Electric Power Systems Research 182 (2020) 106232

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

Towards efficient energy management in smart grids considering microgrids T

with day-ahead energy forecasting
Sheraz Aslama,b,*, Adia Khalidb, Nadeem Javaidb
a
Department of Electrical Engineering, Computer Engineering, and Informatics, Cyprus University of Technology, Limassol 3036, Cyprus
b
Department of Computer Science, COMSATS University Islamabad, Islamabad 44000, Pakistan

A R T I C LE I N FO A B S T R A C T

Keywords: This study proposes an efficient energy management method to systematically manage the energy consumption
Artificial neural network in the residential area to alleviate the peak to average ratio and mitigate electricity cost along with user comfort
Efficient energy utilization maximization. We developed an efficient energy management scheme using mixed integer linear programming
Energy forecasting (MILP), which schedules smart appliances and charging/discharging of electric vehicles (EVs) optimally in order
Home energy management
to mitigate energy costs. In the proposed model, consumer is able to generate its own energy from microgrid
Mixed integer linear programming
Renewable energy generation
consisting of solar panels and wind turbines. We also consider an energy storage system (ESS) for efficient energy
utilization. This work also performs energy forecasting using wind speed and solar radiation prediction for
efficient energy management. Moreover, we perform extensive simulations to validate our developed MILP
based scheme and results affirm the effectiveness and productiveness of our proposed energy efficient technique.

1. Introduction external grid. The microgrid generates electricity from renewable en-
ergy sources (RESs) that include wind turbines (WTs), solar panels,
The traditional power system totally relies on fossil fuel-based tidal, etc., which contribute a lot in carbon emission minimization.
plants and energy is produced at a single place that is far from con- Moreover, RESs are unreliable sources of energy generation due to their
sumers. According to [1], the traditional power system bears energy intermittent nature. So, energy storage systems (ESSs) are considered
losses because of a huge distance among energy users and generation. necessary to utilize renewable energy in an efficient way.
Therefore, transmission and distribution lines are the key reason for Demand response (DR) plays a significant role in a reliable and cost-
blackouts. Therefore, to tackle these situations, electricity plants are efficient energy system operation [3]. DR program motivates the end-
needed that are near to the consumption areas. This definitely refers to users to manage their load demand or provide incentives for electricity
the greener energy sources like microgrid. When a maximum portion of bills and peak alleviation. A consumer can reduce cost and improve
electricity consumption is produced from the microgrid, the high satisfaction-level by the scheduling of home appliances. The DR pro-
carbon emission problem is automatically solved, where approximately, grams considering residential electricity usage have been widely stu-
41% carbon footprint is emitted by the energy sector and transport died. In [4], DR problem has been studied considering multiple homes
sector produces 23% of total greenhouse gases around the globe [2]. with the same living patterns. However, it is not practically possible for
That is why the traditional power system is not feasible to fulfill the multiple homes to have the same appliances with the same length of
future requirements of electricity consumers. The existing power sys- operational time (LOT) and power ratings (kW). An efficient scheduling
tems also cause high carbon emissions due to the heavy dependence on scheme can optimally determine the schedule of appliances consump-
fossil fuels. tion. As a result, the scheduling pattern depicts the same behavior as
The minimum usage of fossil fuels and low carbon emissions are the microgrid electricity generation pattern. In this way, consumers can
possible with the evolution of microgrids. Furthermore, the microgrid save energy costs and imports minimum energy from the external grid.
connects with the smart home in two ways, such as islanded and grid- Furthermore, efficient scheduling may provide reliable and stable mi-
connected modes. In islanded mode, external grid and microgrid cannot crogrid operation.
sell/purchase energy with each other. However, the latter discussed The authors of [5] developed an energy management controller to
mode allows the microgrid to sell and purchase electricity to/from the alleviate the peak for a single home. RESs are also installed in their

⁎
Corresponding author.
E-mail address: [email protected] (S. Aslam).

https://doi.org/10.1016/j.epsr.2020.106232
Received 25 July 2019; Received in revised form 10 November 2019; Accepted 14 January 2020
Available online 22 January 2020
0378-7796/ © 2020 Elsevier B.V. All rights reserved.
S. Aslam, et al. Electric Power Systems Research 182 (2020) 106232

work to mitigate energy cost and efficient system operation. However, developed by the researcher to alleviate electricity cost and PAR with a
they consider day-ahead weather information. [6] also proposes a reduction in imported electricity from the external grid. In this section,
scheme for cost reduction and weather parameters are also known day- there have been few studies on the advantages of DR schemes in the
ahead basis. Mosaddek et al. [7] propose a mathematical based sche- residential area.
duling method that schedules home appliances along with electrical An appliances scheduling problem is solved in [10] for peak load
vehicle (EV) charging/discharging. Furthermore, in their work, an EV and energy cost minimization. Three metaheuristic approaches are
works as a mobile storage. The authors of the paper [8] develop a exploited for appliances scheduling. Experimental analysis shows the
building energy method for peak alleviation. They also considered an efficiency of developed methods with regards to bills and peak reduc-
EV as a storage device to enhance the energy efficiency in building with tion. In [11], a game theory-based appliances’ scheduling scheme is
regards cost reduction. The work [9] develops a hybrid energy system adopted. The primary objectives of their paper are energy cost and PAR
that considers the solar panel along with WT and ESS. In addition, it reduction. Their developed scheme efficiently attains the defined ob-
also considers a diesel generator as an extra source of power generation; jectives. Nevertheless, the comfort of the user is not considered in this
however, it produces energy against heavy cost and it is considered the work. Another optimization technique is developed using the heuristic
main source of carbon footprint. Moreover, energy forecasting is also algorithm in [12]. Particle swarm optimization (PSO) is implemented
performed for efficient energy management. for household appliances’ scheduling. It shifts the smart appliances
The work proposed in [5]– [9] either focused on specific production from ON peak to OFF peak time intervals and eventually, the electricity
application or failed to take full advantages from the smart grid tech- bill is reduced along with the low computational cost. In [13], the
nologies to develop DR techniques. The motivation of our work is to authors proposed a scheduling technique along with ToU pricing signals
devise an optimal home energy management (HEM) scheme to manage to solve the plants scheduling problem in industrial area. The problem
the smart home activities and each user is connected with the micro- is formulated as continuous-time model and the proposed scheme re-
grid. In this work, we investigate a residential sector with microgrid; duces the electricity cost; however, it can be implemented only in small
consisting of WT and solar panel, which is connected to the commercial size problems due to high computation cost. This turned to discrete-
grid. A micro gas turbine (MGT) is installed instead of a diesel generator time models, where discrete events can be handled in a more natural
as a backup resource for fuel cost and carbon emission minimization. and straightforward way.
We install mobile storage (EV) as well as the static storage simulta- Luna Adriana et al. developed an efficient method for energy
neously, because, the EV's presence is uncertain (sometimes, either management for electricity expenses’ reduction and RESs integration
available at home or goes for driving). When EV is not available then [14]. The basic purpose of RESs integration is to minimize imported
demand is met by the static batteries. Furthermore, a day-ahead electricity. Solar panel and WT with battery storage system are con-
weather forecasting model based on artificial neural network (ANN) is sidered in their work as RESs. Simulation results validate their proposed
developed for more accurate wind speed and solar radiation prediction. system model; however, RESs installation cost is not considered in this
This work considers a residential household that has several appli- work. A real time pricing (RTP) based DR management scheme is
ances. All the appliances have various parameters, e.g., start time, end presented using stochastic and robust optimization [15]. They perform
time, power ratings, etc. In addition, the smart home has its own grid- Monte-Carlo simulations and consider time slot 5-min for each interval.
connected microgrid that contains WT and solar panel along with ESS. Results validate their proposed scheme in terms of electricity bill re-
The microgrid is further connected with the EV and commercial grid. duction. In [16], a cost minimization scheme is adopted using PL-
This work also installs the static storage system along with mobile generalized benders algorithm in the residential sector. Multiple homes
storage (EV) for energy storage. The MGT is installed as a backup source with multi types of appliances are considered in the work. Furthermore,
of electricity and enhanced differential evolution (EDE) based ANN simulation analysis shows that the developed method achieves their
(EDE-ANN) is used for more accurate weather forecasting instead of a desired objective efficiently. Another non-sorting GA-based approach is
day ahead assumed weather information. adopted in [17] for building energy management. The basic objectives
of this study are to mitigate cost, energy consumption and carbon
• This paper tackles appliances scheduling problem to mitigate the emission. The standard building model is developed in this work to
peak to average ratio (PAR) and electricity bills in the residential implement their proposed technique. Results uncover that the proposed
sector. method minimizes up to the 40% carbon emission. A multi-agent
• Based on the proposed MILP based scheme, efficient integration of system (MAS) based HEMS is developed in [18] to balance between
microgrid is achieved. demand and supply. The solar panel and fuel cells along with the sto-
• For proper energy management, an accurate weather forecasting rage system are also considered for safe power delivery and recovery.
model (EDE-ANN) is developed. Experimental results show the efficacy of their developed MAS based
• The proposed solution enables a smart home to interact with the HEMS in terms of their defined objectives.
microgrid and power grid, and then make autonomous decisions Mohammad et al. propose a new control algorithm for HEMS to
(purchasing electricity) on the bases of power generation and hourly schedule and monitor the electrical appliances [19]. They develop a
energy tariff. novel architecture while shifting electrical appliances; when the price is
• Results from simulations depict that the newly developed scheme high, the load is shifted to battery bank instead of other time slots to
significantly alleviates the consumer's bill and PAR along with a reduce consumer's discomfort. The solar panel is also considered in this
reduction in load burden on the commercial grid. work and real time testbed is used to implement the proposed archi-
tecture. It is observed from the results, electricity cost is reduced up to
We have organized the remainder of the work as related studies are 15% along with minimum comfort violation. The authors of [20] pro-
presented in Section 2. Section 3 explains the proposed energy efficient pose a MILP based energy management system in the residential sector.
model, and mathematical formulations of the scheduling problem and They also investigate the EVs having bi-directional power flow features.
objective function are uncovered in Section 4. Section 5 unfolds the It is verified from simulation results that the developed scheme shows
simulation results and the key findings of this work are elaborated in efficacy with regards to minimum cost. In [21], an optimal scheduling
Section 6. method for smart buildings is proposed using MILP. Buildings’ heating
and electricity generation are considered simultaneously in this work.
2. Related work The proposed scheme efficiently achieves the defined objectives. An
optimization technique based on genetic algorithm (GA) and MILP is
In the last few decades, numerous DR techniques have been proposed to tackle the unit commitment and economic dispatch

2
S. Aslam, et al. Electric Power Systems Research 182 (2020) 106232

problem in [22]. Both techniques show high performance; however, the Table 1
computation time of MILP increases when constraints are increased. On Nomenclature
the contrary, GA shows efficacy as compared to MILP for maximum Acronyms/Notations Explanations
constraints. A dispatch model based on cost-saving is presented in [23].
They consider electricity selling from the residential sector to office DR Demand response
EDE Enhanced differential evolution
building which is generated from the microgrid. EVs are also taken into
EMC Energy management controller
account in their work. Numerical results verify the effectiveness and ESS Energy storage system
present that their developed model not only minimizes the electricity EV Electrical vehicle
cost, it also maximizes the earning from electricity dispatching to the HEM Home energy management
office building. The microgrid charges EVs at low price hours and sells LOT Length of operational time
MGT Micro gas turbine
the electricity at high price hours by discharging EVs.
MILP Mixed integer linear programming
RES Renewable energy sources
3. Proposed energy efficient model RTP Real time pricing
WT Wind turbine
ANN Artificial neural network
This section first presents the system model and then solar and wind AN Artificial neuron
energy forecasting model has been elaborated. Mathematical notations
U Set of electricity users/homes
u Each user in U
3.1. System model t Single time interval
A Set of all appliance in home u
This work considers a smart home u having a smart meter and a Each appliance in home u
APPsh Set of shiftable appliances
different smart appliances. A microgrid, which consists of WT and solar
APPns Set of non-shiftable appliances
panel is also considered. However, electricity generation from micro- Γtsh Hourly electricity consumption against APPsh
grid is stochastic in nature; hence, for intelligent energy consumption ΓTsh Per day electricity consumption against APPsh
management, an ANN-based prediction model has been developed. Γtns Hourly electricity consumption against APPns
Furthermore, an energy management controller (EMC) is installed for ΓTns Per day electricity consumption against APPns
intelligent appliances’ scheduling according to pricing signals and γsh Power rating of shiftable appliances
electricity generation. The simulation's entire time interval is denoted γns Power rating of non-shiftable appliances
by T and assumed one day in this work. One day is further divided into αmgt ON or OFF status of MGT
sub-intervals shown by t (1 h) and each of which has equal time in- ζt Hourly electricity consumption cost
terval. Furthermore, Table 1 explains the acronyms and mathematical Φt Hourly imported electricity
ΦT Total imported electricity for T
notations, which are used in our proposed model. This work (extended
EPt Electricity price against single time slot
system model is depicted in Fig. 1) is the extension of [24] and the
Et Microgrid generation against single time interval
components of the newly proposed system model have been explained ET Microgrid generation for whole day
below. εm Electricity generation from single microgrid source
ptwt Power generation from WT

3.2. Short-term weather forecasting using ANN Cp Power co-efficient of WT

Vtwt Hourly wind velocity

This section presents the forecasting model of the wind speed and ptpv Produced power from solar panel
Apv Solar panel area
solar radiation for accurate power generation estimation. For this
η pv Solar panel efficiency
purpose, we develop an EDE-ANN based forecasting model to forecast Irrt Irradiation for interval t
short-term weather values. Hereupon, we propose a DR scheme using Tempt Temperature of the day
MILP for minimizing the cost, PAR and imported electricity from the EVe Electricity consumption against EV
commercial grid. The developed model consists of three key modules: i) ψe Charging/discharging of EV in (kWh)
feature selector, ii) forecaster and iii) optimizer. In the first step, the αe Charging/discharging or remaining idle status of EV
inputs (time series wind speed and solar radiation) are given to module ρEV ae Energy storage level at EV arrival time
ρEVde Energy storage level at EV departure time
1; this module uses mutual information (MI) to rank the inputs. Then max
EVdis Maximum EV discharging limit
the ranked inputs are passed to filters for removing irrelevancy and
SE Stored electricity in ESS
redundancy. After that, from the selected inputs, training and valida- ESS charging at slot t
EStch
tion sets are established based on the previously observed data. The
ESS t Total electricity storage in ESS
developed model forecasts the wind speed and solar radiation for the ESS discharging at time t
EStdis
next day on the bases of training and validation samples, and the
forecasted values are given to optimizer. It calculates the error signal
between forecasted and target values, and then this error is reduced by 1
f (S, b) = ,
using iteration based EDE. In particular, PSO, GA, differential evolution 1 + e−β (S − b) (1)
(DE) and ant colony optimization are the widely adapted heuristic
based algorithms in energy and load forecasting problems. Among these where S denotes input signal that consists of selected features (selected
algorithms, DE has relatively better performance in terms of avoiding by first module of our proposed model as discussed above). The term b
premature convergence while searching for optimal solution. The is the bias value and the parameter β controls steepness of the activa-
reason is difference of the vectors-based approach in DE expanding its tion function. Furthermore, as discussed before, the proposed model
search space [26]. In this regard, we are motivated from [27,28] and must be enabled for learning to predict the future values of the weather.
[29,30] that use DE techniques and GAs, respectively, and we used There are three types of learning: i) supervised learning, ii) un-
enhanced DE to optimize the initial weights of proposed forecasting supervised learning and iii) reinforcement learning. In this work, our
model to avoid premature and fast convergence prediction model learns through supervised learning due to time series
In addition, each artificial neuron (AN) in the hidden and output data and we adopt a multivariate autoregressive rule for supervised
layers exploits Sigmoid activation function [31], presented in Eq. (1). learning because of relatively high convergence rate as compared to the

3
S. Aslam, et al. Electric Power Systems Research 182 (2020) 106232

Fig. 1. The developed energy-efficient model.

other rules [32]. The developed forecasting model is presented in Fig. 2. along with optimal EV charging/discharging is formulated in Eq. (3)
Furthermore, the weather data for three years is taken from [33] and and constraints are presented in Eqs. (4)–(12). Furthermore, the for-
the data consists of four values in an hour (each for 15 minutes). The mulated problem contains various decision variables, like linear and
entire data is separated into three datasets, e.g., training, validation and integer variables. Therefore, this problem forms a MILP model. We can
testing sets, as 70%, 15% and 15%, respectively. The dataset division achieve the optimal scheduling for home appliances and EVs according
ratio is adapted from the existing literature [34–36]. Firstly, the to the produced power from RESs; consequently, total electricity cost is
training dataset is used to train the forecasting model and the validation reduced.
dataset helps in overcoming the overfitting problem. After that the
testing dataset is fed to the EDE-ANN based forecasting model to get the 4.1. Objective function
forecasting results and the performance of developed model is mea-
sured by comparing the forecasting results with ground-truth of testing To achieve the desired objectives of this study, e.g., minimum en-
data. Moreover, the mean absolute percentage error (MAPE) and nor- ergy cost and PAR with optimal EV charging/discharging, the objective
malized root mean square error (NRMSE) are considered as a validation function is expressed as:
error (to show the relationship between ANN outputs and target values) T
in our work. The Eq. of MAPE [31] is given below: (Γtsh + Γtns − (E t + ESSt + EV t + MGTt )) × EP,
minimize ∑ t
m t=1 (3)
1 |pactual (i,j) − p forecast (i, j )|
MAPE(i) =
n
∑ pactual (i, j )
,
Subject to:
j=1 (2)
E (t ) = P pv (t ) + P wt (t ), (4)
where pactual and p forecast denote the actual and predicted values of wind
speed and solar radiation, respectively. The number of days under Γns (t ) + Γsh (t ) = (E (t ) + EV(t ) + ESS(t ) + MGT(t ) + Φ(t )), (5)
consideration are demonstrated by m . The values of weights between
n
three layers of ANN are updated (until error is minimized) through
Levenburg Marquart algorithm. So, the forecasted values (for wind ∑ η = LOT(a),
a=1 (6)
speed and radiation) by ANN have the minimum error. The work pre-
sented in [37] describes more detail about ANN and learning techni- n

ques. ∑ α ≤ η ≤ β,
a=1 (7)

4. Mathematical formulations Φt ≤ Kl, (8)

The objective function based on minimum energy cost and PAR V cut − in ≤ V (t )wt ≤ V cut − out, ∀ t ε T, (9)

4
S. Aslam, et al. Electric Power Systems Research 182 (2020) 106232

Fig. 2. Structure of wind speed prediction model [41,42].

Fig. 3. Electricity generation and cut-in, cut-out wind speed.

0 < Irr(t ) < Kc, ∀ t ε T, (10) 4.2. Microgrid

0 < ESSmin < ESSmax , ∀ t ε T, (11) The total electricity generated by all RESs m ε M in a time slot t ε T
is expressed in Eq. (13).

EV emin ≤ ρEV ea + EVe ≤ EV max

e
. (12) E (t ) = ∑ εm (t ).
mεM (13)
All the mathematical notations used in objective function and
constraints are elaborated in Sections 4.2–4.6 as well as presented in where εm expresses the energy generation from single RES. The total
Table 1. electricity generation from microgrid for whole day is calculated by Eq.

5
S. Aslam, et al. Electric Power Systems Research 182 (2020) 106232

(14). appliances Appns , taken from [24]. First class of appliances considered
T soft and We can schedule at any time to any slot. We can calculate the
E (T ) = ∑ ∑ εm (t ). energy consumption of shiftable appliances by Eq. (20) and (21), re-
t=1 mεM (14) spectively.
Furthermore, the RESs are intermittent sources of electricity genera- Γsh (t ) = ∑ (λ sh ·α sh (t )),
tion; therefore, an ANN is implemented in this work for future predic- Appsh (20)
tion. Our proposed system model is able to predict wind speed and solar
radiation for the next T slots (day) on the basis of historical data. The T

detail explanation of RESs is presented below. Γsh (T ) = ∑ ∑ (λ sh ·α sh (t )),

t=1 Appsh (21)
4.2.1. WT where the term T is considered 24 h and λ sh denotes the power rating of
WT produces power on the basis of wind speed and we can calculate each appliance. α sh (t ) presents the ON/OFF status of each smart ap-
energy by the following mathematical formula: pliance at time t .
P wt (t ) = 1/2. Cp. (λ ). ρ . A. (Vt wt )3. (15) 1 If ash is ON
α sh (t ) = ⎧
The P wt
shows the generated energy from WT at time t . WT generates ⎩ 0 If ash is OFF.
⎨ (22)
power on the bases of area swept by rooter blades A , wind speed Vtwt
The appliances from latter category are neither interrupted nor
and ρ that indicates air density. The energy from WT generation is
shifted during their execution. There they cannot contribute in reducing
directly proportional to the wind speed Pt wt ∝ V described in Eq. (15).
energy cost. We can calculate hourly and total cost against this category
Fig. 3 shows the relationship between wind speed and energy genera-
of appliances by Eqs. (23) and (24), respectively.
tion from WT. Furthermore, the wind speed prediction has been per-
formed using ANN model as [38], which is presented in Fig. 2. All the Γns (t ) = ∑ (λ ns ·α ns (t )),
constraints of WT are described below: Appns (23)
V cut − in ≤ V wt (t ) ≤ V cut − out, (16) T

0= V wt (t ) ≥ V cut − out, ∀ t ε T, (17)

Γns (T ) = ∑ ∑ (λ ns ·α ns (t )).
t=1 Appns (24)
0= V wt (t ) ≤ V cut − in, ∀ t ε T. (18) Where the terms α ns and λ ns denote ON or OFF status and power rating
of non-shiftable appliances, respectively. The term α ns (t ) is explained
4.2.2. Solar panel below:
The solar panel produces electricity from sunlight that is calculated 1 If ans is ON
as [39]: α ns (t ) = ⎧
⎩ 0 If ans is OFF.
⎨ (25)
P pv (t ) = ηpv × Apv × Irr(t ) × (1 − 0.005(Temp(t ) − 25)), (19)
where hourly produced energy is shown by Ptpv . The terms ηpv and Apv 4.4. Storage systems
denote the efficiency and area of the solar panel, respectively. The solar
irradiation and outside temperature are denoted by Irrt and Tempt , re- This section uncovered the energy storage systems and their for-
spectively for time interval t . Moreover, we implement the solar cell mulations, which are considered in this work.
performance model to obtain the current-voltage (I-V) curve and
maximum power point (MPP), which are presented in Fig. 4.
4.4.1. ESS
The smart household is also equipped with ESS and 3kWh storage
4.3. Home appliances
capacity is considered as [24]. ESS has various constraints in advance,
like minimum and maximum charging levels denoted by ESSmin and
The household has various smart appliances that are categorized
ESSmax , respectively. Furthermore, there is predefined limit for its dis-
into two classes: shiftable appliances Appsh and non-shiftable
charging because of safety reasons and depth of discharge (DOD) is
considered 90% in this work as [39]. We can express stored energy in
ESS as [39]:

SE(t ) = SE(t − 1) + k . ηESS . ESch (t ) − k . ESdis (t )/ ηESS , (26)

where

ES(t )ch ≤ ESmax , (27)

ESS(t )ch < ESS upl , (28)

ES(t )dis ≥ ESmin . (29)

where SE denotes the stored energy (Ah) at time t , ηESS is the ESS's
efficiency, ESch shows the charging state and ESdis denotes ESS's dis-
charging state at time interval ‘t’. Furthermore, the charging ESch or the
discharging ESdis status of ESS is presented as binary variables:

⎧1 If ESS is charging,
α e (t ) = 0 If ESS is idle,
⎨
Fig. 4. I –V curve. ⎩− 1 If ESS is discharging. (30)

6
S. Aslam, et al. Electric Power Systems Research 182 (2020) 106232

4.4.2. EV Table 2
This work also considers EV as a mobile storage to mitigate the Parameters of appliances [24,25].
energy cost. We consider both functionalities of EV, like driving and Category Appliance name γ (kW) α (h) β (h) LOT (h)
mobile storage. We can calculate the energy utilization of EVe between
arrival and departure: Cooker hub 3 6 10 1
Cooker oven 4 15 20 1
e
td Microwave 1.7 6 10 1
EVe = ∑ (ψe (t ) × α e (t )), Shiftable Laptop 0.1 18 24 2
t = tae (31) appliances Desktop 0.3 18 24 3
Vacuum cleaner 1.2 9 17 1
where tae and tde denote the arrival time and departure time of EVe . The Electric car 3.5 18 8 3
terms ψe and t denote the charging/discharging (Ah) at time interval t Dish washer 3.5 18 8 3
Non-shiftable Lighting 0.84 16 24 6
and EV's status. If the α e = 1, EV is charging, discharging is shown by
appliances Refrigerator 0.3 1 24 24
− 1 and 0 indicates the idle state of EV, which is expressed in Eq. (30).
Furthermore, when we consider EV without discharging capabilities
then we neglect the -1 status from α e . where δ (t ) and δ (T ) show the per hour and total cost for a day, re-
spectively. The terms ϕ (t ) and EP(t ) denote the imported electricity at
4.5. MGT time t and electricity price at time slot t , respectively.

MGTs are electricity generators with easy installation and usually, 5. Simulation analysis
their capacity ranges are 15–300 (kW) [40]. We consider MGT as a
shiftable energy generation system in this work. It performs an opera- We have performed extensive simulations to show the optimal
tion according to the user's requirements. Furthermore, it generates scheduling for home appliances and EV, which are performed on the
electricity with minimum cost as compared to a diesel generator be- computer system intel core i7 with RAM of 8 GB and processor 2.8
cause natural gas is cheaper as compared to diesel [22]. MGTs also play under the MATLAB environment. The performance parameters of this
a significant role in minimizing carbon emission. It is important to note study are PAR and energy cost alleviation with the optimal utilization
that when MGT operates, it generates constant electricity 2 kW other- of electricity generated from own microgrid. A system model is de-
wise 0. Because it has two binary states either ON or OFF; so, it gen- signed for smart home; comprising of 10 smart appliances, which are
erates stable electricity when it has ON state, otherwise no power would connected with an EMC via the internet (WIFI) and EMC takes a deci-
be generated [9]. sion against each appliance according to electricity tariff. Furthermore,
these 10 smart appliances are categorized into 2 different categories
4.6. Electricity cost calculation explained in detail in Section 3. All smart appliances with their para-
meters are presented in Table 2. Non-shiftable appliances are unable to
In this section, we explain per unit energy price EPt that varies in help in managing electricity consumption for cost reduction. In addi-
each time slot t . We utilize RTP tariff for energy cost estimation. We can tion, we also have developed a forecasting model that is used for solar
calculate hourly energy cost, when smart home is not connected with and wind energy prediction for more efficient energy management.
microgrid, as: Table 3 presents the simulation parameters of forecasting model. We
have considered 24 h time for simulations and we have examined our
ζ (t ) = (Γns (t ) + Γsh (t )) × EP(t ), (32)
proposed scheme in 2 case studies: (I) HEM without microgrid (II) HEM
where energy demand of shiftable and non-shiftable appliances is de- with microgrid and EV. In the Section, simulation results of the first
noted by Γtsh and Γtns , respectively. Eq. (33) calculates the total energy case are uncovered.
cost for a day.
T
5.1. HEM without RESs integration
ζ (T ) = ∑ (Γns (t ) + Γsh (t )) × EP(t ).
t=1 (33) This section uncovers the results of our proposed scheme and we did
not consider microgrid in this section.
Moreover, the imported electricity is expressed in Eq. (34) after con-
Fig. 5 shows the energy consumption pattern of smart home with
sidering RESs, ESS, MGT and EV.
and without our proposed scheme. The RTP tariff is also presented in
Φ(t ) = ((Γns (t ) + Γsh (t )) − (E (t ) + EV. α ev(t ) + ESS. α ess(t ) this figure and it is clearly noticed that our scheme schedules load at
low price hours. On the contrary, when our scheme was not im-
+ MGT. α mgt(t ) )), (34) plemented maximum energy is purchased in high price hours, which
may cause peak creation. The hourly cost of this case is presented in
Φ(t ) If Φ(t ) > 0,
Φ(t ) = ⎧
⎨ 0 otherwise.
⎩ (35) Table 3
Simulation parameters of forecasting model.
In Eq. (22), if ϕt > 0 , the energy is imported, otherwise 0. Eq. (36)
expresses the total imported energy in a day. Parameter Value

T Iterations 100
Φ(T ) = ∑ Φ(t ). Hidden layers 1
t=1 (36) ANs 24
Number of ANs in the hidden unit 5
We can calculate per hour and per day energy costs by Eq. (37) and Initial weights 0.1
(38), respectively. Learning rate from hidden unit 5
Learning rate from output unit 2
δ (t ) = (Φ(t ) × EP(t )), (37) Bias value 0
Size of generations (EDE) 100
T
Population size (EDE) 24
δ (T ) = ∑ (Φ(t ) × EP(t )). Number of decision variables 1
t=1 (38)

7
S. Aslam, et al. Electric Power Systems Research 182 (2020) 106232

Fig. 5. Energy consumption. Fig. 8. PAR.

our proposed scheme also helps in efficient and reliable grid operation
by alleviating PAR. It is noticed from Fig. 8, minimum PAR is achieved
(PAR reduction is 48%) by implementing our proposed scheme as
compared to unscheduled case.

5.2. HEM with microgrid integration

In order to validate our newly MILP based developed scheme, we

perform further simulations while considering microgrid as a part of the
system model. Microgrid contains RESs along with storage system and
EV. Electricity generation from WT and solar panel is presented in Fig. 9
and it can be observed from this figure energy generation from the solar
panel is high in day times.
On the contrary, the solar panel generates minimum power or 0
when there are low solar irradiations or at night times. However, WT
totally depends on wind speed. When wind speed is high and below cut
Fig. 6. Energy cost. out speed, it generates maximum energy and vice versa. Furthermore,
day-ahead radiation and wind speed are predicted using an ANN for
Fig. 6. We can observe from the figure, energy consumers need to pay accurate energy management. Fig. 10 shows the predicted and real
the minimum cost with our proposed scheme over the unscheduled values of wind speed. Results present that our proposed historical based
case. Our proposed MILP based scheme optimally schedules the home ANN model shows high accuracy in terms of wind speed prediction.
appliances and eventually, consumers enjoy minimum cost against Furthermore, Figs. 11 and 12 show the relationship between ANN
optimal scheduling. outputs and target values using NRMSE and MAPE. It is clearly seen
Fig. 7 presents the per day energy cost of the consumer while im- from these figures, our proposed EDE based ANN shows minimum error
plementing our proposed scheme. It is observed from the figure, a as compared to a simple ANN model using both NRMSE and MAPE.
significant amount of cost is reduced by implementing our proposed Fig. 13 presents the optimal load demand and electricity generation
approach over unscheduled energy utilization, which is 45%. Moreover, from the microgrid. The results demonstrate that electricity demand is

Fig. 7. Total energy cost for 24 h. Fig. 9. Microgrid generation.

8
S. Aslam, et al. Electric Power Systems Research 182 (2020) 106232

Fig. 10. Wind speed prediction for WT generation using ANN. Fig. 13. Electricity demand and generation.

Fig. 11. Performance validation by NRMSE.

Fig. 14. Hourly imported electricity and optimal demand.

the electricity price is high, our proposed scheme minimizes the load
and the total load is met by the microgrid. When electricity rates are
low then our proposed scheme purchases the electricity from the ex-
ternal grid. Furthermore, Fig. 14 depicts the RES contribution curve
which demonstrates the RES contribution while meeting the electricity
demand for the whole day. We minimize the imported load and even-
tually, energy cost and load burden on the external grid are reduced.
Fig. 15 shows total energy demand, purchased/imported electricity
from the grid without and with considering EV discharging abilities.
This figure also shows the hourly energy generation from WT. Results
clearly express that the proposed approach significantly mitigates the
purchased energy in high peak hours. In addition, the smart home
imports minimum electricity while considering EV with discharging
features. On the contrary, the smart home needs to import more energy
when EV is not able to discharge itself. It is important to note that EV is
Fig. 12. Performance validation by MAPE. assumed for both purposes, e.g., mobile storage and driving purposes.
Fig. 16 shows the EV charging and discharging as well as total elec-
tricity generation from all sources of the microgrid. It is clearly seen
low during high price hours using the proposed technique and elec- from Fig. 16, the EV performs charging operation when electricity
tricity generation is also maximum in these hours. Our developed generation is maximum. On the contrary, EV performs discharging
scheme meets total demand in ON-peak hour from the microgrid. operation when electricity generation is minimum from the microgrid.
Nevertheless, on the whole day, our proposed scheme utilizes electricity Furthermore, in the first hour, we consider EV has full charging which
from microgrid and the remaining demand is met by the utility com- can be seen from Fig. 16.
pany. The contributions of EV and ESS are depicted in Fig. 17 that are
Fig. 14 shows the hourly imported electricity and total electricity minimum cost along with reduced PAR. Results show that the consumer
demand for a single home. The results clearly demonstrate that when utilizes electricity from the storage system or EV instead of importing

9
S. Aslam, et al. Electric Power Systems Research 182 (2020) 106232

Fig. 15. Amount of purchased electricity with and without EV discharging. Fig. 18. Optimal demand and MGT contribution in minimizing PAR.

from Fig. 18, MGT is ON in 16, 22 and 24, time slots, when load de-
mand is greater than defined threshold Kl and due to this, PAR is
minimized.

6. Conclusion

This work proposes a home energy management scheme to mitigate

the energy cost of the residential household. Imported electricity and
PAR reduction are also the key objectives of this study. We consider a
smart home with several smart appliances. The smart home is also in-
tegrated with a grid-connected microgrid that is empowered by RESs.
For efficient utilization of energy produced from microgrid, static sto-
rage system and mobile storage (EV) are considered in this work. A
historical data based EDE-ANN model is also developed for accurate
day-ahead energy prediction for proper efficient energy management.
Moreover, we have formulated and solved the scheduling problem by
MILP. Simulation results have validated that the proposed approach
Fig. 16. EV charging/discharging with electricity generation. efficiently schedules smart appliances as well as EV charging/dischar-
ging according to the microgrid electricity generation pattern.
Consequently, by implementing our proposed scheme, electricity cost is
reduced by 45% and 80% without and with microgrid integration, re-
spectively. Furthermore, our newly developed EDE based ANN energy
forecasting model outperforms over the basic ANN.

References

[1] V.C. Gungor, D. Sahin, T. Kocak, S. Ergut, C. Buccella, C. Cecati, G.P. Hancke, Smart
grid and smart homes: Key players and pilot projects, IEEE Ind. Electron. Mag. 6 (4)
(2012) 18–34.
[2] P. Siano, Demand response and smart grids-a survey, Renew. Sustain. Energy Rev.
30 (2014) 461–478.
[3] S. Aslam, Z. Iqbal, N. Javaid, Z.A. Khan, K. Aurangzeb, S.I. Haider, Towards effi-
cient energy management of smart buildings exploiting heuristic optimization with
real time and critical peak pricing schemes, Energies 10 (12) (2017) 2065.
[4] B.-G. Kim, Y. Zhang, M. Van Der Schaar, J.-W. Lee, Dynamic pricing and energy
consumption scheduling with reinforcement learning, IEEE Trans. Smart Grid 7 (5)
(2016) 2187–2198.
[5] E. Shirazi, S. Jadid, Cost reduction and peak shaving through domestic load shifting
and ders, Energy 124 (2017) 146–159.
[6] J. Silvente, L.G. Papageorgiou, An milp formulation for the optimal management of
Fig. 17. Hourly EV and ESS contribution. microgrids with task interruptions, Appl. Energy 206 (2017) 1131–1146.
[7] M.H.K. Tushar, C. Assi, M. Maier, M.F. Uddin, Smart microgrids: Optimal joint
scheduling for electric vehicles and home appliances, IEEE Trans. Smart Grid 5 (1)
from the commercial grid when power demand is high or generation (2014) 239–250.
[8] S. Umetani, Y. Fukushima, H. Morita, A linear programming based heuristic algo-
from RESs is minimum. In conclusion, electricity costs and PAR are rithm for charge and discharge scheduling of electric vehicles in a building energy
reduced. Furthermore, MGT is considered for peak load reduction. management system, Omega 67 (2017) 115–122.
However, when it is ON, the electricity generation is 2kW otherwise 0. [9] X. Wang, A. Palazoglu, N.H. El-Farra, Operational optimization and demand re-
sponse of hybrid renewable energy systems, Appl. Energy 143 (2015) 324–335.
MGT performs its operation only when the demand is greater than the [10] S. Rahim, N. Javaid, A. Ahmad, S.A. Khan, Z.A. Khan, N. Alrajeh, U. Qasim,
defined threshold Kl of importing load in any time slot. It is clearly seen Exploiting heuristic algorithms to efficiently utilize energy management controllers
with renewable energy sources, Energy Build 129 (2016) 452–470.

10
S. Aslam, et al. Electric Power Systems Research 182 (2020) 106232

[11] S. Pal, S. Thakur, R. Kumar, B. Panigrahi, A strategical game theoretic based de- tasks, Energy Convers. Manag. 110 (2016) 113–124.
mand response model for residential consumers in a fair environment, Int. J. Electr. [26] A.P. Engelbrecht, Computational Intelligence: An Introduction, John Wiley & Sons,
Power Energy Syst. 97 (2018) 201–210. 2007.
[12] Y. Huang, H. Tian, L. Wang, Demand response for home energy management [27] N. Amjady, F. Keynia, H. Zareipour, Short-term load forecast of microgrids by a new
system, Int. J. Electr. Power Energy Syst. 73 (2015) 448–455. bilevel prediction strategy, IEEE Trans. Smart Grid 1 (3) (2010) 286–294.
[13] P.M. Castro, I. Harjunkoski, I.E. Grossmann, New continuous-time scheduling for- [28] Y.-L. Hu, L. Chen, A nonlinear hybrid wind speed forecasting model using lstm
mulation for continuous plants under variable electricity cost, Ind. Eng. Chem. Res. network, hysteretic elm and differential evolution algorithm, Energy Convers.
48 (14) (2009) 6701–6714. Manag. 173 (2018) 123–142.
[14] A.C. Luna, N.L. Diaz, M. Graells, J.C. Vasquez, J.M. Guerrero, Mixed-integer-linear- [29] K.-P. Lin, P.-F. Pai, Y.-J. Ting, Deep belief networks with genetic algorithms in
programming-based energy management system for hybrid pv-wind-battery mi- forecasting wind speed, IEEE Access 7 (2019) 99244–99253.
crogrids: Modeling, design, and experimental verification, IEEE Trans. Power [30] Y. Wang, Y. Shen, S. Mao, G. Cao, R.M. Nelms, Adaptive learning hybrid model for
Electron. 32 (4) (2017) 2769–2783. solar intensity forecasting, IEEE Trans. Ind. Inform. 14 (4) (2018) 1635–1645.
[15] Z. Chen, L. Wu, Y. Fu, Real-time price-based demand response management for [31] A. Ahmad, N. Javaid, M. Guizani, N. Alrajeh, Z.A. Khan, An accurate and fast
residential appliances via stochastic optimization and robust optimization, IEEE converging short-term load forecasting model for industrial applications in a smart
Trans. Smart Grid 3 (4) (2012) 1822–1831. grid, IEEE Trans. Ind. Inform. 13 (5) (2017) 2587–2596.
[16] S. Moon, J.-W. Lee, Multi-residential demand response scheduling with multi-class [32] M. Niedźwiecki, M. Ciołek, Identification of nonstationary multivariate auto-
appliances in smart grid, IEEE Trans. Smart Grid. regressive processes-comparison of competitive and collaborative strategies for
[17] R.D. Bingham, M. Agelin-Chaab, M.A. Rosen, Whole building optimization of a joint selection of estimation bandwidth and model order, Digital Signal Process. 78
residential home with pv and battery storage in the Bahamas, Renew. Energy 132 (2018) 72–81.
(2019) 1088–1103. [33] [online]. available: https://pvwatts.nrel.gov/pvwatts.php. last accessed on:
[18] B. S. Sami, N. Sihem, Z. Bassam, Design and implementation of an intelligent home October 12, 2018.
energy management system: A realistic autonomous hybrid system using energy [34] A. Gensler, J. Henze, B. Sick, N. Raabe, Deep learning for solar power forecasting-an
storage, Int. J. Hydrogen Energy. approach using autoencoder and lstm neural networks, in: 2016 IEEE international
[19] M. Shakeri, M. Shayestegan, S.S. Reza, I. Yahya, B. Bais, M. Akhtaruzzaman, conference on systems, man, and cybernetics (SMC), IEEE (2016) 002858–002865.
K. Sopian, N. Amin, Implementation of a novel home energy management system [35] S. Pulipaka, R. Kumar, Comparison of som and conventional neural network data
(hems) architecture with solar photovoltaic system as supplementary source, division for pv reliability power prediction, in: 2017 IEEE International Conference
Renew. Energy 125 (2018) 108–120. on Environment and Electrical Engineering and 2017 IEEE Industrial and
[20] D. Thomas, O. Deblecker, C.S. Ioakimidis, Optimal operation of an energy man- Commercial Power Systems Europe (EEEIC/I&CPS Europe), IEEE (2017) 1–5.
agement system for a grid-connected smart building considering photovoltaics’ [36] A. Alzahrani, P. Shamsi, C. Dagli, M. Ferdowsi, Solar irradiance forecasting using
uncertainty and stochastic electric vehicles’ driving schedule, Appl. Energy 210 deep neural networks, Procedia Comput. Sci. 114 (2017) 304–313.
(2018) 1188–1206. [37] J. Fulcher, Computational intelligence: an introduction, in: Computational in-
[21] Y. Lu, S. Wang, Y. Sun, C. Yan, Optimal scheduling of buildings with energy gen- telligence: a compendium, Springer, 2008, pp. 3–78.
eration and thermal energy storage under dynamic electricity pricing using mixed- [38] G. Chang, H. Lu, Y. Chang, Y. Lee, An improved neural network-based approach for
integer nonlinear programming, Appl. Energy 147 (2015) 49–58. short-term wind speed and power forecast, Renew. Energy 105 (2017) 301–311.
[22] M. Nemati, M. Braun, S. Tenbohlen, Optimization of unit commitment and eco- [39] E. Shirazi, S. Jadid, Optimal residential appliance scheduling under dynamic pri-
nomic dispatch in microgrids based on genetic algorithm and mixed integer linear cing scheme via hemdas, Energy Build. 93 (2015) 40–49.
programming, Appl. Energy 210 (2018) 944–963. [40] M.A.R. do Nascimento, L. de Oliveira Rodrigues, E.C. dos Santos, E.E.B. Gomes,
[23] M. Meiqin, S. Shujuan, L. Chang, Economic analysis of the microgrid with multi- F.L.G. Dias, E.I.G. Velásques, R.A.M. Carrillo, Micro gas turbine engine: a review, in:
energy and electric vehicles, in: Power Electronics and ECCE Asia (ICPE & ECCE), Progress in Gas Turbine Performance, InTech, 2013.
2011 IEEE 8th International Conference on, IEEE (2011) 2067–2072. [41] A. Khalid, S. Aslam, K. Aurangzeb, Haider S. Irtaza, M. Ashraf, N. Javaid, An effi-
[24] S. Aslam, N. Javaid, M. Asif, U. Iqbal, Z. Iqbal, M.A. Sarwar, A mixed integer linear cient energy management approach using fog-as-a-service for sharing economy in a
programming based optimal home energy management scheme considering grid- smart grid, Energies 11 (12) (2018) 3500, , https://doi.org/10.3390/en11123500.
connected microgrids, in: 2018 14th International Wireless Communications & [42] S. Aslam, An optimal home energy management scheme considering grid connected
Mobile Computing Conference (IWCMC), IEEE (2018) 993–998. microgrids with day-ahead weather forecasting using artificial neural network,
[25] D. Zhang, S. Evangelisti, P. Lettieri, L.G. Papageorgiou, Economic and environ- Master Thesis, COMSATS University Islamabad, Islamabad, Pakistan, 2018.
mental scheduling of smart homes with microgrid: Der operation and electrical

11