Energy 252 (2022) 123991

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Introducing a novel control algorithm and scheduling procedure for

optimal operation of energy storage systems
Armin Ebrahimi*, Masoud Ziabasharhagh
Faculty of Mechanical Engineering, K. N. Toosi University of Technology, Pardis St, Molasadra Ave, Vanak Sq, 19991-43344, Tehran, Iran

a r t i c l e i n f o a b s t r a c t

Article history: One of the principal purposes of using energy storage systems (ESS) are peak shaving and load
Received 9 May 2021 smoothing. In the present study, a novel control algorithm has been developed in order to the optimal
Received in revised form use of ESS for peak load shaving. First, by comparing the complex control algorithm of the present study
21 February 2022
with the control algorithm developed by other researchers (simple control algorithm), the superiority
Accepted 10 April 2022
and advantage of the present work become clear. To do this, five different load demand profiles have
Available online 15 April 2022
been considered and the mentioned algorithms have been applied to them. The results show that the
implementation of the present study novel algorithm on the load demand profiles number one to five
Keywords:
Energy storage system
reduces the maximum of the required grid power to 6.477%, 11.81%, 12.60%, 10.25%, and 3.000% of their
Peak load shaving initial value, respectively. Their variance also reduces their initial value to 14.47%, 14.20%, 28.48%, 39.87%,
Control algorithm and 54.96%, respectively. On the other hand, the implementation of the simple control algorithm has
Power management reduced the maximum of the required grid power to 0.8542% of its initial value only in the fifth profile.
But the variance in the first to fifth profiles is reduced by 9.630%, 5.831%, 14.12%, 21.89%, and 52.00% of
their initial value, respectively. In the following, by applying the novel control algorithm on the first
sample load demand profile and change of various parameters such as ESS capacity, lower and upper
limitation of the grid power, etc. the results are extracted and reported in detail.
© 2022 Elsevier Ltd. All rights reserved.

1. Introduction
Hourly: Consumption increases during the day due to over-

lapping demand in the industrial, commercial and residential
The electrical power systems are essentially dependent on the sectors.
instantaneous balance between supply and demand. Electricity
Transient: Fluctuations for various reasons.
must be used exactly when it is generated [1]. Therefore, it can be
concluded that electricity production at any time is completely a Although the demand for electricity varies from time to time,
function of consumption at that time. On the other hand, the de- electric power systems must supply the demand at all times [3].
mand of customers for electricity, which are divided into industrial, Therefore, using peak load shaving strategies to provide sustainable
residential, and commercial groups, is constantly changing ac- electricity is inevitable. There are three major strategies for peak
cording to the following patterns [2]. load shaving: integration of electric vehicle (EV) to the grid,
demand-side management (DSM), and integration of energy stor-

Seasonal: Due to temperature changes throughout the year that age system (ESS) to the grid [4]. As electric vehicles become more
consumption in the lighting, heating, and cooling sectors are widespread, so do their applications. One of its new applications is
subjected to change. peak load shaving so that their batteries are not only used to supply

Weekly: Demand in the industrial sector decreases significantly the required energy for electric vehicles, but also are considered as
over the weekends, so it will result in a reduction in overall an energy saver. Vehicle to grid (V2G) technology allows a bi-
consumption. directional flow to be created between the grid and the electric

Daily: Demand for electricity varies during the day. vehicle, and both charging and discharging of electrical energy are
possible [5]. This application of EVs faces serious challenges,
including the way of synchronizing charging and discharging op-
* Corresponding author. erations, using a large number of electric vehicles with different
E-mail address: [email protected] (A. Ebrahimi).

https://doi.org/10.1016/j.energy.2022.123991
0360-5442/© 2022 Elsevier Ltd. All rights reserved.
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

owners for peak load shaving, etc. [4]. Demand-side management Fig. 2 categorizes the types of energy storage systems [11]. A grid
(DSM) is a tool for changing the load demand profile to levelize it by energy storage system also called a large-scale energy storage
changing the demand [6]. This is done in different ways including system is used with the aim of storing energy when production
energy price, critical peak pricing (time-sensitive pricing), etc. (from the power plant) is more than consumption and releasing the
Demand-side management is the planning, implementation, and stored energy at the on-peak time during the year. In this way, there
control of all executive activities that have a positive effect on the is no need for electricity production to increase or decrease sharply
electricity customer and motivate and influence them to improve to eliminate the peak consumption time, instead, the production is
their demand level. DSM is one of the most important functions in kept at a constant level and the excess demand for peak load is
the smart grid that allows subscribers to make informed decisions provided by a large-scale energy storage system [12]. Although ESS
about electricity consumption, helping to reduce production at has a high capability for peak load shaving, some significant chal-
peak load demand and reshaping the load demand profile. These lenges should be considered in its application such as ESS sched-
results increase the stability of the smart grid and reduce the cost of uling for optimal operation, and optimal sizing and control of ESS.
electricity and air pollution. Fig. 1 shows the different techniques One of the main challenges of this method is the design of the
for load demand management [7]. Based on this figure, peak clip- operation control algorithm of ESS [4].
ping and valley filling methods focus on reducing the difference Levron and Shmilovitz [13] analytically developed an optimal
between valley and peak levels in order to decrease peak load de- power management strategy for lossless energy storage systems in
mand and increase smart grid security. Peak clipping is a direct load order to peak load shaving. They proved that the optimal method
control method to reduce peak loads, and the valley filling method has two important features. The first feature is that as long as the
at non-peak times is created by applying direct load control. The constraints allow, the generated power, Pg(t) will be constant, or in
load shifting method has wide applications and is the most effec- other words, the generated energy, Wg(t) is a straight line. The
tive load management method in power distribution networks. The second feature is the optimal generated power, Pg(t) as an optimal
strategic conservation method reduces the overall load demand by solution is a continuous function. In other words, the derivative of
increasing energy efficiency. The load building or strategic load the generated energy function, Wg(t) is continuous, so Wg(t) itself
growth method leads to either the replacement of inefficient fossil will be continuous. Therefore, it can be concluded that the optimal
fuel equipment or the improvement of customer productivity by generated energy curve must be tangent to its bounding con-
increasing electric energy. This technique is actually the same as the straints. Bounding constraints of Wg(t) are obtained from the
valley filling technique, except that it is applied to large demands. following relation.
The flexible load method actually enhances smart grid reliability
[8]. DSM is appropriate for peak load shaving but has many limi-
tations, such as the impossibility of shifting customers' activities WL ðtÞ  Wg ðtÞ  WL ðtÞ þ ESScap (1)
from on-peak time to off-peak time in many cases [9]. The third
peak load shaving strategy is the use of energy storage systems where WL(t) is the total energy demand up to the time, t, which
(ESS). Peak load shaving is one of the basic applications of energy should be supplied (is a known function), ESScap is the maximum
storage systems that will play a critical role in the future of smart energy capacity that can be stored in the ESS. According to relation
electric power systems. When electricity supply exceeds demand, (1), it is obvious that at any time, t, the total generated energy
surplus production is stored in the ESS, and when demand exceeds should not be less than the total customers required energy and on
supply, the stored electricity is used to meet the load demand. The the other side, it should not be more than the sum of the customers
purpose of using ESSs is to prevent the increase of grid capacity to needed energy and the ESS maximum capacity. Therefore, relation
the peak consumption rate and also to increase its reliability [10]. (1) shows that at any time, t, Wg(t) is bounded between WL(t) and
WL(t) þ ESScap .

Fig. 1. Demand side management (DSM) different techniques [7].

2
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 2. Various energy storage systems [11].

These features illustrate that the optimal generated power at storage system for Naresuan University was obtained through
any time, t, is either constant or exactly equal to PL(t). There is only optimization by an objective function of minimizing energy costs.
one solution that satisfies both features simultaneously. Therefore, Danish et al. [22] presented a coherent strategy for peak load
the optimal power flow path can be extracted. Although the pro- shaving and smoothing using a battery energy storage system that
posed analytical solution is so instructive and simple, it faces not only achieved the optimal size of the battery energy storage
serious limitations. One of these limitations is that ESS was system (BESS) but also determined its optimal location in the study.
assumed lossless, which is proper for small-scale energy storage Following a proper control algorithm leads to optimal utilization
systems like super-capacitors but not for large-scale energy storage of energy storage system and also generated power of the grid.
systems such as pump-hydro and compressed air energy storage Determining the optimal amount of generated power of the grid as
systems. Numerical methods should be used to find the optimal well as the ESS charging and discharging rate at any time of the
solution for these ESSs. In another study by Levron and Shmilovitz time period are the results of applying the control algorithm to ESS.
[14], a simple power management method was provided to find the In this study, a novel control algorithm and scheduling procedure of
optimal power generation while using a lossy energy storage sys- energy storage systems is developed for the purpose of peak load
tem. In this method, using an assumed generation level, and shaving and smoothing. The proposed algorithm is quite practical
changing it for minimization, the optimal power generation func- yet simple so that the optimal size of ESS as well as its optimum
tion was found. one of the advantages of this method is its operation scheduling can be determined for various load demand
simplicity and the disadvantage of that is incapable of peak load profiles with different characteristics.
shaving in all cases (with different PL(t)) and from this perspective Innovation of the proposed control algorithm can be considered
is completely dependent on the curve of load demand. In the pre- in its comprehensiveness, so that to apply to different load demand
sent study, this control algorithm is named simple control algo- profiles and energy storage systems with various capacity is fully
rithm, so that it can be compared with the novel control algorithm compatible without considering changes in the algorithm. In other
of this study. Chapaloglou et al. [15] investigated load smoothing words, the proposed control algorithm is applicable for peak
and peak shaving by presenting new management of power flows shaving of various domain of usage, including hourly and daily peak
algorithm. First, using an artificial neural network (ANN), load load shaving that requires small energy storage system or annual
forecasting of a power system is performed then using a diesel peak load shaving that requires large energy storage system.
generator, photovoltaic, and battery energy storage system, the Another innovation of the proposed control algorithm is the se-
proposed algorithm is applied and optimal power flow is extracted. lective use of the ESS stored energy before it was fully charged.
Chua et al. [16] developed an optimal sizing method of the energy Most of the developed control algorithms, including the control
storage system for economic benefit. This method examined the algorithm presented in Refs. [13,14], do not allow the energy stored
historical load demand profiles of commercial and industrial cus- in the ESS to be discharged until it is fully charged. However, the
tomers and suggested the most cost-effective ESS size for various proposed algorithm, considering the upper and lower bounds for
tariff patterns. In this regard, the energy storage system was also the generated power profile (Pg (t)), allows the ESS to be discharged
calculated by trading off between costs and benefits. In another optimally when necessary, even during partially charge times.
study by Karmiris and Tengner [17], the optimal shave level was In the following, first, the methodology of the control algorithm
extracted by using historical data as well as considering ESS ca- is described and compared with a similar algorithm, then the
pacity and ESS charge and discharge rates. Lu et al. [18] presented implementation of the algorithm on the hypothetical load demand
the optimal sizing and control method of ESS for peak load shaving. profile is reported and the results are reviewed.
In this regard, two different control models with different objective
functions were developed, one with the approach of minimizing
2. Method statement
peak-valley differences and the other with the approach of mini-
mizing daily load demand variance. Barzkar and Hosseini [19]
Energy storage systems are used for peak load shaving and load
developed a control algorithm for peak load shaving that is
leveling. According to Fig. 3, PL(t), which is the load demand at any
appropriate for a residential distributed energy storage system
time, t, must be supplied by the power system. For this purpose,
(DESS). Baker et al. [20] studied the uncertainty of renewable en-
either grid power rate (Pg(t)) or ESS power rate (Ps(t)) should be
ergy forecasting in peak load shaving and optimized the sizing of an
used directly. The design of how to meet the load demand at any
energy storage system. Prasatsap et al. [21] applied the peak load
time, t, which indicates how to charge and discharge the energy
shaving procedure to Naresuan University as a case study, intend-
storage system, is called the control algorithm and scheduling
ing to reduce electricity cost. The optimal size of the used energy
procedure. By properly designing the control algorithm and

3
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 3. Control algorithm and scheduling procedure role for applying ESSs on load power.

considering the load and ESS characteristics, the optimum solution calculated:
can be found to provide load demand using ESS stored energy. In
the present work, the control algorithm is developed with two
approaches. The first one is suggested without the upper and lower ðt
limitations for Pg(t) and the second by considering that. The WL ðtÞ ¼ PL ðtÞdt (4)
following hypotheses are considered for the development of the
0
proposed algorithm.
To calculate the ESS stored energy (Ws (t)) at the time, t, the ESS

The proposed algorithm for peak load shaving is applied to the charging and discharging efficiencies must be considered as:
annual load demand, which can be used for other types of loads
with slight changes.

The load demand profile is considered as input. DT out

ESS specifications including hch and hdis are provided as input. Ws ðt þ DtÞ ¼ Ws ðtÞ þ hch DTPsin ðtÞ  P ðtÞ (5)
hdis s

where hch and hdis are charging and discharging efficiencies,

respectively. The following equation is used to calculate Ws (t):
2.1. Mathematical model

At any time, t, the generated power of the grid (Pg(t)) is equal to 8 9

the sum of load demand (PL(t)) and ESS power rate (Ps(t)): < hch *Ps ðTÞ Ps ðTÞ  0 >
ðt > =
Ws ðtÞ ¼ 1 dt (6)
>
: * Ps ðTÞ Ps ðTÞ < 0 >
;
Pg ðtÞ ¼ Ps ðtÞ þ Pl ðtÞ (2) 0 hdis
where Ps(t) is positive while ESS is charging and negative while Since at any time, t, the total generated energy of the grid cannot
discharging. It is known that except for ESS charging and dis- be less than the total load energy and on the other side cannot be
charging times, the output power of the grid is equal to load de- more than the sum of load energy and ESS capacity. This constraint
mand. The total generated energy of the grid up to time, t, is is shown as equation (1). At any time, t, ESS stored energy (Ws (t)),
obtained from the following equation: cannot be more than ESS capacity so:

ðt
Wg ðtÞ ¼ Pg ðtÞdt (3) 0  Ws ðtÞ  ESScap (7)
0
The relationship between hch and hdis with round-trip efficiency
In the same way, the total load energy up to time, t, can be hRTE is as follows:
4
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

respectively. By changing the PIG values as well as the size of ESS,

Energy output from the ESS
hRTE ¼ the proposed algorithm can be implemented on PL in different
consumed energy for charching the ESS conditions and the optimum one can be obtained. After calculating
ð
the required inputs, the control algorithm operates in such a way
Ps ðtÞ < 0 Ps ðtÞdt
that initially Pg(t) should be obtained from the maximum of PIG and
¼ ð (8)
PL(t) at any time, t, until the ESS charging is completed. it means:
Ps ðtÞ > 0 Ps ðtÞdt
Pg ðtÞ ¼ MaxfPIG ; PL ðtÞg (14)
hdis :Dw Using this relation, Pg(t) at least will be equal to PL(t) at any time,
hRTE ¼ ¼ hdis :hch (9)
h1
ch
:Dw t, so the ESS charging process will take place gradually. It is
emphasized that by changing the PIG as well as ESScap, the required
According to the above equation, hRTE is obtained by multiplying time of ESS charging can be adjusted. According to Fig. 4, after the
hch by hdis. ESS is fully charged, it should be gradually discharged so that at
times after the completion of the ESS, Pg(t) is obtained from the
2.2. The novel control algorithm definition minimum of PIG and PL (t), that's mean:

As mentioned, it is assumed that load demand can be estimated Pg ðtÞ ¼ Min fPIG ; PL ðtÞg (15)
from historical data in the time period, so it is one of the inputs. In
Using this relation, the maximum amount of Pg(t) at any time, t,
the present study, load power is available per day during a year.
will be equal to PL(t), so the ESS discharging process will take place
First, the statistical characteristics of the load profile must be
gradually. By repeating the explained algorithm until the end of the
extracted. For this purpose, its average and standard deviation are
time period, Pg(t) based on peak load shaving and load smoothing is
calculated. The relations of average and the standard deviation for
generated as one of the basic outputs of the proposed control
load demand profile (PL (t)) are as follows:
algorithm.
1 XN
mðPL Þ ¼ P ðiÞ (10)
N i¼1 L 2.2.2. Complex control algorithm
If the simple control algorithm is followed, the peak load
!12 shaving purpose may not be achieved. Because there is no limit for
N  2
1 X Pg(t). In fact, Pg(t) can even have maximum and minimum of PL(t)
sðPL Þ ¼ PL ðiÞ  mðPL Þ (11)
N i¼1 values. In order to prevent this situation, the upper and lower
limitations are considered for Pg(t) in the complex control algo-
where N, equals the total number of times, t that the algorithm rithm, so that it cannot be more than the upper level and less than
applies to them. After determining the statistical characteristics of the lower level. By applying this limitation, the output Pg(t) will
the load power profile, the proposed control algorithm of this study necessarily be smoother and lead to complete peak load shaving.
should be described. In the present study, first, the control algo- The complex control algorithm flowchart is shown in Fig. 5. Ac-
rithm developed by the research [14] is described as the simple cording to the figure, Pg-lower and Pg-upper, which are the lower and
control algorithm and its advantages and disadvantages are listed. upper limitations of Pg(t), respectively, are calculated from the
It should be noted that in the mentioned reference, its concept is following relations:
given, but in this research, its formulation is completely presented
in the form of an algorithm (see Fig. 4). The novel control algorithm Pglower ¼ Min ðPL Þ þ x0 :sPL (16)
developed in the present work is then described as the complex
control algorithm in detail. In the simple one, there is no limit to the Pgupper ¼ Max ðPL Þ þ y0 :sPL (17)
generated power (Pg(t)) so that Pg(t) can even reach the maximum
and minimum PL(t). But in the complex one, the upper and lower In which x0 and y0 can have different values according to the load
limitations are considered for Pg(t) and it is not allowed to exceed demand profile and its various conditions must be evaluated to
these values in the design. In the following, simple and complex investigate its optimum. The procedure of complex control algo-
algorithms are explained in detail. rithm is similar to the simple control algorithm, except that if
during execution, it reaches a point where Pg(t) should take values
greater than Pg-upper, then Pg(t) is equal to Pg- upper and the differ-
2.2.1. Simple control algorithm
ence will be provided by ESS. If Ws(t) is not enough to compensate,
In the simple control algorithm developed by Ref. [14] (Fig. 4),
then Pg(t-1) is set to Pg-upper and Ws(t) is recalculated. If it still does
the load demand profile, hch and hdis are first received as input.
not suffice, then Pg(t-2) is also equal to Pg-upper, this process con-
Then PL(t) specifications including sPL and mPL as well as WL are
tinues until the energy stored in the ESS can provide the difference
calculated. By these data, the algorithm inputs are completed. In
between Pg(t) and PL(t). The same applies to lower limitation. If
the algorithm, a level of production for Pg (PIG) as well as energy
during the application of the algorithm, it reaches a point where
storage capacity (ESScap) should be considered. The following re-
Pg(t) should take values less than Pg-lower, then Pg(t) is set equal to
lations are used to set the initial values of these variables:
Pg-lower and the difference should be stored in ESS. If Ws(t) is equal
PIG ¼ mPL þ xsPL (12) to its maximum ESScap, then Pg(t-1) is set to Pg-lower and Ws(t) is
recalculated. If the required empty space is still not created in the
ESS, then Pg(t-2) is considered equal to Pg-lower. This process con-
ESScap ¼ yWL (13)
tinues until there is enough free space in the ESS. Fig. 5 shows the
In which x and y can be changed in the intervals [0,1] and [0,0.1], way of applying the provided explanations. It should be noted that

5
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 4. Flowchart of the simple control algorithm developed by this study.

the applicability or non-applicability of the complex control algo- algorithm can be used to apply on energy storage systems with
rithm developed in the present study for small ESSs depends on different capacities to find the optimal solution.
their considered assumptions. Large or small ESS is affected by two In section 3, the control algorithms apply to different load de-
factors: its round-trip efficiency and also its scope of application mand profiles and their results were compared to each other.
(domain of usage) (hourly, daily or more peak shaving). As the ESS
gets smaller, the RTE typically increases, like super-capacitors, and
it is also used for smaller range peak load shavings. In the present 3. Applying the proposed control algorithm on different
algorithm, there are no restrictions on the mentioned factors, so hypothesis load demand profile
that charging and discharging efficiencies or round-trip efficiency
can be changed as the input of the algorithm, and also the proposed In this section, the results of applying complex and simple
algorithm can also be used for hourly and daily, etc. peak load control algorithms on hypothetical load demand profiles are
shavings. Therefore, it can be concluded that the novel control compared with each other. Five different load demand profiles with
different concavities and convexities are considered, the statistical
6
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 5. Flowchart of the complex control algorithm developed by this study.

characteristics of which are shown in Table 1. Table 2 lists the as- Pg(t) variance of complex and simple control algorithms decreased
sumptions for applying the control algorithms to each of the load by 14.47% and 9.630% compared to the PL(t) variance, respectively.
demand profiles. But in terms of peak shaving, the two algorithms have worked
The first hypothesis load demand profile, along with the results differently. The complex control algorithm of the present study was
of applying the complex control algorithm as well as the simple able to reduce the maximum Pg(t) to 9125 MW as the assumptions,
control algorithm [14], are shown in Fig. 6. According to this figure, i.e. 6.477% less than the initial value, but the simple control algo-
both algorithms succeeded in load smoothing so that the output rithm failed in this regard and could not reduce the required

7
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Table 1 maximum grid power. It is necessary to mention that by changing

Statistical data of proposed load demand profiles (PL). the initially assumed parameters, a better response can be achieved
Parameter Sample load demand profile in each of these algorithms, But the complex control algorithm due
First Second Third Fourth Fifth
to the imposed restrictions (considering the upper and lower lim-
itation) will definitely operate better in peak load shaving and load
mPL ðMWÞ 6263 5565 4608 5186 4415
smoothing and under any circumstances, its output will decrease
sPL ðMWÞ 2108 989.6 877.5 729.7 287.3
Max (MW) 9757 7542 6272 6411 4800 the peak value. Fig. 7 shows the results of applying both control
Min (MW) 2243 3152 3018 3783 4000 algorithms to the second hypothetical load demand profile. This
WL (MW.day) 2284149 2030651 1681996 1892214 1611723 profile is faced with peaks with different values. The simple control
algorithm has failed to eliminate these peaks so that it could not
shave the first peak, and in the second peak, due to the small ESS
Table 2 capacity, the ESS is completely discharged during the peak load
Initial value of parameters for designing based on the control algorithms. shaving, so the second peak is incompletely shaved, but by
Parameter Sample load demand profile changing the size of the ESS, a better response can be obtained. In
the third peak of the profile, the simple control algorithm per-
First Second Third Fourth Fifth
formed better and the intended peak load shaving occurred.
hch 0.7 0.7 0.7 0.7 0.7
Regarding the application of the complex control algorithm on the
hdis 0.7 0.7 0.7 0.7 0.7
PIG (MW) 6790 5911 4739 5295 4515
second profile, it can be seen that all the peaks of the profile have
ESScap (MW.day) 68524 20307 16820 18922 19341 been effectively shaved. By applying this algorithm, the maximum
Pg-upper (MW) 9125 6651 5482 5754 4656 Pg(t) is reduced to 11.81% of its initial value and reaches 6651 MW.
Pg-lower (MW) 2876 4043 3808 4439 4144 The results of applying these algorithms on the third hypothetical
Ws(at t ¼ 0) (MW.day) 26725 9138 4205 4731 4835
load demand profile are shown in Fig. 8. According to this figure,
N (days) 365 365 365 365 365
the third profile has more peaks and valleys. As expected, the
simple control algorithm failed in shaving some of these peaks so
that it could not remove the maximum peak of the profile. On the
other hand, the complex control algorithm has been more suc-
cessful and has reduced the maximum Pg(t) by 12.60% from its
initial value to 5482 MW. Fig. 9 shows the results obtained by
applying algorithms to the fourth hypothetical load demand profile.
This profile contains sharper peaks. The performance of the simple
control algorithm is unsatisfactory as expected so that it was un-
successful in peak load shaving but relatively effective in load
smoothing. The results of applying the complex control algorithm
show that it is quite successful in peak load shaving and can reduce
its amount to 4656 MW. Also, the variance of the Pg(t) function
decreased by 54.96% of its initial value. Considering the imple-
mentation of the mentioned algorithms on 4 hypothetical profiles,
it can be concluded that the simple control algorithm operates
better on smoother profiles with fewer peaks and valleys, while the
complex control algorithm due to the considered concept of limi-
tation (upper and lower limitation for Pg(t)) will work well in all
profiles. To determine the proper performance of the simple control
algorithm on smoother profiles, the fifth profile is considered ac-
cording to Fig. 10. According to the figure, the application of the
simple algorithm is quite effective as it reduces the variance and
maximum Pg(t) to 137.9 MW and 4759 MW, respectively. By
considering similar assumptions, the complex control algorithm
performs better than another control algorithm, reducing the
variance and maximum Pg(t) to 129.4 MW and 4656 MW,
respectively.
As a conclusion of the application of control algorithms on
different load demand profiles can express that, in case of facing
simpler profiles (with fewer and smoother peaks and valleys), the
reference control algorithm [14] (simple control algorithm) can be
effective in peak load shaving and load smoothing, while the per-
formance of complex control algorithm not only in the same con-
ditions as observed, can be more effective, but can show the proper
performance of peak shaving and load smoothing by facing with
different profiles with more and sharper peaks and valleys. In other
Fig. 6. a). The first hypothesis load demand profile for applying the proposed control words, it is obvious that by considering Pg-lower and Pg- upper as the
algorithm on it. b). The results of applying the simple control algorithm (PL(t) vs. Pg(t)). allowable range of Pg selection at any time, t, the purpose of
c). The results of applying the complex control algorithm (PL(t) vs. Pg(t)).

8
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 7. a). The second hypothesis load demand profile for applying the proposed control algorithm on it. b). The results of applying the simple control algorithm (PL(t) vs. Pg(t)).c).
The results of applying the complex control algorithm (PL(t) vs. Pg(t)).

9
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 8. a). The third hypothesis load demand profile for applying the proposed control algorithm on it. b). The results of applying the simple control algorithm (PL(t) vs. Pg(t)). c). The
results of applying the complex control algorithm (PL(t) vs. Pg(t)).

complete peak load shaving can be achieved. It is concluded that by 4. Results and discussion
allowing the ESS to be discharged before fully filled in certain
conditions (selectively and intelligently), the results are completely The results of applying the control algorithm developed in this
improved compared to using a simple control algorithm. The re- study on the first hypothetical load demand profile of the previous
sults are briefly listed in Table 3. section are detailed in this section. Taking into account the assumed

10
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 9. a). The fourth hypothesis load demand profile for applying the proposed control algorithm on it. b). The results of applying the simple control algorithm (PL(t) vs. Pg(t)). c).
The results of applying the complex control algorithm (PL(t) vs. Pg(t)).

parameters in Table 2, the changes of total generated energy of grid time, t are displayed in Fig. 11. According to the figure, it is obvious
(Wg(t)), total load energy (WL(t)), ESS stored energy (Ws(t)), that Wg(t) will be greater than or equal to WL(t) at any time. Their
generated power of grid (Pg(t)), and load demand (PL(t)) in terms of final difference is also due to losses during ESS charging and

11
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 10. a). The fifth hypothesis load demand profile for applying the proposed control algorithm on it. b). The results of applying the simple control algorithm (PL(t) vs. Pg(t)). c). The
results of applying the complex control algorithm (PL(t) vs. Pg(t)).

discharging. It should also be noted that there is no limit to the between Wg(t) and WL(t) can be related to the difference between
initial and final values of ESS stored energy and different values can the initial and final values of ESS stored energy (Ws(t ¼ 0), and
be taken. The main reason is the continuity before and after the Ws(t ¼ T)). As expected, the complex control algorithm has been
considered time period, t. Therefore, a small part of the difference more successful than another in peak load shaving. Table 3

12
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Table 3 4.2. Generated power level (PIG) changes effects on the performance
Statistical data of the designed Pg(t) based on the proposed control algorithms. of the complex control algorithm
Parameter Simple control algorithm Complex control algorithm

The first sample load demand profile

The effect of generated power level change (PIG) on peak load
shaving is shown in Fig. 13. PIG represents a base-load power that
mPg (MW) 6486 6484
must be supplied by power plants. An increase in PIG directly in-
sPg (MW) 1905 1803
Max (Pg) (MW) 9757 9125 dicates an increase in the capacity of base-load power plants. On
Min (Pg) (MW) 2243 2876 the other hand, PIG should be selected in such a way that is used
Wg (MW.day) 2390742 2393027 most of the time so that base-load power plants are used in full-
The second sample load demand profile load at most times and their standby time is minimized.
mPg (MW) 5627 5587
sPg (MW) 931.9 849.1 4.3. Lower and upper limitation of the generated power changes
Max (Pg) (MW) 7542 6651 effects on the performance of the complex control algorithm
Min (Pg) (MW) 3152 4043
Wg (MW.day) 2054621 2039831
One of the most important design parameters for using the
The third sample load demand profile complex control algorithm is Pg-lower and Pg- upper. Fig. 14 shows the
mPg (MW) 4688 4642 effect of changing these parameters on the peak load demand
sPg (MW) 753.6 627.6 shaving process. Based on the figure, by reducing the interval be-
Max (Pg) (MW) 6272 5482 tween Pg-lower and Pg-upper, a smoother Pg(t) can be designed.
Min (Pg) (MW) 3018 3808
Definitely, this interval must be selected in such a way that it shaves
Wg (MW.day) 1707689 1690800
the peak and valley loads and subsequently, the loads around the
The fourth sample load demand profile
PIG are shaved by the PIG itself. Therefore, it can be concluded that
mPg (MW) 5312 5243 there is a close relationship between the choice of ESS capacity,
sPg (MW) 570.0 438.8 generated power level, lower, and upper limitation of the generated
Max (Pg) (MW) 6411 5754
Min (Pg) (MW) 3783 4439
power and the interaction between them was described. Therefore,
Wg (MW.day) 1952928 1911142 in order to achieve the optimum design, the changes of the
mentioned parameters should be considered together so that the
The fifth sample load demand profile
peak load shaving method can be discussed more comprehensively.
mPg (MW) 4476 4472 Table 4 shows the standard deviation changes of the generated
sPg (MW) 137.9 129.4
Max (Pg) (MW) 4759 4656
power profile as well as the total generated energy of the grid by
Min (Pg) (MW) 4012 4144 changing the mentioned parameters.
Wg (MW.day) 1629950 1628194
5. Conclusion

compares the statistical characteristics of Pg(t) obtained from each Optimal use of energy storage systems for peak shaving and load
of the algorithms. Then, the performance of the complex control smoothing purposes requires following a proper control algorithm
algorithm was investigated in different conditions by changing the that provides the best solution according to the load demand
ESS capacity, PIG, Pg-upper, and Pg-lower. profile that must be supplied at all times. The appropriate solution
specifies the amount of power generated by the grid and the ESS
charging and discharging rate at any time. In the present study, a
novel control algorithm is developed for the optimal use of ESS for
4.1. ESS capacity changes effect on the performance of the complex peak load shaving. In this regard, first, a simple control algorithm is
control algorithm described. Although the mentioned algorithm is so efficient for load
smoothing, it may not be completely effective for peak shaving in
In Fig. 12, by changing the energy storage system capacity, the some situations covered in section 3. According to the sub-sections
peak load shaving method is reported based on the complex control 2.2.1 and 2.2.2, the control algorithm is described in two modes,
algorithm. According to the figure, it is obvious that increasing the simple and complex, the first provided by the reference [14]
ESS capacity can have both a negative and a positive effect on the without the limitation for Pg(t) and the second one by considering
peak load shaving process, so that if the ESS capacity is considered limitation. According to the algorithm presented in the reference
too large, then the algorithm can be stopped in the ESS charging [14] (simple control algorithm), ESS cannot be discharged until it is
mode, that means most of the examined time period is spent fully charged, even when the ESS charge is nearing completion, it is
charging the ESS. On the other hand, by considering ESS capacity not allowed to use stored energy. A complex control algorithm has
too small, peak load shaving is disrupted so that in on-peak times, been developed to address this defect. This algorithm effectively
the stored energy is not available to supply the required load de- allows the ESS to be emptied even when it is not full. For instance,
mand. It can be concluded that the ESS capacity in the optimal when ESS stored energy does not reach its maximum equivalent to
range, which is directly affected by the values of PIG, Pg-lower, and Pg- ESScap, according to the complex control algorithm, Pg(t) should be
upper, can be changed to achieve better peak load shaving. selected from equation (14) unless PL(t) is greater than Pg- upper, in

13
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 11. a). The simple control algorithm applying effects on Wg(t), Ws(t), and Ps(t). b). The complex control algorithm applying effects on Wg(t), Ws(t), and Ps(t).

14
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 12. ESS capacity changes effects on the peak load shaving trend.

Fig. 13. PIG changes effects on the peak load shaving trend.

15
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

Fig. 14. Pg upper and lower bounds changes effects on the peak load shaving trend.

Table 4
Main parameters simultaneous changes effects on sPg and Wg .

Pg-upper
PIG Max (PL)-0.4* sPL Max (PL)-0.3* sPL Max (PL)-0.2* sPL ESScap
mPL þ0.50*sPL sPg ¼ 1977 sPg ¼ 2008 sPg ¼ 2050 0.01*WL
Wg ¼ 2331200 Wg ¼ 2329200 Wg ¼ 2330900
mPL þ0.25*sPL sPg ¼ 1766 sPg ¼ 1803 sPg ¼ 1860 0.03*WL
Wg ¼ 2389300 Wg ¼ 2393000 Wg ¼ 2390200
sPg ¼ 1555 sPg ¼ 1612 sPg ¼ 1670 0.06*WL
mPL þ0.00*sPL Wg ¼ 2345900 Wg ¼ 2339200 Wg ¼ 2344000
Min (PL)þ0.4* sPL Min (PL)þ0.3* sPL Min (PL)þ0.2* sPL
Pg-lower

which case Pg- upper is selected and its difference with PL(t) is pro- assumptions as before, and the results indicate that this algorithm
vided by ESS stored energy. This means that depletion is allowed is applicable for peak shaving in all conditions. By applying the
before the ESS charge is completed. It is important to note that a complex control algorithm to the first to fifth profiles, the
major advantage of the complex control algorithm is that it allows maximum amount of the grid power is reduced to 6.477%, 11.81%,
the ESS to be discharged before it is fully filled only at critical times, 12.60%, 10.25, and 3.000% of their initial value, respectively. It is
which peak and valley shaving must be done as described above necessary to mention that by changing parameters of the complex
and stored energy is used effectively and efficiently. In other words, control algorithm such as ESScap, Pg-lower, Pg-upper, and PIG, more
it does not allow ESS discharging under any circumstances. appropriate solutions can be extracted. In the complex control al-
The simple control algorithm has been applied on sample load gorithm, in comparison with the simple one, by adding lower and
demand profiles and its advantages and disadvantages are clarified. upper limitations for the variations range of the grid generated
The results show that by considering the first to fifth sample load power (Pg(t)), more effective peak load shaving is presented.
demand profiles and applying the simple control algorithm on Indeed, the complex control algorithm is compatible with all load
them, only in the fifth profile is the maximum reduction of the grid demand profiles with any characteristics and can extract the
power generation observed. In other words, in the other four pro- optimal solution. In the proposed algorithm, the parameters of
files, this algorithm is not capable of peak shaving. It can be lower and upper limitations of the grid generated power, ESS ca-
concluded that the results of the simple control algorithm depend pacity as well as generated power level can all be changed, so by
entirely on the intended profile and it is not possible to expect the changing them, various solutions can be found to better meet peak
appropriate response in all profiles with different specifications. In load shaving purpose, each of which can be effective due to the real
contrast, the complex control algorithm developed by the present limitations of the system. In the results section, by changing these
study has been applied to the load demand profiles with the same parameters, the proposed control algorithm is analyzed from

16
A. Ebrahimi and M. Ziabasharhagh Energy 252 (2022) 123991

various aspects. Taking into account the first sample load demand [14] Levron Y, Shmilovitz D. Power systems' optimal peak-shaving applying sec-
ondary storage. Elec Power Syst Res 2012;89:80e4.
profile, the mentioned parameters have been changed in certain
[15] Chapaloglou S, et al. Smart energy management algorithm for load smoothing
intervals and the results are given. By increasing Pg-upper and Pg- and peak shaving based on load forecasting of an island's power system. Appl
lower, as expected, peak shaving is performed more efficiently, so Energy 2019;238:627e42.
that the maximum and variance of the grid power (Pg(t)) can be [16] Chua KH, Lim YS, Morris S. Energy storage system for peak shaving. Inter-
national Journal of Energy Sector Management; 2016.
reduced to 8.640% and 16.22% of their initial values, respectively. By [17] Karmiris G, Tengne r T. Peak shaving control method for energy storage.
increasing the ESS capacity, the Pg(t) can be obtained smoother, so Vasterås, Sweden: Corporate Research Center; 2013.
that by doubling the ESS capacity compared to Table 2, the variance [18] Lu C, Xu H, Pan X, Song J. Optimal sizing and control of battery energy storage
system for peak load shaving. Energies 2014;7(12):8396e410.
can be reduced by 21.08% of the initial value. It should be noted that [19] Barzkar A, Hosseini SMH. A novel peak load shaving algorithm via real-time
although increasing the ESS capacity can provide a more appro- battery scheduling for residential distributed energy storage systems. Int J
priate response, but this increase directly affects the investment Energy Res 2018;42(7):2400e16.
[20] Baker K, Hug G, Li X. Energy storage sizing taking into account forecast un-
costs. Therefore, considering the economic aspect, a fairer and more certainties and receding horizon operation. IEEE Trans Sustain Energy
accurate judgment can be provided. 2016;8(1):331e40.
[21] Prasatsap U, Kiravittaya S, Polprasert J. Determination of optimal energy
storage system for peak shaving to reduce electricity cost in a university.
Credit author statement Energy Proc 2017;138:967e72.
[22] Danish SMS, Ahmadi M, Danish MSS, Mandal P, Yona A, Senjyu T. A coherent
Armin Ebrahimi: Conceptualization, Methodology, Software, strategy for peak load shaving using energy storage systems. J Energy Storage
2020;32:101823.
Validation, Formal analysis, Data curation, Writing - original draft,
Writing - review & editing - Resources. Masoud Ziabasharhagh:
Nomenclatures
Supervision, Resources.

Declaration of competing interest

Acronyms/Abbreviations

The authors declare that they have no known competing

ANN: Artificial Neural Network
financial interests or personal relationships that could have BESS: Battery Energy Storage System
appeared to influence the work reported in this paper. DESS: Distributed Energy Storage System
DSM: Demand Side Management
ESS: Energy Storage System
References EV: Electric Vehicle
Max: Maximum
[1] Ebrahimi A, Ghorbani B, Taghavi M. Pinch and exergy evaluation of a liquid Min: Minimum'
nitrogen cryogenic energy storage structure using air separation unit, lique- RTE: Round-Trip Efficiency
faction hybrid process, and Kalina power cycle. J Clean Prod 2021:127226. V2G: Vehicle to Grid
[2] Yongliang L. Cryogen based energy storage: process modelling and optimi-
sation. University of Leeds; 2011. Greek letters/symbols
[3] Wang C, Zhao Y, Liu M, Qiao Y, Chong D, Yan J. Peak shaving operational
optimization of supercritical coal-fired power plants by revising control
strategy for water-fuel ratio. Appl Energy 2018;216:212e23. h: Efficiency
[4] Uddin M, Romlie MF, Abdullah MF, Syahirah, Halim A, Tan. A review on peak RS: Summation sign
load shaving strategies. Renew Sustain Energy Rev 2018;82:3323e32. : Integration sign
[5] Arnaudo M, Topel M, Laumert B. Vehicle-to-grid for peak shaving to unlock D: Difference operator
the integration of distributed heat pumps in a Swedish neighborhood. En- m: Mean
ergies 2020;13(7):1705. s: Standard deviation
ndez C, Belusko M, Boer D, Cabeza LF. Optimized de-
[6] Saffari M, Alvaro, Ferna
mand side management (DSM) of peak electricity demand by coupling low Variables/parameters
temperature thermal energy storage (TES) and solar PV. Appl Energy
2018;211:604e16.
N: Number of t
[7] Gellings CW. The concept of demand-side management for electric utilities.
P: Power
In: Proceedings of the IEEE, vol. 73; 1985. p. 1468e70. 10.
t: Time
[8] Ould Abdeslam D, Fatiha L, Haffaf A. Photovoltaic customer generation as a
T: Total time period
concept of demand side management. Electrotehnica 03/29 2019;67:13e20.
W: Energy
[9] Arteconi A, Hewitt NJ, Polonara F. State of the art of thermal storage for
demand-side management. Appl Energy 2012;93:371e89.
Superscripts and Subscripts
[10] Md Mustafizur R, Abayomi Olufemi O, Eskinder G, Amit K. Assessment of
energy storage technologies: a review. Energy Convers Manag 2020;223:
113295. cap: Capacity
[11] Liu J, Hu C, Kimber A, Wang Z. Uses, cost-benefit analysis, and markets of ch: Charging
energy storage systems for electric grid applications. J Energy Storage dis: Discharging
2020;32:101731. g: Generated
[12] Kapila S, Oni AO, Kumar A. The development of techno-economic models for IG: Initial Guess
large-scale energy storage systems. Energy 2017;140:656e72. L: Load
[13] Levron Y, Shmilovitz D. Optimal power management in fueled systems with s: Stored
finite storage capacity. IEEE Transactions on Circuits Systems I: Regular Papers
2009;57(8):2221e31.

17