This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON SMART GRID 1

A Microgrid Energy Management System Based on

the Rolling Horizon Strategy
Rodrigo Palma-Behnke, Senior Member, IEEE, Carlos Benavides, Fernando Lanas, Bernardo Severino,
Lorenzo Reyes, Jacqueline Llanos, Student Member, IEEE, and Doris Sáez, Senior Member, IEEE

Abstract—A novel energy management system (EMS) based on for electric power system management. Smart grids are under-
a rolling horizon (RH) strategy for a renewable-based microgrid stood to be the key enabling technology for renewable energy
is proposed. For each decision step, a mixed integer optimization development, electric vehicle adoption and energy efficiency
problem based on forecasting models is solved. The EMS provides improvements [2]. Smart grids represent a vision for digital op-
online set points for each generation unit and signals for consumers
based on a demand-side management (DSM) mechanism. The pro- timization of electric power distribution and transmission grids
posed EMS is implemented for a microgrid composed of photo- as applied to current operations, enhancing the grid security and
voltaic panels, two wind turbines, a diesel generator and an en- opening up new ways of tapping alternative energy sources.
ergy storage system. A coherent forecast information scheme and By using the Internet protocol on home devices, the smart grid
an economic comparison framework between the RH and the stan- would be able to shuttle information back and forth between the
dard unit commitment (UC) are proposed. Solar and wind energy distributed electric utilities and customers.
forecasting are based on phenomenological models with updated In the specialized literature, there are works related to en-
data. A neural network for two-day-ahead electric consumption
forecasting is also designed. The system is tested using real data
ergy management system (EMS) for microgrids. In [3], an on-
sets from an existent microgrid in Chile (ESUSCON). The results line power energy management for a hybrid fuel cell/battery dis-
based on different operation conditions show the economic sense tributed generation system is presented. The online architecture
of the proposal. A full practical implementation of the system for consists of three layers: the first one captures the possible op-
ESUSCON is envisioned. erations modes, the second is based on a fuzzy controller for
power splitting between batteries and fuel cells, and the last one
Index Terms—Demand-side management, energy management
system, microgrid, photovoltaic, rolling horizon, wind turbine. regulates each subsystem.
In [4], an EMS for controlling a virtual power plant (VPP)
is presented. In this case, the objective is to manage the power
flows to minimize the electricity generation costs and to avoid
I. INTRODUCTION
the loss of energy produced by renewable energy sources. The
VPP is composed of combined heat and power plants, wind

W ITH A LOW population density and many natural

resources, Chile’s electricity supply shortage is a
paradox. The Atacama Desert, Andes Mountains, long coast-
parks and a photovoltaic plant. The surplus energy generated by
renewable units is used for the operation of both a desalination
plant and an electrolyzer that provide hydrogen for industrial
line, and strong agriculture provide the country with abundant processes. Lu & Francois [5] describe an energy management
distributed renewable energy resources that could, according to system based on day-ahead power scheduling for a microgrid
recent estimations, largely surpass the fast-growing electricity (photovoltaic panels, gas turbine) that considers the power pre-
demand. In addition, because of the geographic conditions diction and load forecasting. The EMS provides the real-time
of extreme points in Chile, there are still several small set- power set points for microgrid units and coordinates the droop
tlements deep into the mountains that are isolated from the controllers for the primary frequency control.
interconnected power system. In this context, renewable-based In Teleke et al. [6], a rule-based control strategy is designed
technologies offer a solution for gathering the distributed for a battery energy storage system with photovoltaic arrays and
energy resources [1]. a wind farm. The renewable sources can be dispatched hourly,
A renewable-based microgrid can be understood as a partic- based on forecasting of the solar and wind conditions. The rule-
ular case of a more general concept called a ‘smart grid’, which based controller determines the current reference for the con-
is an interdisciplinary term for a set of technological solutions verter that will charge/discharge the battery bank by using the
state of charge (SOC) and the battery voltage. This system can
deal with variability in the wind and solar generation. Wang et
Manuscript received July 04, 2011; revised November 18, 2011, June 15, al. [7] designed a power management system for a stand-alone
2012, and August 31, 2012; accepted October 14, 2012. This work was sup- grid that is composed of a wind turbine, photovoltaic arrays and
ported in part by the mining company Doña Inés de Collahuasi, the Millen- a fuel cell. The proposed management system is tested for dif-
nium Institute “Complex Engineering Systems” (ICM: P-05-004-F, CONICYT:
FBO16), and FONDECYT project 1110047. Paper no. TSG-00233-2011.
ferent scenarios by using a practical load demand profile and
The authors are with the Energy Center, Faculty of Mathematical and Phys- real weather data. The strategy coordinates the power flows
ical Sciences, School of Engineering, University of Chile (CMM, ISCI, DIE), among the different energy sources.
Av. Tupper 2007, 8370451, Santiago Chile (e-mail: [email protected]; Gupta et al. [8] developed steady-state models for a hybrid
[email protected]).
Color versions of one or more of the figures in this paper are available online energy system for determining its optimal operation. The hy-
at http://ieeexplore.ieee.org. brid system is composed of microhydro, biogas, biomass, pho-
Digital Object Identifier 10.1109/TSG.2012.2231440 tovoltaic panels, a battery bank and a fossil fuel generator. An

1949-3053/$31.00 © 2012 IEEE

 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON SMART GRID

economic dispatch strategy with mixed integer linear program-

ming is implemented. Westermann and John [9] describe a com-
bination of wide-area measurement and ripple control for DSM.
The proposed control systems moderate the impact of increased
renewable sources on adjacent transmission grids. Hamidi and
Robinson [10] propose a responsive demand for a system with
significant intermittent generation, e.g., a microgrid with renew-
able sources. In this system, the demand can shift to reduce the
peak and also be reduced to mitigate the power fluctuations;
however, this action requires evaluating the value of the lost
load for consumers. Fig. 1. Renewable-based microgrid: Power schematic.
Lagorse and Somoes [11] propose a distributed EMS to con-
trol the energy flow in the hybrid energy system (HES). This dis- solution is also able to be interconnected to a power grid as a
tributed controller is based on multiagent-system (MAS) tech- microgrid solution, allowing the integration of the local energy
nology. The hybrid system is composed of photovoltaic cells, resources. In this work, based on a specific real location in the
batteries, fuel cells, supercapacitor and load. The battery unit north of Chile, a novel EMS based on a rolling horizon strategy
is guided by fuzzy-logic rules. Logenthiran and Srinivasan [12] for a renewable-based microgrid is proposed. For each decision
describe a three-step efficient method for the optimal generation step, a mixed integer optimization problem based on forecasting
scheduling of a microgrid in island operation. This microgrid is models is solved. The EMS considers a two-day-ahead forecast
composed of a photovoltaic system, a wind park, thermal units of renewable sources and a two-day-ahead prediction for elec-
and a battery bank. The first step is to set up an initial feasible tric consumption based on a neural network model. In addition,
solution for thermal unit commitment, considering the use of EMS integrates a DSM scheme that provides signals for con-
renewable energy to meet the load. From these solutions, the sumers [1]. Nevertheless, in this proposal [1], the uncertainties
start-up and shut-down of thermal units is determined, again are not considered explicitly.
solving a thermal unit commitment problem. The final step is to The EMS includes an objective function and the nonlinear
optimize a nonlinear renewable-thermal dispatch, considering constraints associated with the modeling of both units. These
the nonlinear constraints associated with the modeling of renew- constraints are represented by piecewise linear models, binary
able energy sources and assuming the start-ups and shut-downs variables and so forth. Therefore, the optimization problem is
of thermal units are known. The nonlinear optimization problem solved using mixed integer linear programming, ensuring the
is solved using genetic algorithms. near-optimality. This advantage is not present in the previous
Tsikalakis and Hatziargyriou [13] describe a centralized con- works described that used Lagrange relaxation, genetic algo-
trol system for a microgrid. The controller is used to optimize rithms and others.
the operation of the microgrid during interconnected operation, The rest of the paper is organized as follows: In Section II,
i.e., the production of local generators and energy exchanges a description of the renewable-based microgrid and its current
with the distribution network are maximized. Two market poli- power system is given. In Section III, a description of com-
cies are assumed to offer options for the demand for controllable munication infrastructure of the renewable-based microgrid is
loads, and this demand-side bidding is incorporated into the cen- presented. Section IV details the proposed energy management
tralized control system. system, including the modeling of each unit and the DSM. In
With regard to EMS with rolling horizon used in the field Section V, simulation results are presented and analyzed under
of smart grids, Dagdougui et al. [14] describe a real time op- different scenarios. Section VI concludes the paper.
erational management system for a hybrid renewable system,
composed of an electrolyzer, a hydroelectric plant, pumping sta- II. DESCRIPTION OF THE RENEWABLE-BASED MICROGRID
tions, a wind turbine and a fuel cell. The goal is to satisfy the The microgrid had been conceived for a small, isolated
variable electric, hydrogen, and water demands hourly. Morais village in the Atacama Desert, Chile, called Huatacondo
et al. [15] present an optimal dispatch of micro-grid renewable . Its electric network was iso-
energy park. The microgrid is composed of photovoltaic panels, lated from the interconnected system and was supplied only
wind turbines, a fuel cell, a load, and a storage battery. The dis- 10 hours a day by a diesel generator. This microgrid takes
patch is based on the minimization of the generation costs and advantage of the distributed renewable resources in the area,
the optimization of the storage charging and discharging cycles. providing 24-hour electricity service. Because the village expe-
Korpas and Holen [16] present an EMS system for a hybrid plant rienced problems with the water supply system, a management
with wind power and hydrogen storage. The system maximizes solution was also included in the microgrid. Additionally, a
the expected profit from power trading in a day-ahead market demand-side option to compensate the generation fluctuations
based on forecast information for generation, load, and elec- due to the renewable sources was considered. The system is
tricity price. In the studies cited, basic forecast data informa- composed of two photovoltaic systems, a wind turbine, the
tion for the electric demand and renewable energy availability existing diesel generator unit of the village (typical in isolated
is used. Additionally, the economic benefits of the control strate- locations), an energy storage system (ESS) composed of a
gies are not explicitly evaluated. lead-acid battery bank (LABB) connected to the grid through a
The aforementioned characteristics and research proposals bidirectional inverter, a water pump and a DSM (loads). Fig. 1
are applicable to a smart microgrid concept. An isolated grid summarizes the power schematic of this microgrid.
with these new features is able to offer new development oppor- The following are the main goals of the EMS:
tunities and perspectives for the local community. The resulting • Minimize the use of diesel
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PALMA-BEHNKE et al.: A MICROGRID ENERGY MANAGEMENT SYSTEM BASED ON THE ROLLING HORIZON STRATEGY 3

Fig. 3. Blocks diagram of the proposed energy management system.

Fig. 2. Renewable-based microgrid: SCADA hardware.
• Some electrical measurements in the grid and control capa-
• Deliver active generation set points for the diesel gener- bilities (connection/disconnection of low voltage network
ator, the ESS inverter, and the PV plant feeders).
• Turn on and off the water pump in order to keep the ele- • Energy consumption measurement of the electrical loads
vated water tank level within predefined limits in the network.
• Send signals to consumers promoting behavior changes • Power control of the ESS inverter and diesel generator.
In field implementation, the diesel generator and the ESS • Grid connection control for all generation units and the
inverter have two configurable droop curves to follow the set water pump consumption.
points. Q-V and P-f droop curves are normally configured for • Sun tracking control for the principal PV plant.
typical operation states that change when receiving adequate • Wireless communication with the interfaces of the DSM
signals from the EMS. system.
When the EMS turns on the diesel generator, it is convenient
to start the LABB charge. In that case, the diesel generator is IV. PROPOSED ENERGY MANAGEMENT SYSTEM
configured to work in isochronous mode (with infinite slope
in both curves, takes all the variations between generation and A. Problem Statement
load, following the rated values ), and the The EMS minimizes the operational costs while supplying
ESS inverter curves are configured to follow the charging pro- the water and load demands, considering a two-day-ahead
file of the batteries. forecasting of the weather conditions, water consumption and
When the diesel generator is turned off, the droop curves of electrical load. Fig. 3 shows the proposed EMS that provides
the ESS inverter change to those for a master operation mode, the power references for the diesel generator, the ESS
with a very low droop characteristic. inverter power , the binary signals for the water supply
system , the desired solar power and the signals
III. COMMUNICATION INFRASTRUCTURE for loads . The EMS inputs are the predicted maximum
In regard to the communications infrastructure of the pro- and minimum attainable solar power , wind
posed renewable-based microgrid, to acquire the field data for power , expected load , water consumption ,
optimizing the operation of the system, a SCADA and several initial conditions for the battery charge , battery bank
measurement devices were installed in the Huatacondo Village voltage and current , water tank level and diesel
(see Fig. 2). on/off state .
The main server (OPC SERVER) runs a Matrikon OPC Data For the design of the EMS, the objective function, the power
Access Service (OPC DA) using an Ethernet MODBUS TCP balance and the dynamic model of the generation units with their
protocol and a Matrikon Historical Data Access Service (OPC corresponding mixed integer programming (MIP) formulations
HDA) to save historical data [17]. are presented. The EMS is solved at the supervisory level using a
All field devices such as controllers, sensors, and actuators unit commitment (UC) with a rolling horizon strategy (UC-RH).
(collecting electrical measurements and status information) The rolling horizon is considered for reducing the effect of the
communicate through the MODBUS RTU protocol, by using uncertainties of the input variable forecasting. Thus, the UC-RH
Ethernet/Serial converters for the needed translation. control strategy at the supervisory level includes the following
For the data transfer, NET applications were written using steps:
existing GPL libraries that also provide communications with 1) Get the initial condition for .
Web Services allowing private web clients to retrieve the real- 2) The EMS optimizes the system operation for the periods
time and historical data via the Internet. from to ( : prediction horizon) using
To collect the information the OPC DA must be configured forecasting for the input variables and the initial conditions
with several tags. Therefore, timestamps can also be retrieved as of the units. In this case, the prediction horizon is two days,
a basis for the forecasting algorithms. The server is configured 192 steps with a sampling time of 15 minutes.
to read variables every second and to store them in the database 3) Obtain the optimal set points of units for .
every 10 seconds. 4) Apply the set points obtained in step 2 for a shorter period
In summary, for the communication between devices, the mi- than , in this case, to .
crogrid uses a SCADA with the following capabilities: 5) Adjust the set points for dispatchable units based on current
• Electrical variable measurements for all generation units, (measured) load and non-controllable renewable injections
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON SMART GRID

in the system. For the ESS the opportunity cost of the stored 2) Wind Generator: Generally, wind power may be obtained
energy is considered. An economic load dispatch (ELD) is using the wind turbine profile given by the manufacturer. In
used for this purpose. The update period of the ELD can this case, a piecewise linear approximation of the profile was
be flexible, selected between 1 and 5 minutes. incorporated into the wind generator model.
6) Update . When this is not possible, the wind power maybe calculated
7) Return to step 1. based on the wind speed and the wind turbine power coefficient
obtained from the basic model expression [18].
B. Objective Function
The objective function of EMS is to minimize the operational E. Forecasting Models
costs of the microgrid in a time scope . The objective func- 1) Solar Irradiance: For horizontal solar irradiance fore-
tion is formulated as casting based on the persistency methodology, the clear-sky
index is used [19], which is an approximate measurement
for cloudy conditions right above the sensor location:

(5)
(1)
is the measured irradiance and is the irradiance
where is the duration of time period , is the cost func- calculated with a phenomenological clear-sky model. With this
tion of the diesel generator, is the start-up cost function model, measured errors increase during the sunset and at dawn.
of diesel generator, is the unserved power in the system, Hence, (5) is used in cases in which the solar altitude is greater
is the price for unserved energy, is the unserved than 20 , as recommended in [19].
water, is the cost of unserved water and is the Then, for the predicted irradiance , is assumed to be an
cost of using the LABB. The first term of the objective function average of the past indices, as shown in (6).
(1) represents the diesel costs, the second term the start-up diesel
cost and the next two terms penalize the unsupplied electric en-
ergy and the unserved water supply, respectively. The last term (6)
represents the cost penalty of using the LABB and affecting its
lifetime in different ways. where n is the number of model autoregressors. The proposed
approach currently has detractors because of the RMSE values
C. Power Balance for the Microgrid obtained in field measurements [19]. Nevertheless, given the
The power balance in the microgrid must be satisfied: special conditions in the north eastern part of the country (At-
acama Desert), the proposed model is suitable for a two-day-
ahead prediction.
(2) 2) Wind Speed: For forecasting the wind speed, this work
considers a global forecast system (GFS) for the bounded con-
where is the electrical load, is the rated pump power, ditions of a weather research and forecast (WRF) model [20],
is the unused power from the energy sources (storage [21].
was not possible). Additional constraints for the problem are The GFS model is used for determining the border conditions
and . of the wind speed that the WRF model used as an initial condi-
tion. Finally, the results of the WRF model are adjusted using
D. Modeling of Renewable Energy Units statistical methods for a 12- or a 48-hour-ahead prediction.
3) Electrical and Water Consumption: For load forecasting,
1) Photovoltaic Panels: In the model proposed, the solar
neural networks have been widely used. Recently, wavelets
power is controlled by the east-to-west inclination angle of the
have been used for characterizing the load, and then, they are
panels . The maximum power is obtained by the
used as inputs for the neural network models [22]. In addition,
optimal orientation of the PV panels. Sometimes, it is not fea-
evolutionary algorithms have been used for determining the
sible to use the maximum solar power because, for example,
inputs for a load model based on fuzzy logic [23]. In this work,
the ESS is completely charged. To prevent this problem, the
the electrical consumption is predicted based on neural network
minimum power is obtained when the photovoltaic
modeling using empirical data. The multilayer architecture
panels are oriented to the minimum irradiance subject to the
is considered with one hidden layer. Backpropagation for
panel structure physical constraints. Therefore, the solar power
determining the weights and a structural optimization process
will be a value between these bounds, which are as fol-
for determining the relevant inputs are used [24]. The same
lows:
procedure for the water consumption can be used.
(3)
F. Modeling of Conventional Generation Units
(4)
1) Diesel Generator: The fuel consumption of a diesel gen-
where is the solar panel efficiency, is the total plant sur- erator can be represented as a non convex function. For the
face and is the irradiance perpendicular to the panel’s MIP formulation, a non convex function can be approximated
plane. by piecewise linear segments. Fig. 4 shows an example for
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PALMA-BEHNKE et al.: A MICROGRID ENERGY MANAGEMENT SYSTEM BASED ON THE ROLLING HORIZON STRATEGY 5

where represents a fixed start-up cost and is the initial

condition ( if the unit was previously turned-on and
if the unit was previously turned-off). This initial con-
dition is obtained from the SCADA.
For example, in (15) if and
, the diesel generator starts up in , and therefore:

(17)

Because the startup costs are represented by inequality con-

straints (14), (15), (16), their values are obtained from the EMS
operation that provides the values of the variables, which
Fig. 4. Nonconvex cost function of the diesel generator, . represent the on-off state of the diesel generator.
The volume of the diesel tank variable is given by:
. Each piecewise linear segment introduces the binary
variables and continuous variable , analytically: (18)

(19)
(7)
Additionally, a minimum volume for the diesel tank at the end
of the prediction horizon is included as:
where is the slope and is the Y-intercept of segment .
The following constraint is added for imposing that only one (20)
variable takes the value 1 in period t:
2) Energy Storage System: A simplified dynamic model of
the ESS:
(8)
(21)
Then, when means the diesel unit is where is the DC power of the LABB, assuming a positive
turned-on, otherwise the unit is turned-off. value if it injects power to the microgrid and a negative value
The generated diesel power is given by: for the charging mode.
The inverter model is:
(9)
(22)
(10)
where is the internal consumption of the inverter. The con-
As shown in Fig. 4, if the unit is turned-on the dispatched stants and are used to represent the losses of the inverter.
power at instant takes a value between and The corresponding MIP formulation for the inverter model de-
, and in the (8), the corresponding binary variables are: scribed in (21) is:

(12) (23)

Contrarily, when the diesel unit is turned-off the corre- (24)

sponding binary variables are: (25)

(13) The power is used when the LABB is discharging, and

is used for the charging model. represents the
Then, the cost function is represented by , maximum charging power and is a large negative parameter
where represents the price of diesel. representing the maximum load power. and are
The positive variable represents the startup cost of the binary variables associated with each operation mode. Only one
diesel generator: operation mode is allowed for a specific time period:

(26)
(14)
The MIP formulation of the LABB energy model described
in (20) is represented by:
(15)
(27)
(16) (28)
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON SMART GRID

the internal components of the battery depends on chemical ef-

fects such as corrosion; sulfating or water losses. Therefore SoH
is mainly determined by both its temperature and its working
zone [26].

(36)

where is the coefficient associated with the battery

temperature and is the working zone co-
Fig. 5. Minimum admissible power in the LABB vs. the battery bank SOC. efficient.

The constants and are used to represent the charge and

discharge energy efficiency of the LABB. is a value be- (37)
tween 0 and 1, and it should be chosen depending on the desired
depth of discharge of the LABB system. is the maximum corresponds to a nonlinear function and to the reference
storage value of the energy bank. is a reference energy value temperature of the model (usually 10 ). (environmental
of the LABB. temperature) can be estimated using the models presented in
The constraints for the allowed battery power are: Section IV-E2,

(29) (38)
The minimum admissible power of the LABB (negative value
in the model) is related to the maximum charge current that where are experimental parameters. The
depends on the end of the charge voltage trigger [25]. This saturation, overload, and load zones are defined based on the
charging constraint is shown in Fig. 5. The maximum LABB voltage; and the discharge, over-discharge and depletion zones
power (positive value in the model) is determined by are based on the energy of the ESS. It is important to note that
the current limit of the inverter. is ten times greater than and is nearly two times greater
For the MIP formulation of (29), a non convex function to than
represent the minimum admissible power of the LABB
can be approximated by piecewise linear segments as (see G. Water Supply Management
Fig. 5):
For the management of the water supply system, the tank
system volume is modeled as:
(30)
(39)
Each piecewise linear segment introduces the binary vari- where represents the water consumption. The water in-
ables and the continuous variable . is the slope flow into the tank is given by:
of segment ( ).
(40)
(31)
(32) where is a constant coefficient, is the pump efficiency,
and is the pump power.
where , and and are Adding a fictitious volume to the tank model, the
the upper and lower bounds of segment . unserved water consumption is incorporated into the balance.
Thus, the maximum charging power is expressed by: Therefore, the volume of the tank is modeled by:

(33) (41)
(42)
The energy balance is represented by:
where is a reference value of the tank level.
For the optimizer, the water consumption can be predicted
(34) using historical data of the village.

The cost penalty of the LABB is given by the decline in the H. Demand-Side Management
state of health (SoH) of the ESS at the end of the evaluation From [1], [27], it was supposed that online signals are sent to
period. the consumers in order to modify their consumption behaviors,
leaving daily energy constant. This modification is modeled by
(35) shifting coefficients that are provided by the EMS. There-
fore, the electric load is given by:
is a factor associated with the investment cost of the LABB.
The SoH is directly related to the ESS lifetime. The lifetime of (43)
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PALMA-BEHNKE et al.: A MICROGRID ENERGY MANAGEMENT SYSTEM BASED ON THE ROLLING HORIZON STRATEGY 7

where is the expected load. To take into account the max- TABLE I
imum expected consumer response to the signals (empirical PREDICTION ERRORS
value), the shifting factor is bounded as:

(44)

Additionally, it is assumed that the expected energy consump-

tion could be bigger than the expected demand for the whole
optimization period. This constraint is given by:

(45)

In addition to using to in (45), different sub periods for

the energy balance can be defined for the optimization horizon
conditions and the system electrical demand of the second day.
(for example, energy balance for each day).
The MIP optimization problem was solved using the commer-
cial package CPLEX version 12.1 [28]
V. CASE STUDY As was mentioned in Section IV-A, a UC-RH is proposed
The following results were obtained for a forecast horizon for reducing the effect of the uncertainties of the input vari-
of 48 hours and a sampling time of 15 minutes. Note that the able forecasting. To validate the proposed EMS-based UC-RH,
use of a 15 minute sampling time is reasonable for the proposed a one-day-ahead UC was also implemented.
rolling horizon strategy as it is focused on energy management. For evaluating both solutions (UC and UC-RH), the follow-
As mentioned in Section IV-A, inside the 15 minute window, an ings steps were achieved:
ELD is used for the set points adjustments of the dispatchable 1) The EMS optimizes the system operation using the fore-
generation units and the ESS. Additionally, the primary control cast data. For practical reasons (simplicity and accuracy),
of frequency and voltage are based on local control strategies the optimization approach presented in Section IV-A was
implemented in each unit to face the faster variations of the re- applied without the consideration of and the related
newable resources. Moreover, for the case study; Huatacondo constraints.
village, at the north of Chile, there are important changes on the 2) The system operation is simulated using the set points ob-
wind speed however the solar resource is more important, in tained in step 1 and real data profiles (validation set). In
terms of energy production, and has a lower variability inside this case, the output power of the units is adjusted to those
the 15 minutes period. defined by each data profile. For this task, a simplified ver-
The stochastic inputs for the optimizer are predicted for 192 sion of an ELD model is used due the fact that only one
steps ahead. For the evaluation purpose, the refresh rate of the unit (ESS or Diesel generator) is dispatchable in one spe-
sliding window was 1 hour, or 4 periods. For the first case cific period.
(PV/Diesel/ESS), the wind energy was not considered because 3) The costs are evaluated from the system operation simu-
the current system implemented in Huatacondo is operating lated with the real data profiles.
with PV panels, a diesel generator and an ESS. In the second 4) The cost is estimated for the different scenarios and
case, we analyze the advantages of incorporating the wind added to the simulation costs for a complete analysis.
energy system into the microgrid (PV/Diesel/ESS/Wind). For To evaluate the EMS performance, the following expected
this study, we considered two 2.5 kW wind turbines. costs must be considered: the cost of operating the diesel gener-
The irradiance for Huatacondo was obtained using the ator (startup cost and operation cost), the unserved energy cost,
models described in Section IV-E, with the photovoltaic plant and the opportunity cost of the energy stored in the LABB (en-
model considered as . The wind speed was built using the ergy deficit cost). Note that the opportunity cost, expressed as an
measured data from Huatacondo and the models described in energy deficit, is related to the difference between the reference
Section IV-E. Load consumption was forecast based on neural energy value and the energy value at the end of the period
networks by considering the historical patterns of the location . This value will be associated with the expected savings
(see Section IV-E). All forecasts are updated every 15 minutes of using the diesel generator for providing the same amount of
using the current data on the forecasting models as the EMS energy.
requires. A typical water consumption profile was considered, Consequently, the resulting EMS performance index is de-
because there are not enough available measurement data for fined as follows:
the training of a forecasting model.
Table I presents the prediction errors at 1 hour, 24 hours,
and 48 hours ahead for the wind power, solar power and elec- (46)
tric load, considering the following performance indexes: mean
absolute percentage error (MAPE) and the root mean square
error (RMSE). We observe that the prediction errors increase
A. EMS Results for a Microgrid (PV/Diesel/ESS)
for longer horizon prediction.
A 48-hour forecast horizon was used because the battery state In this case, the load-shifting capability is not considered
of charge at the end of the first day depends on both the weather . Figs. 6, 7 and 8 show the average value
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON SMART GRID

TABLE II
SUMMER SEASON (PV/DIESEL/ESS)

TABLE III
WINTER SEASON (PV/DIESEL/ESS)

Fig. 6. Average load for the summer and winter seasons.

Tables II and III show that the UC-RH reduces the expected
total cost (18% for summer and 27% for winter) because of the
lower start-up and operation diesel costs.
For the lower solar radiation in winter (see Fig. 8), the ESS
operates close to its minimum energy capacity. Thus, if there is
a forecasting error, it would be possible to reach values below
the minimum. Consequently, the diesel generator starts up off
schedule. Because the input forecasts considered by the UC-RH
are updated at each step, this problem is reduced, and a lower
start-up diesel cost is observed in comparison to the UC. For the
summer season, the energy value is similar for both methods. In
contrast, for the winter season, the UC generates positive energy
Fig. 7. Average water consumption for the summer and winter seasons. values, meaning that the stored level of ESS energy is greater
than the reference value . When using the UC-RH, a greater
value than the energy level is achieved, and therefore, a sig-
nificant energy saving is obtained. This phenomenon happens
because the UC-RH looks further into the future than the UC,
saving more energy for the limited energy of future winter con-
ditions.

B. EMS Results for a Microgrid (PV/Diesel/ESS/Wind)

The energy supply system previously described can be im-
proved by considering the advantage of wind energy integra-
tion. Using measured data from Huatacondo and the estimated
error obtained from models described in Section V-E, 500 wind
power profiles were built. Fig. 9 shows the average value of the
500 samples for wind power in summer and winter. In addi-
tion, RMSE resulting from the comparison between hourly wind
forecasts and hourly measurements increase with the forecast
range as shown in Table I.
Fig. 8. Average solar power for the summer and winter seasons. Tables IV and V show a lower expected total cost (EMS
performance index) than the PV/Diesel/ESS case (Section V-A)
of the 500 samples for load, water consumption, and solar gener- because of the wind energy contribution. Again, a total cost
ation for both the summer and winter seasons. It can be observed (EMS performance index) reduction is obtained using the
a similar profile for the village load. Note that the same water UC-RH compared with the UC approach, which was observed
consumption profile for both seasons is considered. In Fig. 8, for both seasons. However, the achieved savings (5% and 6%)
different patterns for the irradiance according to the season are are smaller than those in the previous case. The following
used. additional interesting results should be noted:
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PALMA-BEHNKE et al.: A MICROGRID ENERGY MANAGEMENT SYSTEM BASED ON THE ROLLING HORIZON STRATEGY 9

TABLE VI
LABB INVESTMENT COSTS

Fig. 9. Average wind power for summer and winter seasons.

TABLE VII
DIFFERENCE IN INVESTMENT AND OPERATIONAL COSTS FOR THE ALTERNATIVE
TABLE IV
CONFIGURATIONS
SUMMER SEASON (PV/DIESEL/ESS/WIND)

TABLE V
WINTER SEASON (PV/DIESEL/ESS/WIND)

C. Overall Analysis
For evaluating the overall performance of the two microgrid
configurations, the investment and operational costs of the main
units are analyzed over a period of 20 years (This period corre-
sponds to the wind turbine lifetime). Specifically, the investment
is calculated for the LABB and the wind turbine units.
• The expected operational diesel costs are higher in the The operational costs presented here are obtained from the
case of UC-RH. This can be explained by seeing the cor- results shown in Sections V-A and V-B. We assume a constant
responding energy deficit costs that also impose “future demand pattern over the evaluation period.
operational costs” for the system. In fact, taking into ac- The LABB costs are detailed in Table VI. It includes the ini-
count the sum of these two costs, it can be concluded that tial investment costs of the battery bank (first row) and the re-
the UC-RH approach shows lower operational costs (op- placement costs based on the life cycles over 20 years obtained
erational diesel plus energy deficit costs) for the period. from the EMS operation results (ex-post evaluation).
For example, from Table IV, these operational costs are Finally, the total cost of the LABB is summarized for the two
for the UC ap- optimization approaches (UC and UC-RH). 850 rated life cycles
proach and using at 80% discharge depth are considered in the evaluation based
the UC-RH. on the manufacturer data. The net present value is calculated
• During the summer, there is unserved energy, which was with a discount rate of 10%, which is a common value used in
not present in the previous case. The reason for this could the Chilean energy sector.
be the uncertainties associated to the forecasted wind The difference in investment and operational costs for the
speed. two microgrid configurations are summarized in Table VII. This
• The UC-RH reduces the energy deficit. This phenomenon table shows a lower total cost for the PV/Diesel/ESS/WIND so-
grows in the winter because there is less available energy lution for both the UC and UC-RH approaches. Additionally,
(Figs. 8 and 9) and the diesel generator is used intensively. for both configurations, a cost reduction using UC-RH over UC
is observed.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON SMART GRID

TABLE VIII TABLE IX

SUMMER SEASON (PV/DIESEL/ESS) EMS/DSM/UC-RH PARAMETERS

From the overall results, we observe that the total costs

(investment and operational) of PV/Diesel/ESS/Wind are up
to 15% lower than the PV/Diesel/ESS for the case of rolling
horizon (RH). Actually, for this case, the additional investment
costs of the wind turbine are less than the reduction in the
operational costs.

D. EMS Results With a DSM

The proposed EMS allows efficient management of the water
To show the advantages of the DSM, Table VIII presents the supply by optimizing the water pump activation as a flexible
results of the proposed EMS using the UC-RH for the microgrid load, especially for periods with energy surplus. The EMS for
(PV/Diesel/ESS). In this table, different shifting factor bounds a microgrid displays the attributes of a smart grid by providing
of the electric demand are applied using a typ- optimized set points for electrical units. Moreover, it also con-
ical demand and an irradiance profile for the summer season. siders other services by integrating water and electric load de-
Additionally, it is important to note that the performance evalu- mands.
ation is based on the optimizer results. In the next step of this work, the proposed EMS will be imple-
The total cost decreases as the demand shifting coefficients mented at the specific location in the Atacama Desert. The effect
are made more flexible. This result is expected, given that EMS of prediction errors and a sliding window optimization will be
was designed for generating signals to consumers. Thus, the included. Field measurements, the behavior of the communica-
system load is shifted to periods when the solar energy is avail- tion platform, the dynamic response of the system and adjust-
able. In addition, the diesel operation is optimized, increasing ments of the proposed EMS will be reported. From the results,
its efficiency. The former improves fuel management usage. the integration of computational intelligence techniques in fore-
casting and control strategies is envisioned. Future research will
VI. CONCLUSIONS also focus on the incorporation of the proposed LABB penalty
This paper presents a novel energy management system for a cost model in the optimization procedure.
renewable-based microgrid. The EMS provides the online set
points for generation units while minimizing the operational APPENDIX
cost and considering the forecast of renewable resources, load Further detailed technical characteristics of the microgrid
and water consumption. units are summarized in Table IX:
The results of the EMS showed the economic benefit of the
proposed unit commitment with a rolling horizon (UC-RH) in REFERENCES
comparison with a standard UC. The rolling horizon presents [1] R. Palma-Behnke, C. Benavides, E. Aranda, J. Llanos, and D. Sáez,
the advantage of dealing with updated data from the forecast “Energy management system for a renewable based nicrogrid with a
variables. The average operational costs of the microgrid (PV/ demand side management mechanism,” in Proc. 2011 IEEE Symp.
Comput. Intell. Appl. Smart Grid (SSCI 2011 CIASG), Paris, France,
Diesel/ESS) using the UC-RH are reduced for certain profiles of Apr. 11–15, 2011.
demand and irradiance in comparison with the UC. For the case [2] Office of Electricity Delivery and Energy Reliability, U.S. Department
of the wind-based microgrid (PV/Diesel/ESS/Wind), the opera- of Energy, “The smart grid: An introduction,” 2010 [Online]. Avail-
tional costs are also reduced because of a reduction in the energy able: http://www.oe.energy.gov/1165.htm
[3] A. Hajizadeh and M. Aliakbar Golkar, “Intelligent power management
deficit, implying a better management of the ESS in comparison strategy of hybrid distributed generation system,” Electr. Power En-
with the UC approach. The previous results are verified for two ergy Syst., vol. 29, pp. 783–795, 2007.
alternative microgrid configurations. In fact, the overall anal- [4] P. Lombardi, M. Powalko, and K. Rudion, “Optimal operation of a
virtual power plant,” in Proc. IEEE Power Energy Soc. Gen. Meet.
ysis, that include both operational and investment costs, re-con- (PES), 2009.
firms that for both microgrid configurations, the UC-RH solu- [5] D. Lu and B. Francois, “Strategic framework of an energy manage-
tion provides lower total costs than a standard UC approach. ment of a microgrid with a photovoltaic-based active generator,” in
Additionally, note that both UC and UC-RH strategies require Proc. EPE Chapter “Electric Drives” Joint Symp. (ELECTROMO-
TION 2009), Lille, France, Jul. 1–3, 2009.
the same communication infrastructure, just a faster updated of [6] S. Teleke, M. Baran, S. Bhattacharya, and A. Huang, “Rule-base con-
unit set-points is required with UC-RH. trol of battery energy storage for dispatching intermittent renewable
The benefits of DSM are also shown in this paper. Those sources,” IEEE Trans. Sustainable Energy, vol. 1, no. 3, pp. 117–124,
benefits are achieved by means of shifting the behavior of con- 2010.
[7] C. Wang and M. Hashem, “Power management of a stand-Alone wind/
sumers to periods in which there are more renewable resources photovoltaic/fuel cell energy system,” IEEE Trans. Energy Convers.,
available. vol. 23, no. 3, pp. 957–967, 2007.
 This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

PALMA-BEHNKE et al.: A MICROGRID ENERGY MANAGEMENT SYSTEM BASED ON THE ROLLING HORIZON STRATEGY 11

[8] A. Gupta, R. Saini, and M. Sharma, “Steady-state modelling of hybrid Carlos Benavides received his B.Sc. and M.Sc. de-
energy system,” in Proc. 3rd Int. Conf. Electr. Eng. (ICEE), 2009. grees in electrical engineering from the University of
[9] D. Westermann and A. John, “Demand matching wind power gen- Chile. His research field is the planning and operation
eration with wide-area measurement and demand-side management,” of electrical systems in power markets.
IEEE Trans. Energy Convers., vol. 22, no. 1, pp. 145–149, 2007.
[10] V. Hamidi and F. Robinson, “Responsive demand in networks with
high penetration of wind power,” in Proc. IEEE Power Energy Soc.
Gen. Meet.—Convers. Del. Electr. Energy 21st Century, 2008.
[11] J. Lagorse and M. Somoes, “A multiagent fuzzy-logic-based energy
management of hybrid systems,” in Proc. IEEE Ind. Appl. Soc. Annu.
Meet. (IAS), 2009.
[12] T. Logenthiran and D. Srinivasan, “Short term generation scheduling
of a microgrid,” in Proc. IEEE Region 10 Conf. (TENCON 2009).
[13] A. Tsikalakis and N. Haatziargyriou, “Centralized control for opti- Fernando Lanas received his B.Sc. degree in
mizing microgrids operation,” IEEE Trans. Energy Convers., vol. 23, electrical engineering from the University of Chile
no. 1, pp. 241–248, 2008. in 2011. His main research interest is the planning
[14] H. Dagdougui, R. Minciardi, A. Ouammi, M. Robba, and R. Sacile, “A and operation of hybrid energy systems based on
dynamic decision model for the real-time control of hybrid renewable renewable resources.
energy production systems,” IEEE Syst. J., vol. 4, no. 3, pp. 323–333,
Sep. 2010.
[15] H. Morais, P. Kádár, P. Faria, Z. Vale, and H. Khodr, “Optimal sched-
uling of a renewable micro-grid in an isolated load area using mixed-
integer linear programming,” Renewable Energy, vol. 35, no. 1, pp.
151–156, Jan. 2010.
[16] M. Korpas and A. Holen, “Operation planning of hydrogen storage
connected to wind power operating in a power market,” IEEE Trans. Bernardo Severino received his B.Sc. degree in electrical engineering from
Energy Convers., vol. 21, no. 3, pp. 742–749, 2006. the University of Chile in 2011. His main research interest is modeling energy
[17] Matrikon OPC [Online]. Available: http://www.matrikonopc.com/ storage systems for microgrids.
[18] R. Teodorescu, M. Liserre, and P. Rodríguez, Grid Converters for Pho-
tovoltaic and Wind Power Systems. New York: Wiley-IEEE, 2011.
[19] H. Beyer, J. Martínez, M. Suri, J. Torres, E. Lorenz, S. Müller, C. Lorenzo Reyes received his B.Sc. degree in elec-
Hoyer-Klick, and P. Ineichen, “Report on benchmarking of radiation trical engineering from the University of Chile
products,” Sixth Framework Programme MESOR, Management and in 2009. His main research interest is microgrid
Exploitation of Solar Resource Knowledge, 2009 [Online]. Avail- development.
able: http://www.mesor.org/docs/MESoR_Benchmarking_of_radia-
tion_products.pdf
[20] V. Zavala, E. Constantinescu, and M. Anitescu, “Economic impacts of
advanced weather forecasting on energy system operations,” in Proc.
Innov. Smart Grid Technol. (ISGT), 2010.
[21] W. Skamarock, J. Klemo, J. Dudhia, D. Gill, D. Barker, M. Duda, X.
Huang, W. Wang, and J. Powera, “A description of the Advanced Re-
search WRF Version 3,” Tech Rep. 475+STR, NCAR, 2008.
[22] Z. Bashir and M. El-Hawary, “Applying wavelets to short-term load
forecasting using PSO-based neural networks,” IEEE Trans. Power Jacqueline Llanos received her electronic engi-
Syst., vol. 24, no. 1, pp. 20–27, 2009. neering degree from the Escuela Politécnica del
[23] V. Hinojosa and A. Hoese, “Short-term forecasting using fuzzy induc- Ejercito, Ecuador, and her M.Sc. in electrical engi-
tive reasoning an evolutionary algorithms,” IEEE Trans. Power Syst., neering from the University of Chile. Her research
vol. 25, no. 1, pp. 565–574, 2010. fields are predictive control, forecasting of electricity
[24] M. Norgaard, O. Ravn, N. Poulsen, and L. Hansen, Neural Networks demand, and renewable resources.
for Modelling and Control of Dynamic Systems. New York: Springer-
Verlag, 2000.
[25] H. Bergveld, Battery Management Systems Design by Modelling.
Eindhoven, The Netherlands: Univ. Press Facilities, 2001.
[26] D. Guasch and S. Silvestre, “Dynamic battery model for photovoltaic
applications,” Progr. Photovoltaics: Res. Appl., vol. 11, pp. 193–206,
2003. Doris Sáez received her M.Sc. and Ph.D. degrees
[27] R. Palma-Behnke, D. Ortiz, L. Reyes, G. Jiménez-Estévez, N. Garrido, in electrical engineering from the Pontificia Univer-
and N. , “A social SCADA approach for a renewable based microgrid,” sidad Católica de Chile.
in The Huatacondo Project, 2011 IEEE Power Energy Soc. Gen. Meet., She is currently an Associate Professor in the Elec-
Detroit, MI, Jul. 24–28, 2011. trical Engineering Department at the University of
[28] CPLEX [Online]. Available: http://www-01.ibm.com/software/inte- Chile. Her research fields are fuzzy systems control
gration/optimization/cplex-optimizer/ design, fuzzy identification, predictive control, con-
trol of power generation plants, and control of trans-
Rodrigo Palma-Behnke received his B.Sc. and port systems.
M.Sc. degrees in electrical engineering from the Pon-
tificia Universidad Católica de Chile and a Dr.-Ing.
from the University of Dortmund, Germany.
He is currently an Associate Professor in the Elec-
trical Engineering Department at the University of
Chile. His research field is the planning and opera-
tion of electrical systems in competitive power mar-
kets and new technologies.