of their complexity method of conventional
GANs
[47] Semi - GANs Adding labels of real data to the Generates a higher quality Weighting of discrimination
training of discriminator sample than conventional and classification
GANs Generating examples with
Reduces training times for class labels
the generator
[48] Bidirectional GANs Mapping the real data to the No assumptions of Testing of the Bidirectional
latent variable space in an underlying structure of data GANs under other space of
unsupervised learning are needed architecture models
environment Outperforms many
unsupervised feature
learning approaches
[49] Info GANs Capturing mutual information Learns interpretable and Applying mutual information
between a small subset of latent disjointed representations and induce representation to
variables and observations on challenging datasets other methods such as
completely unsupervised variational autoencoder
Negligible increment in
computational expenses
compared to conventional
GANs and easy to train
[50] Auxiliary e Classifier GANs Incorporating label information Generation and Improving the reliability of
into the generator and discrimination capability of the proposed GANs
adjusting objective function for GANs are enhanced Improving visual
the discriminator Produces a more diversified discriminability
samples of data
[51] Sequence GANs Generating sequences of Excellent performance in Monte Carlo tree search in
discrete tokens synthesizing speech, poem, improving the action
and music generation decision making for large
scale data in cases of long e
term planning
[52] Boundary e Equilibrium GANs Equilibrium enforcing method Balances the discriminator Determining the best latent
paired with a loss derived from and generator in training space size for a given dataset
the Wasserstein distance Provides trade e offs Determining when and how
between samples' diversity noises should be added to
and quality the input
Variance Reduction Techniques

(VRTs)
Common Antithetic
Stratified Latin Hypercube Control Variates Dagger
Random Variates
Sampling (SS) Sampling (LHS) (CV) Sampling (DS)
Numbers (CRN) (AV)
Importance
Sampling
(IS)
Cross Entropy Adaptive Importance Sequential Importance

(CE) Sampling (AIS) Sampling (SIS)
Fig. 5. Overview of variance reduction techniques (VRTs).
complex non e linear interactions (Generator). The idea behind might only marginally increase the quality of the solution and the
GANs is to set up a minimax game of two DNNs which are in an objective function until a certain threshold [6]. One need to care-
adversarial relationship with each other. The Generator's DNNs fully evaluate the trade e offs between the accuracy and the rate of
updates its weights during each training epochs to “trick” the convergence of a given algorithm. One popular technique to in-
discriminator by generating “fake” samples of scenarios, while the crease the sampling precision is called Variance Reduction Tech-
Discriminator's DNNs attempts to distinguish between true his- niques (VRTs) [12]. VRTs can be broken down into several main sub
torical scenarios and the “fake” ones. Theoretically, after reaching e categories as shown in Fig. 5. The estimates of scenario genera-
equilibrium, the optimal solution of GANs will yield scenario dis- tions’ precision depend on standard deviation between the sam-
tributions from Generators which are hardly distinguishable from ples. The standard deviation can be expressed in equation (1)
an authentic real historical data. below:
Hence, the Discriminator can no longer differentiate the origin
pffiffiffiffiffiffiffiffiffiffi
of the data, whether it is from the generator or the real historical VðzÞ
training data distributions. It is easier to imagine the GANs as a s¼ pffiffiffiffi (1)
N
counterfeiter and a police game where the counterfeiter (Gener-
ator) keeps on improving its technique to deceive the police, while where VðzÞthe unbiased sample variance and N is the sample
the police (Discriminator) are also getting better at catching the number.
counterfeiter. The counterfeiter in the end would produce a “fake According to equation (1), the precision of the estimates can be
product” that resembles the authentic product successfully, which intuitively increased by increasing the number of samples, N.
no longer can be identified by the police. However, increasing the samples’ size would mean reducing the
To summarize, the GANs method in scenario generation can efficacy of computation. In cases of sequential sampling process
leverage the power of DNNs and vast sets of historical data in throughout a year, with 8760 h steps, each hour containing its own
performing the tasks for directly generating scenarios conforming multivariate properties, a sample increase of 1 would mean a
to the same distribution of historical data, without the need of repetition of 8760 h of sampling process. Therefore, another way to
explicit modelling of the distribution [43]. However, it is important keep the sample size small yet still maintaining a desired precision
to note that the architecture of DNNs is complex in nature and is to reduce the variance between the samples. The main idea
requires high computational efficacy in solving GANs problems. behind VRTs is to decrease the amount of sampling needed to the
Yize et al. in his work [41] has suggested the usage of efficient desired level of accuracy or increasing the accuracy of the expected
GPU(s) to accelerate the DNNs training procedures. Future works in value for a given number of samples. There are various VRTs which
renewable energy systems using GANs would be in the decision e have been reported in literature in renewable energy applications,
making strategy for unit commitments with high penetration of as depicted in Fig. 5.
renewable energy generations and incorporation of the method in The authors in Ref. [53] used a range of random variables (RV) to
probabilistic forecasting problems. Interested readers are directed develop an improved stochastic model for power system sched-
to the work of [44] for a comprehensive overview of GANs as well as uling in the presence of uncertain renewables. A work in Ref. [54]
its future trending. Hitherto, several main variants of GANs have focused on reliability evaluation through sequential Monte Carlo
been identified and are summarized in Table 4. Characteristics, simulation to address cascading failure in power systems operation.
main advantages, and identified future works of the GANs' variants The Weibull distribution together with antithetic variates (AV) is
are highlighted for the perusal of interested readers. It is to be noted implemented in order to reduce the large computational burden in
that the future works identified are mostly in the realms of simulations. Kardooni et al. [55] conducted a survey on climate
computational and mathematical sciences. However, implementa- change and renewable energy in Peninsular Malaysia based on
tions of these GANs’ variants in renewable energy systems are yet stratified sampling (SS). The authors in Ref. [56] identified the
to be tested. factors shaping public opinion based on random stratified sampling
to examine willingness to pay for expansion of renewable energy
4. Sampling methods in scenario generations sources in the electricity mix. A novel modified Latin hypercube-
important (LHS) sampling method is suggested in Ref. [57] to
Increasing scenario generations would intuitively mean a closer enhance the accuracy and speed of correlation processing under
and more comprehensive representation of possible futures. low sampling times. A LHS method is proposed in Ref. [58] to
Nonetheless, increment of scenario generations (samples taken) analyse the reliability of power systems considering the
intermittent behaviour of renewable generations such as wind, 4.1.1. Type of IS method

solar power and fluctuation of bus loads. Dahlblom [59] applied “Trainings” are encouraged with a trial distribution to capture
control variates (CV) for Monte Carlo-pricing on three-asset spread the appropriate estimate distributions using MCS. With repetitions
options with a view towards energy markets. A control variable of MCS simulations, a better trial distribution can be drawn out
based dagger sampling (DS) technique is proposed in Ref. [60] to based on the weighted MCS samples. The process is repeated until
decrease the computational effort in Monte-Carlo reliability eval- termination criteria are met. The “Trainings” and the trial distri-
uation for composite systems. Apart from the methods mentioned bution procedure is called the Adaptive Importance Sampling (AIS)
above, Importance Sampling (IS) Method has become popular in method, as the proposal distribution is updated adaptively. Another
renewable energy applications. IS method and sub-division are typical form of IS is called the Sequential Importance Sampling
explained in the following section. (SIS). As the name suggests, SIS constructs the proposal distribution
sequentially and typically requires a decomposition procedure. SIS
is normally implemented in high e dimensional problems in
4.1. Importance Sampling (IS) method building up proposal distributions sequentially. Cross e Entropy
(CE) was proposed by Ref. [62] to enable the inclusion of very un-
Recent surveys from the literatures have shown that the IS likely events in computations. CE is a popular sub e category of IS
method boost the sampling efficiency [61]. Typically, in MCS, the method in VRTs to account for the optimizations of rare events [63].
sampling representations would be excellent, if and only if samples Based on repeated sampling, the method utilizes each iteration in
can be drawn from the target distributions. However, certain rare two steps; random data generation using a specific random method
cases in renewable energy applications such as extreme wind cy- and updating the specific method's parameters to yield an
cles, sudden power outages, and rare occurrences of device failures improved sample in the next iteration. According to Ref. [64], IS
are difficult to account for. IS method focuses on sampling the method is the hardest variance reduction method to use, therefore
important region (usually named “proposal distribution”) in which expertise in the field is a necessity. Readers are advised to read the
the important region have greater occurrence probabilities in works of [46] for the detailed mathematical representations and
comparison to the original distribution. The intuition is to construct implementations of IS methods. The following paragraph briefly
a proposal distribution that “boosts” the sampling of important presents the recent literatures in IS, CE and SIS applications in
regions. The method can bring enormous advantage, making an renewable energy systems. From our extensive literature searches,
otherwise seemingly impossible problem for typical MCS, only a few recent literatures existed in the implementation of IS in
amenable. Nonetheless, applying IS method requires experience in renewable energy applications.
sampling due to its double e edged characteristics. One could easily
go wrong by yielding an estimate with infinite variance, when a 4.1.2. IS method implementations in renewable energy
simple sampling method could have yielded a finite one. Therefore, IS in reducing the computational time of MCS has been imple-
a well e chosen proposal distribution is the key to maximize mented by Ref. [65] in a probabilistic security management for
computation efficiency. power system operations with large amounts of wind power.
Fig. 6. Framework of CE based dispatch model to handle uncertainties in PHEVs and renewable generation [66].
Author has found out that the IS method significantly reduced the
computational time needed in sampling by two to three orders and
has shifted the original distribution to the desired proposal distri-
bution. In the work of [66], CE method is utilized in hybrid
renewable generation's optimal dispatch strategy of Plug e in
Hybrid Electric Vehicles (PHEVs) to improve the voltage profile and
the power flow with a 33 e nodes distribution systems. Authors
have found that the proposed method has managed to decrease the
power flows in heavy loaded lines and renewable generation fluc-
tuations. The objective function is developed using two parts. The
first part presents the expectation of population variance of
renewable generation outputs, while the second part denotes the
expected operation cost including battery degradation, PHEV
owner benefits and control of the fleet of the vehicles. The objective
function is formulated as follows:
" #
X
T
Mina1 E DP Pe;t þ a2 E CBDP ð1 þ u%ÞPdch;t Dt (2)
t¼1
Where, E denotes the outputs of PV system/wind generator, CBDP is

the per MWh cost of the battery degradation, Pdch;t is the average
power consumption of a single PHEV in time interval t, Pe;t is the
total charging/discharging power of PHEVs in time interval t, Dtis
the time interval, a1 and a2 are the probability density function
(PDF) parameters.
The comprehensive framework is developed with the multiple
cases such as typical situations of seasonal renewable generation
and vehicle usage, as shown in Fig. 6. Different renewable market
share and peak generation or demand circumstances are also
discussed.
An efficient sampling method for MCS in Ref. [67] has been
investigated using CE and Copula theory to analyse generation
adequacy of multi e area power systems with high penetrations of
wind power. Results have shown that the sampling method
significantly reduced the number of samples required to estimate
reliability parameters of interest. A robust Multi e Objective CE Fig. 7. Flowchart of SIS based hybrid probabilistic method for electricity market [71].
(MOCE) algorithm is proposed by Ref. [68] in integrated scheduling
approach to solve for microgrid supply and demand scheduling
problem under uncertainties. A multi-objective function is devel- emissions generated from electricity and natural gas, respectively.
oped using fuel price, maintenance cost, buying and selling elec- Authors have shown that the proposed algorithm has managed
tricity price, depreciation cost of battery and penalty cost which can to simultaneously minimize operation costs and emissions under
be presented in the following equation, the worst e case scenario of fluctuating renewable generations and
uncertain loads.
8
XjTj < X Leite and Castro [69] has presented a new probabilistic method
dg pg buy pg pg
F1 ¼ cfuel ftmt þ cdg pt þ zt ct pt þ 1 zt in evaluating spinning reserve margins using CE in renewable en-
: ergy systems with transmission restrictions. CE is utilized in
t¼1 dg2A
9 treating the rare events and identifying necessity equipment for
=
sell dg bt bt es es chs chs operation in such events. Authors have shown that the CE method
þ ct pt þ s pt þ s pt þ s pt (3)
; was successful in managing higher penetration of renewable
sources and ensuring a reliable operation. Graf. et al. [70] has uti-
lized the Adaptive Stratified Importance Sampling (ASIS) method in
where cfuel is the natural gas fuel price,ftmt is waste heat by burning
hybrid extrapolation and MCS method for estimating wind turbine
natural gas, A is DG unit set, cdg presents the DG unit maintenance
dg buy extreme loads. Authors have shown that the variance of the hybrid
cost, pt is DG power output, ct and csell t denote the buying and
method are reduced swiftly with the implementation of the ASIS.
selling electricity price, respectively, ppg t is power grid power
The minimal variance importance distribution can be derived as
output, sbt is depreciation cost of battery, ses and schs are the penalty
follows,
cost of shortage/excess electricity and cooling/heating respectively,
and pbt
t is the battery power.
YðxÞf ðxÞ
Another objective function F2 containing coal and natural gas q* ðxÞ ¼ (5)
combustion emissions, can be expressed as follows Ef ½YðxÞ
jTj
X
pg 1
F2 ¼ εpg pt þ εfuel ftmt (4) Ef ½YðxÞ ; with xi drawn form f (6)
t¼1
P
M tot
Mtot Yðxi Þ
i
where εpg and εfuel denote the conversion factor of carbon
Table 5
IS, CE and SIS implementations in recent renewable energy applications.
References Method Objective IS/CE distribution parameters Main Results Future work/Gaps
[70] Adaptive stratified e IS To estimate wind turbine Extreme loads, wind speed The proposed method Root causes of extreme
extreme loads outperforms sample e based IS response variation in wind
e MCS method turbine loads
[65] Risk assessment e IS To estimate very low operating Load, Wind power Decrease in computational Robustness tests with different
risks in power systems expenses of two to three orders values of controllable active
of magnitude with respect to power outputs and wind power
crude MCS forecast distributions
[66] Normal Distribution To provide an optimal dispatch PHEV's driving behaviour, wind With introduction of Vehicle 2 Consider future studies
Parameterized e CE strategy for PHEV speed, solar irradiance, system, Grid (V2G), PHEV could act as intervals in seconds and
and load data storage devices and proposed minutes relevant to power
CE models solved for multiple markets like spinning reserves
patterns of seasonal profiles for
PHEV dispatch cases
[67] Copula Theory & CE To analyse generation adequacy Conventional generation, Load, Proposed method outperforms N/A
of multi e area power systems Wind power generation crude MCS in terms of efficiency
with high penetrations of wind and accuracy by three to four
power orders. Number of samples
required does not increase with
the decrease of probability
interests' level
[68] Multi e objective e CE To schedule energy supply and Load profiles, Solar PV power Total cost and carbon emissions Large scale integration of
demand in integrated are significantly reduced using distributive resource and
scheduling under uncertainty proposed method renewable energy in regional
integrated energy systems
[69] MCS e CE To assess probabilistic spinning Wind generation capacity, Using risk assessments and The configuration of the
reserve considering renewable Equipment failures, capacity knowing the critical elements Brazilian interconnected
resources and transmission limits of transmission of the system, planners can system to demonstrate the
restrictions equipment better manage the high practicality of the proposed
penetration of renewable approach
sources, ensuring sustainable
and reliable operation
[63] MCS-SIS To assess the deviation of price, System load, renewable energy Estimations for both expected The proposed is promising to be
possible occurrence of price output, generator bidding normal price and price spike implemented on online
spike in electricity market strategy, and outage rate probability are accurate and fast applications.
using less than 3% of the MC
simulation time.
Then for any given load Yj , The estimation of ASIS can be distribution.
expressed as follows Huang et al. [71] established a hybrid probabilistic assessment
method based on SIS for electricity market risk management. The
P Y < Yj ¼ Ef I Y < Yj (7) proposed method has considered various uncertainties such as
system load, renewable energy output, generator bidding strategy,
ð and outage rate. The performance is checked under Australian

¼ I YðxÞ < Yj f ðxÞdx (8) National Electricity Market consisting of 59 buses, with 38 con-
ventional generation units and one wind farm. The authors have
ð found that the method has resonance accuracy similar to MCS re-
f ðxÞ
¼ I YðxÞ < Yj gðxÞdx (9) sults and fast execution with regard to normal price and price spike
gðxÞ probability. The implementation flow is illustrated in Fig. 7 where
system load is classified into “STATE”, and reported price of each
ð
f ðxÞ 1 X f ðxi Þ unit into “ACTION”.
¼ I YðxÞ < Yj gðxÞdx; I Yðxi Þ with xi Vast amount of recent literatures pertaining to recent IS adap-
gðxÞ Mtot i gðxi Þ
tations and improvements have been found outside of the renew-
drawn from g (10) able energy applications which has proven to be efficient and
robust to implement, mainly in the fields of signal processing and
8
> computational sciences. Recent adaptations of various IS methods
>
>
> f ðxi Þ
1 X< if i < f in the renewable energy applications are still scarce. Readers are
¼ gðxi Þ (11) encouraged to read the work of [39] which provides a compre-
>
Mtot i >
>
> hensive overview of IS methods. In this work, the IS methods'
:0 otherwise
(mainly AIS) scopes are discoursed at great depths from the past,
the present, and on to the future. Future works in IS involves the
implementations of proposed IS methods with different and wide
1 X f ðxi Þ
¼ (12) ranges of distribution parameters in high dimensional problems in
Mtot i < j gðxi Þ
which the characteristics of the problems are very similar to the
renewable energy applications. IS method's promising new appli-
where, q* ðxÞis the auxiliary importance variable, YðxÞis load, f ðxÞ is cations involves utilization of the method in the deep learning field
the distribution of wind speed, Ef denotes expectation with regard for computing the weights of hidden layers.
to f, Mtot is the total number of samples and gðxÞ is the arbitrary
Summary of the references mentioned above is specified in .

kW Cap kW Cap OM
Table 5. CESði;sÞ ¼ CESðiÞ PESðiÞ þ CESðiÞ EESðiÞ LV þ CESðiÞ (16)
4.2. Notable sampling method: Markov Chain Monte Carlo method .

CEVði;sÞ ¼ CCH cEVðiÞ LV (17)
The Markov Chain Monte Carlo (MCMC) is a popular and rapidly
growing sampling method which combines the properties of
Markov Chain and MCS [72]. The intuition behind the Markov Chain !!
is to generate random samples through a special sequential pro- h i X X X
cess. The next generated random sample depends only on its pre-
ClossðsÞ þ CconsðsÞ ¼ days
NAðdÞ rkWh
Gridði;h;d;sÞ PGridði;h;d;sÞ
i d h
vious random sample and not affected by any samples prior to the
previous ones, thus creating a chain of random generated samples (18)
until the end of iterations. This is the well e known “Markov”
Where, CEV , CRE and CES denote the capital and operating costs for
property. MCMC proved to be advantageous especially in Bayesian
PEV chargers, renewable energy resource units and DESS, respec-
inference due to the difficulty of predicting the posterior distribu-
tively, PRE is the active power provided by renewable energy re-
tions via analytic methods. MCMC grants the user the ability to
sources in kW,RRE is the return of selling energy, PES is the active
approximate the posterior distribution, with minimal number of
power consumed by the DESS, EES denotes capacity of DESS in KWh,
samples [73].
Closs and Ccons represent the costs of energy losses and energy
consumed by PEVs, normal load, and DESS, respectively, rkWh
4.2.1. Overview of MCMC Gridði;h;d;sÞ
is the selling price of renewable energy resources, PGridði;h;d;sÞ is the
A simple yet concise introduction to MCMC was written by days
energy cost distributed from the grid in $/kWh, NAðdÞ is the number
Ref. [72]. The goal of the authors was to demystify MCMC sampling
of actual day, PGrid is the generated active power from grid, LVis the
method and provide a comprehensive example to encourage new
levelized cost factor,CCH is the capital cost of PEV chargers, and cEV
researchers/users in adopting the MCMC method for their own
is number of charging stations installed at bus i .
research purposes. Interested readers are directed to the work of
A work in Ref. [77] presented a review of the measurement
[74] for in depth analysis and advanced coverage of MCMC. A more
uncertainty in energy monitoring, where the MCMC method's
technical approach of MCMC method can be found in the work of
contributions in this area are elucidated. The authors in Ref. [78]
[75].
used MCMC method in simulating the wind speed data and
In recent renewable energy applications, the MCMC method has
implemented an embedded Markov Chain to incorporate the long
been implemented as follows. MCMC simulation model has been
term effects in modelling the turbulent wind flow, as depicted in
utilized by Ref. [76] to consider the uncertainties of renewable
Fig. 8. Authors have discovered that the proposed embedded
energy generation outputs and plug-in electric vehicle (PEVs)
Markov chain outperform the conventional MCMC method.
charging demand in a combined resource allocation framework in
A slice sampling in MCMC simulation in a case study of proba-
distributed energy storage systems (DESS). The objective function is
bility assessment for power system voltage stability margin with
expressed as follows:
renewable energy source has been presented in Ref. [79]. The slice
X Xh i sampling method performs better than Gibbs sampling method
min ProðsÞ CREðiÞ RREði;sÞ þ CESðiÞ þ CEVði;sÞ with respect to smaller simulation size, and the calculation effi-
U1 ;U1 s i ciency. Besides, the proposed slice sampling method is more effi-
!
h i cient and simpler to implement in the power system probabilistic
þ ClossðsÞ þ CconsðsÞ (13) case study. The execution process of the proposed algorithm for
power system voltage stability margin using slice sampling in
MCMC is illustrated in Fig. 9.
Where
.
kW Cap
CREðiÞ ¼ CREðiÞ PREðIÞ LV (14) 4.2.2. MCMC sampling procedures
Typically, the MCMC sampling is broken down in three main
X days X sampling procedures namely; the basic Metropolis e Hastings al-
RREði;sÞ ¼ NAðdÞ rkWh
REði;h;d;sÞ PREði;h;d;sÞ (15)
gorithm, Gibbs sampling algorithm, and Differential Evolution [72].
d h
Each has its own advantages and complexity as well as types of
applications. The basic Metropolis e Hastings algorithm is known
Data partitioned
Measured wind Embedded Markov Monte Carlo Simulated output
into large time step
speed data chain generation Simulation data
periods
Generation of Mean wind speed

transition matrix simulation for large
for large time step time step
Generation of Large time step

transition matrix populated using
for short time step short time step
Fig. 8. Markov Chain development and related Monte Carlo simulation [78].
Step 1: Weibull PDF

and the beta PDF are Step 2: The slice
Step 3: The extended
constructed by sampling algorithm is
continuation power
collecting the historical utilized to obtain the
flow (CPF) with the
data from photovoltaic wind speed and
sampling matrix is
power plant(s) and the illumination intensity
executed to obtain the
wind farm(s) locations samples from their
voltage stability.
for a period of one PDFs.
year.
Step 4: The power flow

Step 5: The sensitivity equation in critical
Step 6: The distribution matrix is calculated point is linearized to
of voltage stability corresponding to achieve the eigenvector
margin is achieved by stability margin of wind related to zero
using statistical method speed and illumination eigenvalue and evaluate
intensity distribution of the load
margin variable.
Fig. 9. The voltage stability margin estimation process for power system with renewable energy source using slice sampling.
Table 6
Main advantages vs disadvantages of main MCMC variants.
References Type of MCMC Advantages Disadvantages
[80] Metropolis Hastings Knowing the posterior distribution without knowing all the The values calculated must be proportional to the posterior
mathematical properties through random sampling despite likelihood
only knowing the density for different samples Only applicable to very strongly correlated parameters
Particularly useful in representing posterior distributions that Requires a suitable step size to avoid too many rejections
are hard to determine using analytical means from the next sampling sequence or resulting in a poor
Simple implementations for highly correlated distributions exploration
Struggles in multi e modal distributions
[80] Gibbs Sampling Produces posterior distribution with good accuracy Suffers from Computational efficiency in a long run
Easy to evaluate the conditional distributions Suffers from manoeuvrability in cases of strong variables'
Conditional distributions will be in lower dimensions and dependencies
rejection sampling or importance sampling can be applied
to these dimensions
[81] Differential Evolution Faster convergence rate in a higher dimension sampling Cross e over and exchange between parallel chains of
problem sampling needs to be addressed for better convergence
Reduction in rejection rate of proposal distributions due to
multiple chains of sampling learning from each other
Requires simple tuning parameters
[82] Slice Sampling Does not require much tweakable parameters such as Suffers from curse of dimensionality
proposal distributions Sampling is done for each variable in turn using one
No rejections of samples dimensional sampling in a multi e dimensional distribution
Suitable when little is known about the sampling distribution
[83] Annealing Methods Suitable for sample transitioning from high probability region May be developed by trial and error
to another high probability region Moving in small steps from one iteration to the next
Does not suffer greatly from curse of dimensionality Requires knowledge in tuning its parameters
A heuristic method that is easy to implement
Stochastic Optimization
methods
(Approximate)
Stochastic
Stochastic Dynamic Robust Optimization
Programming
Programming
Policy Iteration /
Value Function State – Space
Model Predictive
Approximation Approximation
Control
Fig. 10. General overview of stochastic optimization [85].

for its simplicity but lacks the ability to converge properly in operating these conventional generators depends on up/down time
problems where parameters are highly correlated. Therefore, a requirements of the generators and the various starts up and
more complex approach would be suitable in a multivariate envi- shutdown costs. Therefore, the uncertainties and quality in fore-
ronment. The Gibbs sampling method by Ref. [80] separates the casting plays a major role in stochastic optimization as it effects the
multivariate problem and treats them independently by sampling prior decision that must be made.
from conditional distributions of parameters. The Gibbs method is In the second stage which is the real time operations (i.e. the
known for its accuracy but suffers in computational efficiency in a expected real time operations' costs), the input variables’ PDF must
long run. The Differential Evolution sampling procedure is a heu- be known beforehand to generate large number of relevant sce-
ristic method that generates random chained samples that “learn” narios relating to the output PDFs. The 2nd stage normally involves
from each other. Instead of relying on a single random sample and the strategy in dispatching renewable resources and reserves (e.g.
creating a chain from that random sample, multiple random sam- Pump e hydro storage) over multiple periods of time while
ples with multiple chains are generated using this method. By considering uncertainties involved. Despite the huge number of
learning from other chains of samples, the parameter's correlations scenarios generated in the 2nd stage, the scenarios are not inter-
between the samples are respected. Hence, it causes the chains of twined with each other, implicating that each scenarios outcome is
sampling to be formed within the parameter's correlations limits, independent of each other. Once the decision has been made for the
leading to a greater efficiency of sampling within the true under- 1st stage problems, decomposition method is generally used in two
lying distribution. However, the Differential Evolution algorithm e stage models to treat the 2nd stage scenarios independently,
requires a certain “tuning” parameter to sample efficiently. More resulting in a cluster of much lesser scenarios needed to be opti-
information regarding the DE sampling procedure in MCMC can be mized. Common decomposition methods used in two e stage
found in Ref. [81]. Many other main variants of MCMC exist hitherto programming models are the Benders Decomposition (BD) method
and is summarized in Table 6. The table highlights the main MCMC [75,76], Lagrangian Relaxation (LR) method [77,78], Bundle
sampling variants' advantages and disadvantages. methods [92], and Sample Average Approximation (SAA) method
[88]. A stochastic two-level model is offered in Ref. [93] to maxi-
5. Stochastic optimization methods mize the profit of the EV aggregator in the upper level and mini-
mize the cost paid by the EV owners in the lower level in the
As opposed to the deterministic optimization method which competitive electricity markets. The upper level problem relates to
assumes a perfect knowledge of a system, the stochastic algorithm the revenue obtained from selling energy to the EV owners and
models include uncertainties either in predictions, in the decision e from reducing load based on negative imbalance prices. The upper
making processes, or both. In optimizing the problem formulations level problem can be modelled as follows:
under uncertainties in stochastic models, the main approaches are
Xh þ;B þ;B
ch ch DA DA
divided in three categories, namely; stochastic programming, Maximize Et;w lt;w Et;w lt;w Et;w ; Et;w
þ;B ;B
;Et;w ;lt;w ;Et;w ;z;tðwÞw2T
ch
robust optimization, and (approximate) stochastic dynamic pro- DA
Et;w ;Et;w ch
gramming (ASDP) as shown in Fig. 6. The paper's scope is focused " #

i 1 X
;B ;B
on the renewable energy applications which are in the field of þ Et;w ; Et;w þ b z pðwÞlðwÞ (19)
ASDP. Brief information on the stochastic programming methods 1a
w2U
which are still prevalent in renewable energy applications are
ch is EV demand provided by the aggregator, lch is the
Where Et;w
shown in the next section. Robust optimization approach is not t;w
DA day ahead (DA) EV demand, lDA is the
aggregator selling price Et;w
considered in the paper's scope. The robust optimization approach t;w
þ;B ;B
generally produced over e conservative results, needed expertise DA price at time t, Et;w ; Et;w are the positive/negative energy bal-
as well as rationale in uncertainty set construction, and difficult to ance, zis the rival aggregator scenario index, Uis the number of
implement in dynamic uncertainty cases [84]. Nonetheless, inter- scenarios with regard to price and demand, ais confidence level of
ested readers are directed to the recent notable works of stochastic conditional value at risk (CVaR), bis the risk factor and pðwÞis the
robust optimization in renewable energy applications as probability of occurrence with respect to demand and price and
mentioned in the works of following authors [85e87] (See. Fig. 10). lðwÞdenotes the auxiliary variable to control CVaR.
The lower-level problem narrates the decision-making of EV
5.1. Stochastic programming owners and their reaction to the prices which can be expressed as
below:
In dealing with power generation problems, stochastic unit 2 3
commitment in the form of stochastic programming has been
6 7
implemented as a promising tool [88]. The utilization of scenario e 6 7
6 7
based/tree uncertainty's representation and probabilities in the n _D 66 ch X 7
optimization is the main idea of the stochastic programming. The Xs0 ðzÞ 2 arg Minimize E t 6lt;w Xs0 ðzÞ þ ls;t;z 7
7
0
XS ðzÞ;Z S;S ðzÞ 6 7
stochastic programming models are divided into two e stage 6 s2S 7
6 7
models as well as multistage models. The methods were mainly 4 ss0 5
used as stochastic mixed integer programming (SMIP, linear or non
e linear SMIP are denoted as SMILP or SMINLP) problems formu- X X _D 0 0
lations in renewable energy applications. þ E t K s;s Z S;S ðzÞ (20)
s0 2S s0 2S
5.1.1. Two e stage models ss0
The former two e stage models separate the optimizations in
day e ahead (1st stage) versus real e time (2nd stage) decisions. Where_X Ds0
is the EV demand percentage delivered by the aggre-
Typically, in the 1st stage (day e ahead), decisions for conventional gator, E t is the total EV demand, ls;t;z is 0the electricity selling prices
generators such as coal power plants and nuclear generators are offered by each rival aggregator, K s;s is the cost relates to the 0
made beforehand as the start e up and shutdown times for these reluctance of EV owners for shifting between the aggregators Z S;S ðzÞ
generators are not immediate. The commitment decision in is the EVs percentage that are shifted between the aggregators.
Start
Collect load data obtained from customers'

registration
Determine the demand

response structure
Identify types of loads and their average power
consumption
Stage 1
Identify demand density of end-users based on
historical data
Determine total demand of all customers (shiftable,

sheddable and non-sheddable loads)
Input historical data of wind and

PV power
Obtain mean and standard deviation through

Scenario generation and
forecasted and historical data

reduction
Stage 2
Scenario generation for load, electricity price, wind

and PV power by using MCS and RWM
Scenario reduction to Ns Scenario by using K-means

method
Master Problem
(Find the optimal solution for UC and economic
Scheduling
Stage 3
dispatch without network constraints)
t=0
s=0
Subproblem for scenario s

(Hourly evaluation of AC constraints for scenario s )
Create cut for NO

Is it feasible?
scenario s, time t
YES
YES
s=s+1 S < NS
NO YES
t < NT t=t+1
NO
Add cuts to master YES Number of cuts > ()
NO Get the optimal
problem solution
Fig. 11. Implementation framework of stochastic model to solve the optimal scheduling problem in autonomous microgrids [94].
A risk-constrained two-stage stochastic programming is sug- The authors in Ref. [95] developed a stochastic model of AC
gested in Ref. [94] to maximize the expected profit during micro- security-constrained unit commitment (AC-SCUC) problem related
grid operator considering uncertainties such as renewable with demand response (DR) considering uncertainties of wind, PV
resources, demand load and electricity price. A three-stage efficient units and demand-side participation for the day-ahead energy and
flow diagram is developed to represent the underlying the optimal reserve scheduling in an islanded residential microgrid. In addition
scheduling problem, as shown in Fig. 11. In the first phase, the to that, an economic model of responsive loads is established based
customers electrical devices and equipment demand are assessed. on real-time pricing (RTP) scheme in view of the price elasticity of
In the second phase, the scenario generation and reduction process demand and customers' utility function. The objective function of
are executed for stochastic parameters. In the third phase, the the proposed model is designed with two terms including the
optimization problem is solved by employing a risk-constraint profits associated with here-and-now (H&N) and wait-and-see
stochastic programming approach. (W&S) decisions. The objective function includes the purchasing
energy cost from renewable, dispatchable units and buying reserve depends on its previous states at time, t. Decision e making pro-
from DG. cesses are adjusted and updated hourly (or multi e hourly or sub e
hourly). Therefore, the multi e stage models represent a more ac-
Max P H&N þ EP W&S (21) curate and realistic interactions between decision e makings and
X
NT X
NJ X NG h
NT X D i
P H&N ¼ rj;t Dj;t R
Ai :ui;t þ Bi :Pi;t þ SUCi: yi;t: þ SDCi:zi;t Ci;t :RD RU U RNS NS
i;t þ Ci;t :Ri;t þ Ci;t :Ri;t
t¼1 j¼1 t¼1 i¼1
" # (22)
D X
NT X
NW X
NV
rw;t: Pw;t þ rv;t Pv;t
U
R D R U
þ Cj;i :Rj;i þ Cj;t :Rj;t
t¼1 w¼1 v¼1
XNS X NT X NG h i
EP W&S ¼ ps: SUCi yi;t;s yi;t þ SDCi : zi;t;s zi;t þ rDep i;t
U
: ri;t;s NS
þ ri;t;s D
ri;t;s
2 s¼1 t¼1 i¼1 3 " # (23)
XNs XT X
nJ X
Ns XT XnW X
Nv X
NJ
4ps: : rj;t : ri;t;s ri;t;s 5
Dep U D
ps: : rDep DP w;t;s þ r Dep
D P v;t;s EENSj
w;t v;t
S¼1 t¼1 j¼1 S¼1 t¼1 w¼1 v¼1 j¼1
unfolding uncertainties as time goes by. Each scenario generated

In the above equations, the profit of microgrid operator, P H&N is takes a unique path starting from its root node, x1 to corresponding
assessed using sum of 5 terms. The first one denotes the electricity end nodes (i.e., x6, x8, and x15), where each node along the path
consumption revenue from customers. The second term denotes represents the time at which decisions were made. For each cor-
the cost associated with distributed generations (DGs) and their responding scenario, n (i.e. n1 taking the node from x1 / x2 / x3
start-up/shut-down. The third and the fourth term represents the / x5), the problem is treated as an individual deterministic
scheduled reserve cost of generating units and loads, respectively. problem. The difficulty of the multi e stage models rises from the
Finally, the last term represents the energy cost delivered by wind non e anticipative constraints, which means that only one set of
and PV units. Similarly, expected profit of microgrid, P H&N is eval- decision variables are permissible at each node. The advantages of
uated based on the total of 5 terms. The first one denotes the unit the multi e stage models come with a huge computational expense.
commitment (UC) cost. The second and the third terms represent The number of scenarios grows exponentially as shown in Fig. 13.
the deploying reserves cost from DG units and loads, respectively. Hence, multi e stage models are harder to solve than the two e
The fourth term stands for the power cost delivered from wind and stage models. Advanced decomposition models/algorithms are
PV units in real-time and the day-ahead energy forecasted. Finally, typically introduced in these cases. Often, the techniques used are
the last term corresponds to the expected energy cost which is not nested or multi e layered decompositions and are further divided
served (EENS). The detail parameter description of the in equations into scenario e based decomposition and unit e based decompo-
(22)-(23) can be found in Ref. [95]. sition targets [6]. Common advanced decomposition algorithms in
A work in Ref. [96], proposed a risk constrained two-stage sto- multi e stage stochastic programming and its past notable works
chastic programming model to determine the optimal scheduling are shown as follows; Augmented LR [105], Dantzig e Wolfe
to maximize the expected profit of operator. The flow diagram of decomposition (Column Generation (CG)) [106], Progressive
the propose framework is operated in two stages, as illustrated in Hedging [107], Nested CG [108], Stabilized LR or CG [109].
Fig. 12. As it can be observed that, the input data is categorized into The algorithms summarized in Table 9 are used in the past
two groups, deterministic data and stochastic data. After, a set of notable works of multi e stage e stochastic programming. In
scenarios is generated using MG uncertainties. Then, an appro- Table 9, readers are also enlightened with the qualitative advan-
priate scenario-reduction algorithm is employed to reduce the tages and disadvantages of the highlighted algorithms in multi e
generated scenarios into an optimal subset. In the next stage, the stage stochastic programming, while Table 8 presents the recent
stochastic security and risk-constrained scheduling problems are works of multi e stage stochastic programming in renewable en-
addressed. The optimal scheduling of the generating units is per- ergy applications. Quantitative comparisons of the two e stage and
formed based on unit commitment (UC) algorithm and AC/DC multi e stage models can be found in the past works of [110,111].
optimal power flow (OPF) procedure by taking into account of Qualitative advantages and disadvantages of these methods are
objective function and constraints. summarized in Table 9.
Readers are encouraged to read the works specified for each From the literatures surveyed based on Table 7 & Table 8 in
decomposition algorithm, which highlights the past notable renewable energy applications, it is apparent that the two e stage
implementations of the two-stage methods in power generations stochastic models are preferably implemented due to its simplicity
and renewable energy applications. Table 7 presents the recent in implementations and a guaranteed convergence in obtaining the
works of two-stage stochastic programming in renewable energy solution. However, the multi e stage stochastic models are
applications. becoming more reliable as it better represents the complexity of
renewable systems with significant increase in renewable re-
sources and storages. Various advanced decomposition in two e
5.1.2. Multi e stage models stage and multi e stage models have proven to yield better results
In multi e stage stochastic programming models, uncertainties than the deterministic as well as perfect foresight cases (i.e.
are captured dynamically as possible events branched out of a [101,114]). The literatures in two e stage stochastic models pro-
scenario tree. Each uncertainty in events at a later time tþ1, vided a rather conservative solution with respect to multi e stage
Input data (Wind and PV power, load and demand

elasticity, electricity price and etc.)
Scenario generation (Generate scenario data based on

normal operation contingency - based uncertainties)
Scenario reduction (Reducing the number of scenarios

to NS based on PDF algorithm)
Stage I
Master problem
(find network optimal solution for UC and economic
dispatch)
Sub problem (I) Cut

(Hourly network evaluation for DC constraints)
NO
Is it feasible?
YES Cut
Sub problem (II)
(Hourly network evaluation for AC constraints)
NO
Is it feasible?
YES
Sub optimization
(Hourly network evaluation for AC constraints for all scenarios)
Stage II Scenario S1 ... Scenario SN
Master problem
(Evaluation of DCOPF) ... Master problem
(Evaluation of DCOPF)
Sub problem
(Evaluation of ACOPF)
Cut ... Sub problem
(Evaluation of ACOPF)
Cut
Is it feasible?
NO ... Is it feasible?
NO
YES YES
NO
Is it feasible?
YES
Final results
Fig. 12. Methodological framework of a stochastic model for energy and reserve scheduling considering risk management strategy [96].
stochastic models that may lead to inefficiencies in generating the literatures on stochastic programming (i.e. [82,83,88,89]) have
best solutions. With advancements of computational efficacies, started to consider the demand side uncertainty and managing the
multi e stage stochastic models are becoming more viable in demand side in optimizing the renewable's system. Main advan-
solving stochastic renewable energy problems. Applicability of tages of a responsive demand side management are the reduction
multi e stage stochastic models (short e term and long e term) in costs and minimization of energy wastage. It is to be noted that
especially in big e scaled renewable economic dispatch are yet to literatures combining the stochastic programming methods with
be fully explored. Demand side uncertainty and considering de- metaeheuristic algorithms were not being considered in this sec-
mand side response has been gaining a lot of attentions in formu- tion and only SMIP method variations were highlighted.
lating the stochastic renewable systems' problems. Many recent
Table 7
Two e stage stochastic programming methods in renewable energy applications.
References Methods Structures Objective(s) System's uncertainty Main Result(s) Future work(s)
[97] SMILP Two e stage Minimize daily operational Wind power and Energy The proposed stochastic N/A
costs storage methods reduced the total
daily costs and load
shedding
[98] Multi e objective SMILP Two e stage Minimize operational cost Demand side, supply side Applying portable N/A
and pollution (renewable), and energy renewable energy
storage resources have decreased
the cost and increased
profits
[99] Novel decomposition e Two e stage Minimize NPV of total Solar irradiance, wind, and The proposed method Envisaged to be used in MG
SMILP expected costs load demonstrated the planners, Govt. and private
effectiveness in finding the agencies
optimal battery energy
storage system (BESS)
power and energy sizes
[100] BD e SMIP Two e stage Minimize day ahead Demand side, supply side Day e ahead power Demandeside procurement
purchase cost and expected (RE), electricity prices procurement and the by twoestage stochastic am
resource cost formulation as a two e
stage SMIP problem is
proposed
[101] BD e SMINLP Two e stage Minimize expected total Demand, EV availability and Using the energy from EV N/A
operation costs including storage capacity, renewable reduces the total operation
generation, day e ahead energy resources cost of the microgrid. The
market, and battery wear results yielded better cost
for the next 24 h savings than a
deterministic model
[102] BD e SMILP Two e stage Minimize environmental Wind speed, solar Including demand response N/A
and social impacts irradiation, and demand as a flexible load reduces
load curtailment and
reduces energy generation
needed
[103] ε e Constraint multi e Two e stage Maximize DG owners' Wind speed, load, Solving the reconfiguration N/A
objective SMILP profits and minimize electricity price of the network and
Distribution Company's allocation of DG
(DisCo) costs simultaneously produced a
more desired scheduling
between the stakeholders.
The stochastic optimization
is compared to a
deterministic optimization
with an improved profit on
behalf of the DG owners
[104] Scenario e based SMINLP Two e stage Minimize active and Load demand, wind power Reduction of expected costs N/A
reactive power purchasing of energy and reactive
costs power as well as emission
costs
5.2. Approximate stochastic dynamic programming unintentionally increase the total costs throughout the whole
period of optimizations. One need to evaluate decisions made at all
Stochastic dynamic programming is an optimization method in stages carefully to obtain the most cost e efficient objective func-
solving discrete multi e stage decision e making processes with tion. Stochastic dynamic programming method can capture the
underlying uncertainties or probabilities. Decisions made to lower trade e off between decisions made in the present and the future
the objective function's costs at a current stage might stages. Due to these properties, it is instinctive that stochastic
t x1
t+ 1 x2 x9
t+ 2 x3 x4 x10 x11
t+ 3 x5 x6 x7 x8 x12 x13 x14 x15
Fig. 13. Scenario tree with multiple stages (4 stages, 8 scenarios, and 15 nodes).
Table 8
Multi e stage stochastic programming methods in renewable energy applications.
References Methods Structures Objective(s) System's uncertainty Main Result(s) Future work(s)
[112] Dynamic Response e Multi e stage Maximize net social benefit Demand side and Energy A responsive demand side Enhancing planning
SMILP Storage provided a more flexible methodologies using k e
and smarter power systems means and system states
[113] Decision dependent e Multi e stage Maximize total profit Wind capacity penetrations The proposed method Developing new models
SMILP and demand provided effective with uncertainties
optimization information constraints
on investment and long e
term expansion planning
[114] Two e period multi e Multi - stage Minimize NPV related to Generation sources, The proposed method N/A
stage SMILP losses, emission, electricity demand, produced significantly
maintenance, operation, emission prices, demand better results in terms of
and unserved energy growth objectives and yielded
robust decision - makings
in comparison to
deterministic methods
[115] Piecewise multi e stage Multi - stage Minimize operational costs Load and Water inflow Inclusion of piecewise N/A
linear stochastic and computational time of linear approximation
optimization long e term generation boosted the computational
scheduling of hydropower efficacy and minimized the
operational costs in
operating large storage
capacity hydro power
plants
Table 9
Qualitative advantages and disadvantages of the two e stage and multi e stage stochastic programming algorithms.
Stochastic Optimization Type References and Algorithms Advantages Disadvantages

Method
Stochastic Programming Two e stage [80,81] Lagrangian Simple Implementations and Probabilities of scenario
Relaxation (LR) [88] Sample easier to understand generated need to be known
Average Approximation Convergence and good Computationally expensive for
(SAA) [78,79] Benders performances are guaranteed as large number of scenarios
Decomposition various decomposition methods generated
[92]Bundle Methods have been tested Complexity in dealing with
Robustness issues can be integer variables during the 2nd
addressed using risk stage (i.e. unit rescheduling in
measurements real e time)
Value of stochastic solution and Assumption of static
expected value of perfect uncertainties
information can be provided
Multi e stage [105] Augmented LR Considering over multiple time Size of problems grows
[106] Column Generation intervals of uncertainties in exponentially with increasing
(CG) [107] Progressive decision e making processes time intervals and scenarios
Hedging Uncertainties and decisions can Requires explicit scenario tree
[109] Stabilized LR or CG be adjusted dynamically as representations
scenarios unfold Difficulties increase with integer
Advantageous for systems that variables present in all stages
needs quick rescheduling
Providing value of perfect
information and value of
stochastic solution
dynamic programming is suitable in the applications of renewable three categories as shown in Fig. 6. The scope of the paper is within
energy optimizations. the policy function approximation in the form of stochastic Model
The usage of dynamic programming can be dated back to late Predictive Control (MPC) in renewable energy applications. Readers
1970s [90] in solving deterministic problem. The solving approach are directed to the recent renewable energy applications
was based on Bellman's Principle of Optimality [91] which uses the mentioned which highlights the usage of value function approxi-
backward induction method. The past works of dynamic pro- mation [117e119] and state e space approximation [93,94]
gramming suffer from heavy computational expenses due to the methods. A comparison of approximate dynamic programming
curse of dimensionality. As the number of scenarios and states in- techniques was carried out by Ref. [122]. Authors have compared
creases as stages unfolds, the time needed in yielding a solution various policy iteration and value function approximation tech-
grows exponentially. Hitherto, various methods and broad class of niques. Authors have argued that new theory and methodology are
algorithms have been tested to overcome the computational needed for these techniques in order to solve real e world prob-
expenses. lems, which are becoming more difficult.
Approximate stochastic dynamic programming (ASDP) has
proven to lighten the burden of dimensionality's curse of dynamic 5.2.1. Model predictive control (MPC)
programming and is well suited for models with uncertainties and MPC, also known as receding control horizon approximates
stochasticity [116]. Generally, the ASDP method can be divided in policies by iteratively solving a finite horizon optimal control
Fig. 14. Framework of the MPC optimization-based heating, ventilation, and air conditioning (HVAC) systems. The boxes highlighted with blue denote the factors that have impact
to the optimization problem directly; the boxes highlighted with green indicate the optimization problems results [123].
problem. The horizon recedes once the optimal control for a current the main author in his past work [127] with regards to theory and
stage, t has been found moving on to another finite horizon at a applications of HVAC control systems using MPC and was regarded
later stage, t þ 1. The process is repeated until the optimal control as the most remarkable review on MPC due to clear classifications
has been found for all stages; t (initial stage) until tmax (final stage). and comprehensive scheme of MPC implementations.
Serale et al. [123] have suggested several parameters which have a
direct impact to the MPC optimization problems namely; Objective 5.2.3. Stochastic MPC implementations in renewable energy
function, receding horizon, control model, constraints, and distur- Stochasticity of MPC in recent renewable energy applications
bances, where the optimization problems of MPC can be further are typically represented as probability e constrained scenarios or
divided into Linear and Non e Linear problem formulations. The forecasts, uncertainty modelling of scenario generations, and
framework of the proposed method is illustrated in Fig. 14. random disturbances. Stochastic MPC can be further derived into
Many works in the scope of MPC have been found in literatures three main categories which are tree e based, chance e con-
and are highlighted in the next paragraph. Recent literatures in strained, and multi e scenario MPC. The summary of recent liter-
stochastic MPC are mentioned in the later paragraphs of this sec- atures pertaining to these categories is mentioned in the next
tion and summarized in Table 10, Table 11, and Table 12. sections.
5.2.2. Notable overview of MPC 5.2.3.1. Tree e based MPC implementations in renewable energy.
Several recent reviews in MPC have been published in the lights Tree e based MPC works with an assumption of time dependant
of power generations, building and environments, and renewable events can be synthesized from a rooted tree, where the most
energy applications. Interested readers are encouraged to read the relevant possible disturbances can be captured. The concept of tree
works of following authors where theoretical modelling and ap- e based MPC is quite similar to the multi e stage stochastic pro-
plications of MPC are further discoursed. The state e of e the e art gramming approach (refer Fig. 7). Each root to different nodes’
development of MPC has been reviewed by Ref. [124] for renewable paths represents a possible disturbance scenario, where the
energy applications. The authors have presented a systematic re- branching of the paths symbolizes the different forecast possibil-
view of MPC applications in the field of solar PV and wind energy ities and uncertainties along a given prediction horizon. Each node
renewable systems. The authors aimed to help researchers in at a given point in time, t corresponds to a control action that can be
further exploring the flexibility of MPC for design, implementation, taken at that time. One must note that the control action taken
and analysis in renewable energy applications. In Ref. [125], hier- must not be allowed to diverge before the bifurcation points. The
archical energy management strategy based on model predictive tree e based MPC utilizes the ensembles of forecasts and solves it
control is proposed for microgrid management operation consid- by considering the sequences contained in the tree. Different paths/
ering different endogenous and exogenous sources of uncertainties. branches of the tree nodes are treated as individual deterministic
In Ref. [123], MPC in the themes of enhancing building and HVAC problems. The path with the least costs or the most efficient in
system energy efficiency have been systematically reviewed. The terms of given objective functions are implemented at current time,
potential benefits and future of MPC are discoursed at a great depth t as a control action. The process is repeated until the control op-
in authors’ work. ANNs based MPC has been reviewed by Ref. [126] timizations over the entire horizons have been obtained.
in a case study of a residential HVAC system. The authors have A hybrid robust and stochastic accelerated MPC have been
utilized new ANNs algorithm to design a supervisory MPC which implemented in the work of [128] with 24 h horizon window for EV
successfully reduced operating costs of equipment while con- integrated microgrid energy management considering demand
straints are not violated with a cost reduction percentage range of response. The authors have utilized the hybrid MPC with forecasts
6%e73% depending on the season. A similar review was made by coupled with Benders decomposition (BD) method to
Table 10
Tree e based stochastic MPC in renewable energy applications.
References Method Objective MPC Type Control horizon Sampling System's uncertainty Main Results Future Work
resolution
[128] Stochastic Minimize total daily Tree e based 16 h ½ hour EV charging demand, load, The stochastic MPC Extending the proposed
accelerated MPC operational costs real e time electricity price, outperforms the method with available EV
renewable energy output deterministic MPC by lower charging load prediction
total daily operational cost models
in all cases
[129] Risk e averse Maximize profit and Tree e based 24 h 1h Wind power forecasts, price The proposed method Application of the proposed
stochastic MPC minimize risks (CVaR) of energy outperforms all mentioned method to real e world
methods and marginally cases and other renewable
expected profit compared applications
to perfect solution
[130] CVaR fault tolerant Optimize CVaR Tree e based 4 steps ahead 1s Wind power forecasts The proposed method has Solving the proposed
stochastic MPC achieved a control stochastic MPC in one step
performance of 40% higher to yield a higher practical
than the common Min e value
Max performance index
A. Zakaria et al. / Renewable Energy 145 (2020) 1543e1571

Table 11
Chance - constrained stochastic MPC in renewable energy applications.
References Method Objective Type of MPC Control Sampling System's uncertainty Main Results Future Work
horizon resolution
[131] Multi e time scale Minimize weekly Chance e constraint 12 h 5 min/1 h PV power forecast, plug in Shifting the hourescale and dayescale Applying the proposed method in
stochastic e heuristic operational costs EV, deferrable and non e appliances to the optimal hours and multi e scale microgrids
MPC deferrable appliances in week of the day can substantially
smart home reduce the weekly operational costs
[132] Stochastic two e stage Minimize cost of energy Chance e constraint 6h 1h Renewable energy Experimental results have proven the Analysing the scalability of
MPC and emissions of resources, demand feasibility and implementation ability proposed framework and
greenhouse gases of proposed stochastic MPC that investigating distributed methods
outperforms the deterministic MPC
[133] Stochastic warping Minimize wind power Chance e constraint 1e12 h 5 min Wind power forecasts The proposed stochastic MPC The proposed control system can be
function MPC tracking error outperforms the deterministic MPC in integrated into currently existing
power tracking errors system
[134] Stochastic e EMPC Minimize microgrids' Chance e constraint 72 h 1h Renewable supply, Load The proposed method achieved a better Incorporating the topology of
operating costs demand trade e off between performance and distribution network, energy
computational efficacies in comparison exchange between MG and
to centralized scheme fluctuating prices of energy
[135] Stochastic MPC Minimize operational costs Chance e constraint 24 h ½ hour RE generations, load, The stochastic MPC framework Applying the stochastic MPC in a
demands, EV, and outperforms the traditional day e multi scale microgrid systems
electricity prices ahead programming strategy in terms
of minimizing the operational costs
1563
simultaneously reduce total operational cost in energy manage- trial solution at iteration m, εm;p;w represents the sensitivity for the
ment as well as improve computational efficiency. Simulation re- corresponding Qm;p;w;tþkjt , andEex;t indicates the energy purchased
sults showed that the proposed method outperforms the standard from the external gird.
deterministic MPC method with regard to total operational cost by A risk e averse stochastic MPC based on real e time operation
a margin of around 10%. The algorithm of stochastic BD applied to has been developed by Ref. [129] for a wind energy generation
MPC is shown below. system combined with a pumped hydro storage unit to maximize
Algorithm: Stochastic BD
At each time step t do
Initialize 1, Upper Bound (UB) , Lower Bound (LB) , 0
For k 1 to N
Set 0, do
Solve master problem in Equation (24) and determine a trail solution X , g ,t kt
Update the value of LB

Solve all sub-problems in Equation (25) with trail solution X , g ,t k t
Update the value for UB

if LB / UB then
Add the new optimally cut associated with iteration to the master problem
Set 1 , and repeat the solution procedures
else
an optimal solution is obtained, and implement the control signals at time step t
end for
profit and minimize risks in day e ahead bidding strategies. Au-

thors have compared the results of stochastic MPC method with
N h
X i several other methods such as deterministic MPC, bid e following
Zm ¼ min lf Gg;tþkjt Xg;tþkjt þ F Xg;tþk1jt ; Xg;tþkjt þ qk
heuristic and open e loop algorithms. The stochastic MPC method
k¼0
Up Uw h outperforms all other methods and reached an expected profit
X X
qm pp;tþkjt pw;tþkjt Qm;p;w;tþkjt εm;p;w Gg;tþkjt close to the perfect information solution with a margin of around
p w 2%. Fault tolerant control problem of wind energy conversion sys-
i tems have been addressed by Ref. [130] using stochastic MPC based
Xm;g;tþkjt X m;g;tþkjt (24)
on CVaR. Authors have implemented the Markov jump linear
model to model randomness of the wind energy conversion sys-

p p;r p;r tems. A scenario e tree is created within the prediction horizon to
Qm;p;w;tþkjt ¼ min lex;tþkjt Eex;tþkjt þ lf Gl;tþkjt (25)
transform the stochastic MPC problem to a deterministic MPC. The
method produced a better fault tolerant control performance in
Where lf is fuel price, Gg;t is the gas input of combined heat and comparison to the Min e Max performance index. The objective
power (CHP) units, FðXg;tþk1jt ; Xg;tþkjt Þ is penalty function used to function formulation of CVaR using SMPC algorithm is shown
control the frequency changes during the on/off operating state, below.
qm is Benders cut at iteration m, Qm;p;w;tþkjt denote the sub-problems
value at iteration m under pth and wth scenario, Xm;g;tþkjt is the
Algorithm: SMPC algorithm of CVaR objective function

1. Prepare the controller Ci
1.1 Generate m scenario trees according to different root node w k i i w k 1, 2,..., m
1.2. Calculate corresponding Controllers Ci
2. Estimate VaR
set 90%
Solve SMPC problem in Equation (26)
The VaR is given by Equations (27) and (28)
3. Estimate SMPC of CVaR
For i=1:3
CS . . f ones 1, s * nu , 1 2 ,..., s ,% decision vectors weights
Calculate other parameters;
end for
set T; %simulation time
for k=1:T
measure x k
solve CVaR SMPC problem in Equation (29) and obtain u1
apply u k u1
end for
X X
min pðxi xr ÞT Q ðxi xr Þ þ pi uTi Rui
u
i2TjT1 ∪S i2TjS
8
>
>
< x1 ¼ xðkÞ (26)
s:t xi ¼ AðwðkÞÞxpreðiÞ þ BðwðkÞÞupreðiÞ þ DðwðkÞÞ þ D1 ðwðkÞÞeðkÞ þ Iw*yrðkÞ; i2TjfT1 g
>
>
: Gx xðkÞ þ Gu uðkÞ g; k ¼ 0; :::; N; cwðkÞ2W
output of a given process within the constraints of ymin & ymax , and
l is the confidence level of such constraints that can be satisfied.
According to equation (2), the basic idea of a chance e con-
Gxj; Gvj; Gdj; Gd1j; Gd3j (27) strained MPC is to solve the optimization problem in each horizon
while guarantying the satisfaction of the constraints with a certain
probability. It is to be noted that the chance e constrained MPC
abðuÞ ¼ minfa2ℝ : jðu; aÞ bg (28) involves the careful selection of future output predictions and its
mins ½p1 p2 :::ps m

u;fvj gj¼1
s:t m Gx xðkÞ þ Gx UðkÞ þ Gd Iw1 þ Gd1 ðIw1*eðkÞÞ þ Gd3 *ðIw3*yrðkÞÞ Lðr þ aÞ
m Gx xðkÞ Gv UðkÞ Gd Iw1 þ Gd1 ðIw1*eðkÞÞ Gd3 *ðIw3*yrðkÞÞ þ Lðr þ aÞ (29)
m o; j ¼ 1; :::; s
UðkÞ u
UðkÞ u
uncertainties. Since exact future output predictions can't possibly

be captured, uncertainties are represented in either of these two
Where the xk is the present state, pi represents the realization ways; which is either the uncertainty in future disturbances or
probability of scenario i, Q and R denote weight matrixes, Gx 2 uncertainty of the process outputs due to manipulated variables.
Rnx þnu and Gu 2Rnx þnu stand for coefficient matrixes used in state Within this realm of solving probabilities and uncertainties in
and input constraints, f ðu; wÞis the estimation error, b is the prob- chance e constrained MPC, several recent publications have been
ability factor. By using the probability density of error,a can be identified and listed below.
found as a ¼ b VaR . The jumping information in Markov jump A multi e time scale stochastic MPC combined with genetic
linear model is denoted by Gxj; Gvj; Gdj; Gd1j; Gd3j . algorithm (GA) is proposed in Ref. [131] in order to perform
scheduling deferrable appliances and energy resources of a smart
home (SH) system. The stochastic parameters namely; solar irra-
5.2.3.2. Chance e constrained MPC implementations in renewable diances and its prediction uncertainties are forecasted using neural
energy. Chance e constrained MPC relies on the formulation of network toolbox in MATLAB. The uncertainties of the appliances’
output constraints with a given type ymin y ymax as chance usage as well as the economic and technical constraints of other
constraints as shown below: energy sources such as diesel generators, batteries, and PV panels
are also modelled by the author. The objective function is devel-
Pr ðymin y ymax Þ l (30a) oped for SH with a goal to minimize the value of the stochastic
forward-looking objective function subject to various constraints. A
where Pr ðxÞ is the probability of an event X occurring, y is the
Table 12
Multi e scenario stochastic MPC optimizations in renewable energy applications.
References Method Objective MPC Type Control Sampling System's uncertainty Main Results Future
horizon resolution Work
[136] Adaptive constrained Minimize operation Multi - scenario 1e24 h 0.01e1 s Renewable energy The method produced a N/A
stochastic MPC costs sources, electrical loads less conservative
solution compared to
the robust MPC
approach
[137] Various stochastic MPC Compare multiple All types 5 steps ahead 30 s Renewable resources, Chance e constrained N/A
types of MPC Load, hydrogen e MPC outperforms other
based PEM electrolyser MPC types resulting in a
and fuel cells, lead acid lower cost and less
batteries' state of energy exchange in a
charge hydrogen based
microgrid
total of 6 cost terms are taken into consideration to develop the aand b denote the confidence and trade-off parameter respectively.
objective function including fuel costCd;t F:DG , carbon esmsmions cost A new distributed chance e constraints stochastic EMPC scheme
E:DG , start up cost C STU:DG shut down cost C SHD:DG of DG, switching
Cd;t has been presented in Ref. [134] for coordinated stochastic multiple
price of PEV battery C SW:PEV and cost or income due to the power microgrids energy management. The supply and demand side un-
distribution with the grid Pd;t Grid p0 DISCO certainties were handled using the probabilistic forecasts of
d;t
X renewable generations and load of each microgrid that is in a
FL
minFd;t ¼ min FL
Fd;t UPV
d;t ; d2D; t2ft1 ; t2 g; ct1 2T1 ; ct2 2T2 cooperation scheme with each other. The proposed method suc-
PV cessfully reduced the system operating costs and achieved the
Pd;t supply and demand balance in each microgrid within the control
horizons through the development of distributed network operator
(30b)
(DNO) controller. DNO acts as an intermediary between microgrids,
8h i h i 9 thus the energy selling between any two microgrids is performed
>
> >
> indirectly through DNO. The mathematical expression of cost
> Cd;t
>
F:DG
þ Cd;tE:DG >
>
>
> h i >
>
>
> þ 1 xDG >
> function is presented in the following equation,
< DG STU:DG =
d;t1 xd;t1 C 0 1
Fd;t ¼ h i d2D; t2
> > XM
>
> þ xDG DG
d;t1 1 xd;t1 C
SHD:DG
>
> ðkþhjkÞ h þ ðkþhjkÞh
>
> h i h i >
> HX B pur;D
D 1B
E pur;D Esel;m sel;m C
> þ x0 PEV C SW:PEV þ P Grid p0 DISCO >
> > C
>
: >
; min JD ¼ B m¼1 C
d;t d;t d;t B X
M C
Epur;DðkþhjkÞ h¼0 @ A
Esel;D ðkþhjkÞhsel;D Epur;m ðkþhjkÞhpur;m
ft1 ; t2 g; ct1 2T1 ; ct2 2T (31) Esel;mðkþhjkÞ
m¼1
Esel;DðkþhjkÞ
The multi e time scale MPC divided the control optimization in Epur;mðkþhjkÞ
scale of minutes and hours in a weekly operation, where usage of
certain appliances is dominant in their respective time scale. The (35)
author has shown that the proposed MPC method has managed to
Where,Epur;D and Epur;m denote the energy purchased from the grid
notably decrease the weekly operational cost of the smart home
and DNO respectively, Esel;D and Esel;m represent the energy sold
system.
back to the main grid and DNO respectively. hpur;D and hpur;m stand
An experimental case study was conducted by Ref. [132] in the
for energy price purchasing from the grid and DNO respectively
operation management of microgrids using stochastic MPC to
while hsel;D and hsel;m signify the energy price selling to the main
optimize the economic and environmental parameters. Un-
grid and DNO respectively.
certainties due to renewable energy resources and demand were
The optimal operation of a smart residential microgrid based on
considered and the stochastic optimization was solved by using
stochastic MPC has been conducted in the work of [135]. The res-
mixed e integer linear programming toolbox via commercial
idential microgrid comprised of renewable energy resources,
solvers. Experimental results have proven the feasibility and
distributed energy generators, energy storage, electrical vehicle,
implementation ability of stochastic MPC that outperforms the
and smart loads. The uncertainties are modelled in a short e term
deterministic MPC. Kou et al. [133] proposed a stochastic MPC for
forecasts of renewable energy generations, load demand, and
wind farm energy dispatch strategy with BESS using probabilistic
electricity prices. The proposed method aimed to reduce the total
wind power forecasts. The method considers the non e gaussian
daily operational costs of the microgrid. The simulation results by
wind power forecast uncertainties using chance e constraints
the authors have shown the economic advantages of the method in
warping function. The authors have shown that the proposed
comparison to the traditional day e ahead programming approach.
method outperforms the deterministic MPC method in terms of
power tracking errors while maintaining the state of charge (SOC)
of the battery within operational limits. The chance constraint
5.2.3.3. Multi - scenario MPC implementations in renewable energy.
optimization problem is developed to enhance the wind power
Multi e scenario MPC utilizes multiple scenario generations within
dispatchability and lessen its oscillation, as shown in equation (32).
a given optimization horizon to implement a control action at
In addition, SOC constraints and charge discharge power con-
present time, t. Similar to uncertainty modelling (Refer to Section
straints are assigned in order to protect the battery from being
3), the independent multiple scenarios generated are synthesized
overcharged and over discharged, as expressed in equation (33) and
from random input variables of PDFs to produce PDFs of output
equation (34) respectively.
variables in representing the uncertainties. Ranges of solutions
exist, each with its own probability as represented in the output
XH
min J¼b eðk þ hjkÞ2 PDFs. The most cost e effective scenario in terms of objective
uðkþhjkÞ;eðkþhjkÞ functions are chosen to be the control action within the optimiza-
k¼1
tion horizon.
X
H 1
þ ð1 bÞ uðk þ hjkÞ2 ; Subject to (32) An adaptively constrained stochastic MPC has been proposed in
h¼0 Ref. [136] for optimal dispatch of microgrid. The objective function
is formulated to minimize the total operation cost including cost of
h i running generator and cost of purchasing electricity form DG, as
Pr yðt þ kjtÞ yref ðt þ kÞjt eðt þ kjtÞ a; h ¼ 1; 2; :::; H
expressed in the following equation.
SOCmin x2 ðk þ hjkÞ SOCmax; h ¼ 1; 2; :::; H (33) X

T
min fccon Pcon ðt þ ijtÞg þ cGrid ðt þ ijtÞPGrid ðt þ ijtÞ (36)
i¼1
PB;max uðk þ hjkÞ PB;max; h ¼ 1; 2; :::; H (34)
Where T represents the length of time horizon, i denotes the time
Where eðk þ hjkÞis the set of auxiliary variables, yðt þ kjtÞ is the step index, Pcon ðt þ ijtÞ stands for power discharge from the
stochastic variables, yref denotes the reference, Pr is the probability, controllable generator in i-step ahead, cGrid ðt þ ijtÞ denotes the
electricity price for energy exchange in i-step ahead, PGrid ðt þ ijtÞ is determined in order to produce the needed solution.
power exchange between MG and DG in i-step ahead.
The method adaptively/dynamically changed the rate of 5.2.4.2. Future MPC's trending. The future trends in stochastic MPC
constraint violation in the microgrid operation to improve the are converging towards a multi e scale and multi e time based
performance of the energy dispatch. In comparison to the robust optimizations as stated in the works of [125,126,129]. In a renew-
MPC method, the authors have shown that the method can improve able energy system, where multiple sources of energy generations
the dispatch performance (less conservative) in cases of uncertain are present, a realistic representation must consider these sources
renewable generations and loads. Furthermore, with increment of in order to provide an insight closer to real e world applications.
prediction horizon, computational efficacies were not significantly Managing surges of dynamic demands and supplies from plug in EV
affected. (V2G), varying behaviours of energy consumers, smart appliances,
Stochastic MPC control strategies in a case of hydrogen e based demand response, and intermittent multiple renewable energy
microgrid have been compared in the work of [137]. The three resources are the challenges that must be addressed together in
categories of stochastic MPC mentioned in the previous paragraphs future smart grid e systems. In addition, these challenges are all
were compared in the work of the authors in an experimental setup time e dependant variables in which, each of them possesses traits
including a PEM electrolyser, lead e acid batteries, and a PEM fuel e with dominance in certain time e steps. Addressing the challenges
cell as the main equipment. For each category of the stochastic MPC in a multi e time scale approach could capture the undisclosed
effectiveness, performances, advantages, and disadvantages were dynamic behaviour of the system.
discoursed. Authors have discussed extensively the valid criteria In such systems where the dynamics are complex, multivariate,
needed when selecting the appropriate stochastic MPC method. and time dependent, exact solutions are difficult to obtain. There-
fore, approximate solutions to such cases are more feasible in the
5.2.4. MPC's comparison and future trending forms of ASDP. The works of [19,138] combines the stochastic
5.2.4.1. MPC's comparison. It is apparent from the trending of method with a (meta)heuristic methods. The stochastic method is
recent stochastic MPC in renewable energy applications that the hybridized with genetic algorithm to produce ranges of relaxed
tree e based and chance e constrained MPC were the most used solutions. Trending in hybridization of stochastic and meta-
methods in recent studies. The multi e scenario MPC yielded a heuristic methods are relatively new but promising in the field of
robust but over e conservative solutions. Therefore, this category of stochastic optimizations to improve the ASDP algorithms. Inter-
MPC is not preferred due to the need of an accurate representation ested readers are encouraged to read the works of [2,139e141] for
of the system, in which the tree e based and chance e constrained recent reviews of (meta)heuristic methods and intelligent searches
MPC could provide better. in the field of renewable energy applications.
Furthermore, in cases of stochastic MPC applications, the pre-
diction/control window is typically within 24 h. Despite the heavy 6. Conclusions
computational expenses of the tree e based compared to multi e
scenario MPC, the calculation time within the mentioned window Stochastic optimizations in renewable energy applications have
is still relatively inexpensive. The multi e scenario MPC is more shown its successful implementations in recent surveys that are
suitable in cases of huge numbers of scenarios needed to be presented in the paper. Almost always, based on the works of many
considered (i.e. Optimization within 8760 h in a year, 1 e hour time authors, the stochastic optimization techniques exhibit enhanced
step, and multivariate properties). The multi e scenario MPC could performances and can deliver accurate representations in capturing
provide a certain robustness of system's representation to the po- the uncertainties of renewable systems. Despite its advantages, due
tential disturbances and provide a trade e off between the best to numerous amounts of samplings and unfolding events, which
solution and the computational expenses. are discussed in the works of many authors to improve or develop
The chance e constrained MPC offers the lowest computational novel algorithms in increasing the efficiency of stochastic optimi-
expenses compared to the other two. It formulates the optimization zation techniques. Within these contexts, the relevant research
problem by considering the probabilities of the uncertainties themes going into the future based on stochastic optimization al-
without adding the variables' size. In the work of [137], the chance gorithms are concluded as follows:
e constrained MPC outperforms the other MPC methods by offer-
ing a reduced computational time, lower operational costs, and i. Novel scenario generations and uncertainty modelling ap-
minimal energy exchanges with the networks. These advantages of proaches; These are necessary in renewable systems in-
chance e constrained MPC are one of the reasons of frequent usage tegrations where trending in the future involves stochastic
of this MPC method as shown in recent literatures stated in Table 11. multi e scale modelling. With rapid increment of data and
However, the chance e constrained MPC requires an explicit sta- size of renewables' problem, perhaps model e driven ap-
tistical characterization of the systems’ disturbances. For the se- proaches alone could not fully address and cope with the
lection of the suitable MPC method, priority factors such as underlying complexity in vast multivariate and expanding
operational costs and computational expenses must be taken into renewable systems. Data e driven scenario generations
considerations [137]. provided a general guideline in choosing the could provide a pivotal role as highlighted in the works of
best stochastic MPC for a given priority factors. Nonetheless, in [41,142].
general categories of stochastic dynamic programming, an efficient ii. Unfolding dynamic uncertainties in multi e stage problems;
method lies often on the specific problems at hand as stated by Addressing dynamic probability issues as scenarios/new
Ref. [6]. forecasts unfolds have been addressed by several authors
The prediction/control window played an important role in [103,104] in the paper. Better weather and power forecasts
determining the accuracy of the solution as well as computational which provide information with dynamic uncertainties as
time. A long prediction/control window would mean a more ac- events unfold would incorporate a more robust real e time
curate representation of unfolding events, thus yielding a greater decision e making strategy for generation companies in
accuracy in finding the best solution. However, the computational handling stochastic renewable generations.
expenses increase as the window increases. Trade e offs between iii. Implementing new recent notable algorithms in the field of
prediction/control window and computational expenses must be renewable energy optimizations; Recent work by Ref. [143]
in the form of proximal policy optimization (PPO) has shown [152], accurate modelling of consumer's behaviour [142],
great promise in updating multiple epochs per data sample. security and privacy and scalability [155].
The method boasts the ability of simple implementations iii. Multi e scale and multi e time e scale distributed renewable
and great stability as well as better overall performance in energy systems; vast amount of literatures have supported
comparison to its predecessor, trust region policy optimiza- the claim that having hybrid or combination of renewable
tion (TRPO). The PPO algorithm has attracted many authors systems would allow for a higher fraction of renewable
especially in the field of computational sciences. No works of generation in a distributed renewable energy system. How-
PPO have been published in renewable energy applications. ever, increasing the scale of distributed generation from
iv. Improvements of existing sampling and decomposition housing, to district, and finally to national scale would mean
methods; Parameters such as number of scenarios needed, addressing new challenges such as ensuring the grid and
scenario reduction techniques, quality of scenarios gener- market stability in a growing complex socio e techno e
ated, and relevant scenarios generated are still under economic system with underlying dynamic uncertainties
extensive study as highlighted by the literatures in these and probabilities. Furthermore, each renewable component,
sections. Acceleration techniques and efficient cuts have consumer's appliances, and electricity market all have
been developed by several authors in the decomposition different time e scales in which they are dominant.
methods approaches to speed up calculations (see Table 7). Addressing both multi e scale and multi e time e scale
Where some sampling and decomposition methods proved problems with high penetration of intermittent renewable
to be advantageous, further testing of the methods to other resources in distributed generation are the future research
renewable applications are still required. areas in the field.
v. Hybridizing existing methods with intelligent search (meta
e heuristic method); Especially in problems with higher di- The paper highlighted the recent and notable stochastic opti-
mensions (Non e Linear), accurate representation of mization approaches in renewable energy applications. The ad-
renewable systems is difficult. Intelligent searches find a vantages and challenges of stochastic optimization methods are
relaxed approximation to a solution and can reduce carefully evaluated, and its recent trending and future works are
computational expenses while increase accuracy as high- summarized in this paper. An intuitive approach was presented to
lighted in the works of [19,138,144]. The works of meta enlighten new researchers in venturing into the stochastic opti-
heuristic method in renewable energy applications are mization methods within the domain of renewable energy
mainly in the field of deterministic optimization problems applications.
[139].
Acknowledgements
While algorithms are important in solving the stochastic
renewable energy problems, future research areas in this field have The authors would like to acknowledge the financial support
also been identified from the surveys conducted. The trending from BOLD grant provided by Universiti Tenaga Nasional (UNITEN),
themes moving forward can be broken down in three main Malaysia and 20190101LRGS grant provided by Ministry of Higher
categories: Education, Malaysia.
i. Plug in EVs integration to microgrid; The surge of plug in EVs

References
are expected in the nearest future as these vehicles are more
efficient and produces relatively less greenhouse gases [145]. [1] IEA, Marke T Report Series Renewables 2O17 Marke T Report Series Re-
These EVs will lead to unique future challenges as well as newables 2O17, 2017.
[2] T. Tezer, R. Yaman, G. Yaman, Evaluation of approaches used for optimization
opportunities in future MG systems. The plug in EVs are
of stand-alone hybrid renewable energy systems, Renew. Sustain. Energy
mostly stochastic problems as charging demand profiles vary Rev. 73 (2017) 840e853. January.
from one user to another. Main themes regarding the inte- [3] A. Ioannou, A. Angus, F. Brennan, Risk-based methods for sustainable energy
system planning : a review, Renew. Sustain. Energy Rev. 74 (2017) 602e615.
gration of Plug in EVs in microgrids are identified namely;
December 2016.
Charging and scheduling of Plug in EVs [146e148], renew- [4] S. Spiecker, C. Weber, The future of the European electricity system and the
able energy integration via vehicle to grid operation [149], impact of fl uctuating renewable energy e a scenario analysis, Energy Policy
and willingness of participation towards the usage of Plug in 65 (2014) 185e197.
[5] P.A. Gooding, E. Makram, R. Hadidi, Probability analysis of distributed gen-
EVs [150]. Interested readers are encouraged to read the eration for island scenarios utilizing Carolinas data, Electr. Power Syst. Res.
recent notable works mentioned pertaining to these high- 107 (2014) 125e132.
lighted themes and its solution methodologies. [6] Q.P. Zheng, J. Wang, S. Member, A.L. Liu, Stochastic optimization for unit
commitment d a review, IEEE Trans. Power Syst. 30 (4) (2015) 1913e1924.
ii. Demand side management (DSM); Multiple authors have € Karsu, A.S. Kocaman, Bi-objective optimization of a
[7] O. Altıntaş, B. Okten, O.
started to consider load demand as an active entity as sur- grid-connected decentralized energy system, Int. J. Energy Res. 42 (2) (2018)
veyed. Results have shown reduction in peak demands, lower 447e465.
[8] Z. Abdmouleh, A. Gastli, L. Ben-Brahim, M. Haouari, N.A. Al-Emadi, Review of
costs, and reduction of generation capacities. However, the optimization techniques applied for the integration of distributed generation
success of DSM highly depends on the policies involved and from renewable energy sources, Renew. Energy 113 (2017) 266e280.
active participation of consumers. Recent review by [9] N. Sharma, Stochastic techniques used for optimization in solar systems : a
review, Renew. Sustain. Energy Rev. 16 (2012) 1399e1411.
Ref. [151] identified the consumers as one of the three main
[10] M. Aien, M. Rashidinejad, M. Fotuhi-firuzabad, On possibilistic and proba-
aspects of smart grid management. Authors have also high- bilistic uncertainty assessment of power fl ow problem : a review and a new
lighted that the acceptance of DSM varies from one con- approach, Renew. Sustain. Energy Rev. 37 (2014) 883e895.
sumer to another. It is critical that the focus of future [11] S.S. Reddy, V. Sandeep, C. Jung, Review of stochastic optimization methods
for smart grid, Energy 11 (2) (2017) 197e209.
researches is consumer e centred to improve acceptance in [12] A. Ali, N. Izzri, A. Wahab, I. Aris, J. Jasni, N. Abdalla, Computational techniques
DSM for a better management of the electrical grid. The main for assessing the reliability and sustainability of electrical power systems : a
future research directions identified regarding DSM and its review, Renew. Sustain. Energy Rev. 80 (2017) 1175e1186. June 2016.
[13] M. Aien, A. Hajebrahimi, M. Fotuhi-Firuzabad, A comprehensive review on
related recent works are; Consumer engagement methods uncertainty modeling techniques in power system studies, Renew. Sustain.
Energy Rev. 57 (2016) 1077e1089.
[15] G.E. Constante-flores, S. Member, Data-Driven Probabilistic Power Flow [43] I. Goodfellow, Generative Adversarial Networks, Dep. d’informatique Rech.
Analysis for a Distribution System with Renewable Energy Sources Using Oper. Univ. Montr., 2016.
Monte Carlo Simulation, 2017, pp. 1e8. [44] A. Creswell, T. White, V. Dumoulin, K. Arulkumaran, B. Sengupta,
[16] D.D. Chen, J. Dean Chen, Monte - Carlo Simulation - Based Statistical A.A. Bharath, Generative adversarial networks: an overview, IEEE Signal
Modeling, Springer, Singapore, 2016. Process. 18 (2018) 53e65. January.
[17] G. Carpinelli, P. Caramia, P. Varilone, Multi-linear Monte Carlo simulation [45] M. Arjovsky, S. Chintala, B. Leon, Wasserstein GAN, Courant Inst. Math. Sci.
method for probabilistic load fl ow of distribution systems with wind and Faceb. AI Res. (2017) [stat. ML], 1701.07875.
photovoltaic generation systems, Renew. Energy 76 (2015) 283e295. [46] G. Qi, Loss-Sensitive Generative Adversarial Networks on Lipschitz, (MAPLE),
[18] C. Lucia, T. Borges, An overview of reliability models and methods for dis- Univ. Cent. Florida, 2018, pp. 1e34. Lab. Mach. Percept. Learn.
tribution systems with renewable energy distributed generation, Renew. [47] A. Odena, Semi-Supervised Learning with Generative Adversarial Networks,
Sustain. Energy Rev. 16 (2012) 4008e4015. Cornell Univ, 2016, pp. 1e3, arXiv 1606.01583v2 [stat.ML].
[19] R. Dufo-Lo pez, I.R. Cristobal-Monreal, J.M. Yusta, Stochastic-heuristic meth- [48] J. Donahue, T. Darrell, P. Krahenbuhl, Adversarial Feature Learning, Cornell
odology for the optimisation of components and control variables of PV- Univ, 2017, pp. 1e18, arXiv 1605.09782v7 [cs.LG].
wind-diesel-battery stand-alone systems, Renew. Energy 99 (2016) [49] X. Chen, Y. Duan, R. Houthooft, J. Schulman, I. Sutskever, P. Abbeel, InfoGAN :
919e935. Interpretable Representation Learning by Information Maximizing Genera-
[20] S. Garshasbi, J. Kurnitski, Y. Mohammadi, A hybrid Genetic Algorithm and tive Adversarial Nets, OpenAI, 2016 arXiv: 1606.03657v1 [cs.LG].
Monte Carlo simulation approach to predict hourly energy consumption and [50] A. Odena, C. Olah, J. Shlens, Conditional Image Synthesis with Auxiliary
generation by a cluster of Net Zero Energy Buildings, Appl. Energy 179 Classifier GANs, Cornell Univ, 2017 arXiv 1610.09585v4 [stat.ML].
(2016) 626e637. [51] L. Yu, W. Zhang, J. Wang, Y. Yu, SeqGAN: Sequence Generative Adversarial
[21] J.M. Lujano-Rojas, G. Zubi, R. Dufo-Lo pez, J.L. Bernal-Agustín, J.P.S. Catala
~o, Nets with Policy Gradient, Shanghai Jiao Tong Univ. Univ. Coll, London, 2016
Novel probabilistic optimization model for lead-acid and vanadium redox arXiv 1609.05473v6 [cs.LG].
flow batteries under real-time pricing programs, Electr. Power Energy Syst [52] D. Berthelot, T. Schumm, L. Metz, “BEGAN : Boundary Equilibrium Generative
97 (2018) 72e84. June 2017. Adversarial Networks, Google, 2017, pp. 1e10, arXiv:1703.10717v4 [cs.LG].
[22] Y. Zheng, B.M. Jenkins, K. Kornbluth, C. Træholt, Optimization under uncer- [53] M. Paridah, A. Moradbak, A. Mohamed, F. abdulwahab taiwo Owolabi,
tainty of a biomass-integrated renewable energy microgrid with energy M. Asniza, S.H. Abdul Khalid, Improved Stochastic Modeling: an Essential
storage, Renew. Energy 123 (2018) 204e217. Tool for Power System Scheduling in the Presence of Uncertain Renewables,
[23] A. Mohamed, S. Kazim, L. Alvarado, Optimal sizing for a hybrid power system in Intech, 2016.
with wind/energy storage based in stochastic environment, Renew. Sustain. [54] Q. Chen, L. Mili, Composite power system vulnerability evaluation to
Energy Rev. 59 (2016) 1149e1158. cascading failures using importance sampling and antithetic variates, IEEE
[24] V.S. Lopes, C.L.T. Borges, S. Member, Impact of the combined integration of Trans. Power Syst. 28 (3) (2013) 2321e2330.
wind generation and small hydropower plants on the system reliability, IEEE [55] R. Kardooni, S.B. Yusoff, F.B. Kari, L. Moeenizadeh, Public opinion on
Trans. Sustain. Energy 6 (3) (2015) 1169e1177. renewable energy technologies and climate change in Peninsular Malaysia,
[25] H. Hashemi-dezaki, M. Hamzeh, H. Askarian-abyaneh, H. Haeri-khiavi, Risk Renew. Energy 116 (2018) 659e668.
management of smart grids based on managed charging of PHEVs and [56] S. Ntanos, G. Kyriakopoulos, M. Chalikias, G. Arabatzis, M. Skordoulis, Public
vehicle-to-grid strategy using Monte Carlo simulation, Energy Convers. perceptions and willingness to pay for renewable energy: a case study from
Manag. 100 (2015) 262e276. Greece, Sustain. Times 10 (3) (2018).
[26] P. Sarajcev, J. Vasilj, D. Jakus, Monte e Carlo analysis of wind farm lightning- [57] Q. Li, X. Wang, S. Rong, Probabilistic load flow method based on modified
surge transients aided by LINET lightning-detection network data, Renew. Latin hypercube-important sampling, Energies 11 (11) (2018) 3171.
Energy 99 (2016) 501e513. [58] Z. Shu, P. Jirutitijaroen, Latin hypercube sampling techniques for power
[27] R. Dufo-lopez, E. Pe rez-cebollada, J.L. Bernal-agustín, I. Martínez-ruiz, Opti- systems reliability analysis with renewable energy sources, IEEE Trans. Po-
misation of energy supply at off-grid healthcare facilities using Monte Carlo wer Syst. 26 (4) (2011) 2066e2073.
simulation, Energy Convers. Manag. 113 (2016) 321e330. [59] O. Dahlblom, Control Variates for Monte Carlo-Pricing of Three-Asset Spread
[28] B.R. Prusty, D. Jena, An over-limit risk assessment of PV integrated power Options with Application in the Energy Markets Oscar Dahlblom Supervised
system using probabilistic load fl ow based on multi-time instant uncertainty by, 2016.
modeling, Renew. Energy 116 (2018) 367e383. [60] R. Sun, C. Singh, L. Cheng, Y. Sun, Short-term reliability evaluation using
[29] K.-H. Chang, A quantile-based simulation optimization model for sizing control variable based dagger sampling method, Electr. Power Syst. Res. 80
hybrid renewable energy systems, Simul. Model. Pract. Theory 66 (2016) (6) (2010) 682e689.
94e103. [61] Y. Wang, An adaptive importance sampling method for spinning reserve risk
[30] M. El Ha, O. Farges, J. J. B, Global optimization of solar power tower systems evaluation of generating systems incorporating virtual power plants, IEEE
using a Monte Carlo algorithm : application to a redesign of the PS10 solar Trans. Power Syst. 33 (5) (2018) 5082e5091.
thermal power plant, Renew. Energy 119 (2018). [62] R. Rubinstein, The cross-entropy Method : a unified approach to combina-
[31] E. Zio, M. Delfanti, L. Giorgi, V. Olivieri, G. Sansavini, Monte Carlo simulation- torial optimisation, motnte - Carlo simulation and machine learning, Smart
based probabilistic assessment of DG penetration in medium voltage dis- Environ. Technol. Protoc. Appl. 34 (5/6) (2005) 903.
tribution networks, Int. J. Electr. Power Energy Syst. 64 (Jan. 2015) 852e860. [63] D.Y. Yang, J.G. Teng, D.M. Frangopol, Cross-entropy-based adaptive impor-
[32] G. Mokryani, A. Majumdar, B.C. Pal, Probabilistic method for the operation of tance sampling for time-dependent reliability analysis of deteriorating
three-phase unbalanced active distribution networks, IET Renew. Power structures, Struct. Saf. 66 (2017) 38e50.
Gener. 10 (2016) 944e954. [64] M. Leimeister, A. Kolios, A review of reliability-based methods for risk
[33] M. Di Somma, G. Graditi, E. Heydarian-forushani, M. Sha, P. Siano, Stochastic analysis and their application in the off shore wind industry, Renew. Sustain.
optimal scheduling of distributed energy resources with renewables Energy Rev. 91 (April) (2018) 1065e1076.
considering economic and environmental aspects, Renew. Energy 116 (2018) [65] C. Hamon, M. Perninge, L. So €der, An importance sampling technique for
272e287. probabilistic security assessment in power systems with large amounts of
[34] Y. Degeilh, G. Gross, Electrical Power and Energy Systems Stochastic simu- wind power, Electr. Power Syst. Res. 131 (2016) 11e18.
lation of power systems with integrated intermittent renewable resources, [66] G. Wang, J. Zhao, F. Wen, Y. Xue, G. Ledwich, S. Member, Dispatch strategy of
Electr. Power Energy Syst. 64 (2015) 542e550. PHEVs to mitigate selected patterns of seasonally varying outputs from
[35] U. Arnold, Economic risk analysis of decentralized renewable energy in- renewable generation, IEEE Trans. Smart Grid 6 (2) (2015) 627e639.
frastructures - a Monte Carlo Simulation approach, Renew. Energy 77 (2015) [67] T. Egill, Generation adequacy analysis of multi-area power systems with a
227e239. high share of wind power, IEEE Trans. Power Deliv. 33 (4) (2018)
[36] D. Pinheiro, E. Geraldo, A. Paulo, A. Traça, D. Almeida, Portfolio optimization 3854e3862.
of renewable energy assets : hydro , wind , and photovoltaic energy in the [68] L. Wang, Q. Li, R. Ding, M. Sun, G. Wang, Integrated scheduling of energy
regulated market in Brazil, Energy Econ. 64 (2017) 238e250. supply and demand in microgrids under uncertainty : a robust multi-
[37] O. Hanbury, V.R. Vasquez, Life cycle analysis of geothermal energy for power objective optimization approach, Energy 130 (2017) 1e14.
and transportation : a stochastic approach, Renew. Energy 115 (2018) [69] A.M. Leite, C. Castro, Probabilistic assessment of spinning reserve via cross-
371e381. entropy method considering renewable sources and transmission re-
[38] S. Conti, S.A. Rizzo, Monte Carlo simulation by using a systematic approach to strictions, IEEE Trans. Power Syst. 33 (4) (2018) 4574e4582.
assess distribution system reliability considering intentional islanding, IEEE [70] P. Graf, K. Dykes, R. Damiani, J. Jonkman, P. Veers, Adaptive stratified
Trans. Power Deliv. 30 (1) (2015) 64e73. importance sampling : hybridization of extrapolation and importance sam-
[39] M. Bugallo, V. Elvira, L. Martino, D. Luengo, J. Míguez, P.M. Djuric, Adaptive pling Monte Carlo methods for estimation of wind turbine extreme loads,
Importance Sampling: the past, the present, the future, IEEE Signal Process. Wind Energy Sci. 3 (1) (2018) 475e487.
17 (2017) 60e78. July. [71] J. Huang, Y. Xue, Z.Y. Dong, K.P. Wong, An efficient probabilistic assessment
[41] Y. Chen, Y. Wang, D. Kirschen, B. Zhang, Model-Free renewable scenario method for electricity market risk management, IEEE Trans. Power Syst. 27
generation using generative adversarial networks, IEEE Trans. Power Syst. 33 (3) (Aug. 2012) 1485e1493.
(2018) 3265e3275. [72] D. Van Ravenzwaaij, P. Cassey, S.D. Brown, A Simple Introduction to Markov
[42] S. Talari, M. Sha, G.J. Oso rio, J. Aghaei, Stochastic modelling of renewable Chain Monte e Carlo Sampling, 2018.
energy sources from operators ’ point-of- view : a survey, Renew. Sustain. [73] P.J. Green, K. Łatuszy, M. Pereyra, C.P. Robert, Bayesian computation : a
Energy Rev. 81 (2018) 1953e1965. June 2017. summary of the current state , and samples backwards and forwards, Stat.
Comput. (2015) 835e862. under uncertainty using Benders decomposition method, Energy 143 (2018)
[74] A. Gelman, D.B. Rubin, A. Gelman, D.B. Rubin, Inference from iterative 1029e1047.
simulation using multiple sequences, Inst. Math. Stat. 7 (4) (2018) 457e472. [103] E. Kianmehr, S. Nikkhah, A. Rabiee, Multi-objective stochastic model for joint
[75] W.R. Gilks, S. Richardson, D.J. Spiegelhalter, Markov chain Monte Carlo in optimal allocation of DG units and network recon fi guration from DG owner
practice, J. Appl. Stat. 25 (4) (1998) 575. ’ s and DisCo ’ s perspectives, Renew. Energy 132 (2019) 471e485.
[76] S.M. Kandil, H.E.Z. Farag, M.F. Shaaban, M.Z. El-sharafy, A combined resource [104] A. Samimi, M. Nikzad, P. Siano, Scenario-based stochastic framework for
allocation framework for PEVs charging stations , renewable energy re- coupled active and reactive power market in smart distribution systems
sources and distributed energy storage systems, Energy 143 (2018) with demand response programs, Renew. Energy 109 (2017) 22e40.
961e972. [105] P. Carpentier, J. Chancelier, G. Cohen, M. De Lara, Stochastic Multi - Stage
[77] H. Carstens, X. Xia, S. Yadavalli, Measurement uncertainty in energy moni- Optimization: at the Crossroads between Discrete Time Stochastic Control
toring : present state of the art, Renew. Sustain. Energy Rev. 82 (2018) and Stochastic Programming, Springer, France, 2010.
2791e2805. October 2017. [106] T. Shiina, J.R. Birge, Stochastic unit commitment problem, Int. Trans. Oper.
[78] S.P. Evans, P.D. Clausen, Modelling of turbulent wind flow using the Res. 11 (1) (Jan. 2004) 19e32.
embedded Markov chain method, Renew. Energy 81 (2015) 671e678. [107] S. Takriti, J.R. Birge, E. Long, A stochastic model for the unit commitment
[79] X. Zhang, B. Wang, Probability assessment for power system voltage stability problem, IEEE Trans. Power Syst. 1 (1996) 1497e1508. August.
margin with renewable energy source using slice sampling, Int. Trans. Electr. [108] S. Sen, L. Yu, T. Genc, A stochastic programming approach to power portfolio
Energy Syst. (2017) 1e13, no. March. optimization, informs 54 (1) (2006) 55e72.
[80] A.F.M. Smith, G.O. Roberts, Bayesian computation via the Gibbs sampler and [109] M.P. Nowak, W. Romisch, Stochastic Lagrangian relaxation applied to power
related Markov chain Monte Carlo methods, J. R. Stat. Soc. Ser. B 55 (1) scheduling in a hydro-thermal system under uncertainty, Ann. Oper. Res.
(1993) 3e23. 100 (2001) 251e272.
[81] C.J.F. Ter Braak, A Markov chain Monte Carlo version of the genetic algorithm [110] R. Jiang, S. Member, J. Wang, S. Member, M. Zhang, Two-stage minimax
Differential Evolution : easy Bayesian computing for real parameter spaces, regret robust unit commitment, IEEE Trans. Power Syst. 28 (3) (2013)
Stat. Comput. 16 (2006) 239e249. 2271e2282.
[82] M. Neal, Slice sampling, Inst. Math. Stat. 31 (3) (2003) 705e741. [111] R. Jiang, S. Member, J. Wang, Y. Guan, A. Sets, Power and pumped storage
[83] S. Kirkpatrick, C.D. Gelatt, M.P. Vecchi, Optimization by simulated annealing, hydro, IEEE Trans. Power Syst. 27 (2) (2012) 800e810.
Science 220 (1983) 4598 (80-. ). [112] M. Asensio, P.M. De Quevedo, G. Mun ~ oz-delgado, S. Member, J. Contreras,
[84] A. Street, F. Oliveira, J.M. Arroyo, Contingency-constrained unit commitment Joint distribution network and renewable energy expansion planning
with n K security Criterion : a robust optimization approach, S. Member, and considering demand response and energy storage d Part I : stochastic pro-
A. Functions, IEEE Trans. Power Syst. 26 (3) (2011) 1581e1590. gramming model, IEEE Trans. Smart Grid 9 (2) (2018) 655e666.
[85] S. Choi, A. Hussain, H. Kim, Adaptive robust optimization-based optimal [113] Y. Zhan, Q.P. Zheng, J. Wang, S. Member, Generation expansion planning
operation of microgrids considering uncertainties in arrival and departure with large amountsof wind power via decision-dependent stochastic pro-
times of electric vehicles, Energies 11 (2646) (2018) 1e16. gramming, IEEE Trans. Power Syst. 32 (4) (2017) 3015e3026.
[86] Y. Zhang, J. Le, F. Zheng, Y. Zhang, K. Liu, Two-stage distributionally robust [114] F. Santos, D.Z. Fitiwi, S. Member, A.W. Bizuayehu, M. Shafie-khah, M. Asensio,
coordinated scheduling for gas- electricity integrated energy system J. Contreras, C. Manuel, P. Cabrita, P.S. Catal, Novel multi-stage stochastic DG
considering wind power uncertainty and reserve capacity con fi guration, investment planning with recourse, IEEE Trans. Sustain. Energy 8 (1) (2017)
Renew. Energy 135 (2019) 122e135. 164e178.
[87] A. Hussain, V.-H. Bui, H. Kim, Multi-microgrids considering uncertainties, [115] G. Luiz, M. Fredo, E. Cristian, V. Luiz, D. Matos, Assessing solution quality and
Energies 9 (2016) 278. computational performance in the long-term generation scheduling problem
[88] M. Sharafi, T.Y. ElMekkawy, Stochastic optimization of hybrid renewable considering different hydro production function approaches, Renew. Energy
energy systems using sampling average method, Renew. Sustain. Energy Rev. 131 (2019) 45e54.
52 (2015) 1668e1679. [116] H. Abdi, S.D. Beigvand, M. La Scala, A review of optimal power flow studies
[89] Q.P. Zheng, J. Wang, P.M. Pardalos, A Decomposition Approach to the Two- applied to smart grids and microgrids, Renew. Sustain. Energy Rev. 71 (2017)
Stage Stochastic Unit Commitment Problem, Springer , Sci. þ Bus. Media, 742e766. May 2015.
2013, pp. 387e410. [117] J. Go€nsch, M. Hassler, Sell or Store ? an ADP Approach to Marketing
[90] R.A. Jabr, Tight polyhedral approximation for mixed-integer linear pro- Renewable Energy, vol. 38, Springer, OR Spectr., 2016, pp. 633e660.
gramming unit commitment formulations, IET Gener., Transm. Distrib. 6 [118] J. Shin, J.H. Lee, M.J. Realff, Operational planning and optimal sizing of
(2012) 1104e1111. July. microgrid considering multi-scale wind uncertainty, Appl. Energy 195
[91] A. Papavasiliou, S.S. Oren, Multiarea stochastic unit commitment for high (2017) 616e633.
wind penetration in a transmission constrained network, Oper. Res. 61 (3) [119] S. Elfwing, E. Uchibe, K. Doya, From free energy to expected energy :
(2013) 578e592. improving energy-based value function approximation in reinforcement
[92] M.P. Nowak, R. Udiger, M. Westphalen, A stochastic integer programming learning, Neural Netw. 84 (2016) 17e27.
model for incorporating day-ahead trading of electricity into hydro-thermal [122] D.P. Bertsekas, Approximate policy iteration : a survey and some new
unit commitment, Optim. Eng. 6, 2005, pp. 163e176. methods, J. Control Theor. Appl. 9 (2011) 310e335. January.
[93] H. Rashidizadeh-Kermani, M. Vahedipour-Dahraie, H. Najafi, A. Anvari- [123] G. Serale, M. Fiorentini, A. Capozzoli, D. Bernardini, A. Bemporad, Model
Moghaddam, J. Guerrero, A stochastic Bi-level scheduling approach for the predictive control ( MPC ) for enhancing building and HVAC system energy
participation of EV aggregators in competitive electricity markets, Appl. Sci. 7 efficiency, Energies 11 (631) (2018) 1e35.
(10) (2017) 1100. [124] W.R. Sultana, S. Kumar, S. Sukchai, S. Yamuna, D. Venkatesh, A review on
[94] M. Vahedipour-Dahraie, A. Anvari-Moghaddam, J.M. Guerrero, Evaluation of state of art development of model predictive control for renewable energy
reliability in risk-constrained scheduling of autonomous microgrids with applications, Renew. Sustain. Energy Rev. 76 (2017) 391e406. January 2016.
demand response and renewable resources, IET Renew. Power Gener. 12 (6) [125] N. Bazmohammadi, A. Tahsiri, A. Anvari-Moghaddam, J.M. Guerrero,
(2018) 657e667. A hierarchical energy management strategy for interconnected microgrids
[95] M. Vahedipour-Dahraei, H.R. Najafi, A. Anvari-Moghaddam, J.M. Guerrero, considering uncertainty, Int. J. Electr. Power Energy Syst. 109 (2019)
Security-constrained unit commitment in AC microgrids considering sto- 597e608. October 2018.
chastic price-based demand response and renewable generation, Int. Trans. [126] A. Afram, F. Janabi-sharifi, A.S. Fung, K. Raahemifar, Artificial neural network
Electr. Energy Syst. 28 (9) (Sep. 2018) e2596. ( ANN ) based model predictive control ( MPC ) and optimization of HVAC
[96] M. Vahedipour-Dahraie, H. Rashidizadeh-Kermani, H.R. Najafi, A. Anvari- systems : a state of the art review and case study of a residential HVAC
Moghaddam, J.M. Guerrero, Stochastic security and risk-constrained sched- system, Energy Build. 141 (2017) 96e113.
uling for an autonomous microgrid with demand response and renewable [127] A. Afram, F. Janabi-shari, Theory and applications of HVAC control systems e
energy resources, IET Renew. Power Gener. 11 (14) (2017) 1812e1821. A review of model predictive control ( MPC ), Build. Environ. 72 (2014)
[97] R. Hemmati, H. Saboori, S. Saboori, Assessing wind uncertainty impact on 343e355.
short term operation scheduling of coordinated energy storage systems and [128] Z. Ji, X. Huang, C. Xu, S. Houtao, Accelerated model predictive control for
thermal units, Renew. Energy 95 (2016) 74e84. electric vehicle integrated microgrid energy Management : a hybrid robust
[98] V.S. Tabar, M.A. Jirdehi, R. Hemmati, Energy management in microgrid based and stochastic approach, Energies 9 (973) (2016) 1e18.
on the multi objective stochastic programming incorporating portable [129] U. Yildiran, I. Kayahan, Risk-averse stochastic model predictive control-
renewable energy resource as demand response option, Energy 118 (2017) based real-time operation method for a wind energy generation system
827e839. supported by a pumped hydro storage unit, Appl. Energy 226 (2018)
[99] H. Alharbi, A. Indices, B. Parameters, Stochastic optimal planning of battery 631e643. October 2017.
energy storage systems for isolated microgrids, IEEE Trans. Sustain. Energy 9 [130] Y. Shi, X. Xiang, L. Wang, Y. Zhang, D. Sun, Stochastic model predictive fault
(1) (2018) 211e227. tolerant control based on conditional value at risk for wind energy conver-
[100] S. Kwon, L. Ntaimo, N. Gautam, S. Member, Optimal day-ahead power sion system, Energies 11 (2018) 193.
procurement with renewable energy and demand response, IEEE Trans. [131] M. Rahmani-andebili, Scheduling deferrable appliances and energy re-
Power Syst. 32 (5) (2017) 3924e3933. sources of a smart home applying multi-time scale stochastic model pre-
[101] E. Mortaz, J. Valenzuela, Microgrid energy scheduling using storage from dictive control, Sustain. Cities Soc. 32 (2017) 338e347. January.
electric vehicles, Electr. Power Syst. Res. 143 (2017) 554e562. [132] A. Parisio, E. Rikos, L. Glielmo, Stochastic model predictive control for eco-
[102] S. Hemmati, S.F. Ghaderi, M.S. Ghazizadeh, Sustainable energy hub design nomic/environmental operation management of microgrids : an
experimental case study, J. Process Control 43 (2016) 24e37. Optimization Algorithms,” Prox. Policy Optim. Algorithms, Cornell Univ. Libr.
[133] P. Kou, F. Gao, X. Guan, Stochastic predictive control of battery energy arXiv.org, Ithaca, 2017, pp. 1e12.
storage for wind farm dispatching : using probabilistic wind power forecasts, [144] M. Sharafi, T.Y. ElMekkawy, Stochastic optimization of hybrid renewable
Renew. Energy 80 (2015) 286e300. energy systems using sampling average method, Renew. Sustain. Energy
[134] P. Kou, D. Liang, L. Gao, Distributed EMPC of multiple microgrids for coor- Rev. 52 (2015) 1668e1679.
dinated stochastic energy management, Appl. Energy 185 (2017) 939e952. [145] S. Howell, Y. Rezgui, J. Hippolyte, B. Jayan, H. Li, Towards the next generation
[135] Y. Zhang, T. Zhang, R. Wang, Y. Liu, B. Guo, ScienceDirect Optimal operation of smart grids : semantic and holonic multi- agent management of distrib-
of a smart residential microgrid based on model predictive control by uted energy resources, Renew. Sustain. Energy Rev. 77 (2017) 193e214.
considering uncertainties and storage impacts, Sol. Energy 122 (2015) March.
1052e1065. [146] A. Ovallle, A. Hably, S. Bacha, Grid Optimal Integration of Electric Vehicles :
[136] X. Guo, Z. Bao, Z. Li, W. Yan, Adaptively constrained stochastic model pre- Examples with Matlab Implementation, Springer, Poland, 2018.
dictive control for the optimal dispatch of microgrid, Energies 11 (243) [147] D. Wang, J. Coignard, T. Zeng, C. Zhang, S. Saxena, Quantifying electric vehicle
(2018) 1e17. battery degradation from driving vs . vehicle-to-grid services, J. Power
[137] P. Velarde, L. Valverde, J.M. Maestre, C. Ocampo-martinez, C. Bordons, On the Sources 332 (2016) 193e203.
comparison of stochastic model predictive control strategies applied to a [148] A.W. Thompson, Economic implications of lithium ion battery degradation
hydrogen-based microgrid, J. Power Sources 343 (2017) 161e173. for Vehicle-to- Grid ( V2X ) services, J. Power Sources 396 (2018) 691e709.
[138] M. Rahmani-Andebili, Stochastic, adaptive, and dynamic control of energy June.
storage systems integrated with renewable energy sources for power loss [149] S. Rajakurna, F. Shahnia, A. Ghosh, Plug in Electric Vehicles in Smart Grids,
minimization, Renew. Energy 113 (2017) 1462e1471. Springer, Singapore, 2015.
[139] S. Twaha, M.A.M. Ramli, A review of optimization approaches for hybrid [150] J. Geske, D. Schumann, Willing to participate in vehicle-to-grid ( V2G )? Why
distributed energy generation systems: off-grid and grid-connected systems, not !, Energy Policy 120 (2018) 392e401. May.
Sustain. Cities Soc. 41 (2018) 320e331. May. [151] O. Ellabban, H. Abu-rub, “Smart grid customers ’ acceptance and engage-
[140] S.M. Zahraee, M. Khalaji Assadi, R. Saidur, Application of artificial intelligence ment : an overview, Renew. Sustain. Energy Rev. 65 (2016) 1285e1298.
methods for hybrid energy system optimization, Renew. Sustain. Energy Rev. [152] A. Mengolini, F. Gangale, J. Vasiljevska, Exploring community-oriented ap-
66 (2016) 617e630. proaches in demand side management projects in Europe, Sustainability 8
[141] B. Bhandari, K.-T. Lee, G.-Y. Lee, Y.-M. Cho, S.-H. Ahn, Optimization of hybrid (2016) 1266.
renewable energy power systems: a review, Int. J. Precis. Eng. Manuf. [155] B. Amarasekara, C. Ranaweera, R. Evans, A. Nirmalathas, Computers and
Technol. 2 (1) (2015) 99e112. Operations Research Hierarchical aggregation method for a scalable imple-
[142] Y. Chen, An unsupervised deep learning approach for scenario forecasts, Inst. mentation of demand side management, Comput. Oper. Res. 96 (2018)
Electr. Electron. Eng. Inc. (IEEE), Piscataw. (2018) 1e7. 188e199.
[143] J. Schulman, F. Wolski, P. Dhariwal, A. Radford, O. Klimov, “Proximal Policy

Paper - 11

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

Paper - 11

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Paper - 11

Uploaded by

Copyright:

Available Formats

Renewable Energy 145 (2020) 1543e1571

Contents lists available at ScienceDirect

Uncertainty models for stochastic optimization in renewable energy

5.1.1. Two e stage models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1556

1. Introduction from huge computational expenses due to large number of sce-

Fig. 1. Stochastic optimization ﬂowchart.

Fig. 2. Uncertainty modelling overview.

determine the optimal amount of power generation to be

reduction technique (Roulette e Wheel Mechanism) to reduce the

geothermal plants LCA with

scenarios that spans over a period of 300 months. Authors in

processes of intermittent renewable resources and simulated the

ysis of decentralized renewable energy infrastructures has been

MCS in terms of risks

3.1.2.3. Non e sequential MCS applications in renewable energy.

Few recent works that have implemented the non e sequential

Drilling time, geothermal

Well ﬂuid composition,

Vasquez, 2018 employed the usage of MCS in geothermal plant's

construction to stochastically capture the environmental impact

in terms of complete Life Cycle Analysis (LCA) relative to other

methods of energy production [37]. A systematic approach of MCS

to address the system distribution reliability considering inten-

3.1.2.4. Trending of MCS applications in renewable energy.

heuristic method typically acted as a space search reducer for the

MCS method in performing the stochastic optimization as shown

and SAIDI (refer to paper)

Random MCS (RMCS) with

Historical Samples Generated Samples

Real Output Prediction Fake

Fig. 4. Example of GANs structure for wind scenario generation.

References Type of GANs Characteristic Main advantage(s) Future work(s)

Variance Reduction Techniques

Cross Entropy Adaptive Importance Sequential Importance

Fig. 5. Overview of variance reduction techniques (VRTs).

intermittent behaviour of renewable generations such as wind, 4.1.1. Type of IS method

Where, E denotes the outputs of PV system/wind generator, CBDP is

Summary of the references mentioned above is speciﬁed in  .

4.2. Notable sampling method: Markov Chain Monte Carlo method .

Generation of Mean wind speed

Generation of Large time step

Step 1: Weibull PDF

Step 4: The power flow

References Type of MCMC Advantages Disadvantages

Fig. 10. General overview of stochastic optimization [85].

gramming (ASDP) as shown in Fig. 6. The paper's scope is focused " #

Collect load data obtained from customers'

Determine the demand

Determine total demand of all customers (shiftable,

Input historical data of wind and

Obtain mean and standard deviation through

forecasted and historical data

Scenario generation for load, electricity price, wind

Scenario reduction to Ns Scenario by using K-means

dispatch without network constraints)

Subproblem for scenario s

Create cut for NO

Summary of the references mentioned above is speciﬁed in .

SOCmin x2 ðk þ hjkÞ SOCmax; h ¼ 1; 2; :::; H (33) X