1 s2.0 S0306261923011650 Main

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Applied Energy 350 (2023) 121801

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Dynamic game optimization control for shared energy storage in multiple


application scenarios considering energy storage economy
Xiaojuan Han , Jiarong Li *, Zhewen Zhang
School of Control and Computer Engineering, North China Electric Power University, Beijing 102206, China

H I G H L I G H T S

• The profit relationship between multiple stakeholders in auxiliary services and energy storage needs is explored.
• Double-level optimization control model for shared energy storage system in multiple application scenarios is established.
• The combinatorial optimal scheduling problem in decision sets is solved by adaptive greedy search algorithm.
• The non dominated sorting beluga whale optimization algorithm effectively solves multi-objective optimization problems.

A R T I C L E I N F O A B S T R A C T

Keywords: In response to poor economic efficiency caused by the single service mode of energy storage stations, a double-
Shared energy storage system level dynamic game optimization method for shared energy storage systems in multiple application scenarios
Non-dominated sorting beluga whale considering economic efficiency is proposed in this paper. By analyzing the needs of multiple stakeholders
optimization algorithm
involved in grid auxiliary services, fully tap into the profitability potential of energy storage stations. The ca­
Adaptive greedy search algorithm
Peak regulation and frequency regulation
pacity of the shared energy storage system is optimized by the non-dominant sorting beluga whale optimization
Energy market algorithm (NSBWOA) in the upper level, and the operation strategy under multiple scenarios is optimized by the
Capacity configuration adaptive greedy search algorithm (AGSA) in the lower level. With the goal of maximizing the gross annual total
income and high-value peak regulation ratio, and minimizing the cost- income ratio, the optimal capacity
configuration and operation strategy of the shared energy storage system are obtained through collaborative
optimization between upper and lower level models. The effectiveness of the proposed method is verified
through the simulation testing of actual operating data of a certain power grid in China. Simulation results show
that the gross annual income and high-value peak regulation ratio across multiple scenarios (Scenario III) are
the highest, and the cost-income ratio is at an acceptable low level, which can provide a theoretical basis for the
large-scale application of energy storage systems in new power systems.

1. Introduction energy storage system for multiple application scenarios participating in


power grid auxiliary services has advantages such as more efficient
Under the background of dual carbon goals and new power system, scheduling and operation, more controllable safety and quality [3], and
local governments and power grid companies in China proposed a more prominent economic benefits [4]. Therefore, according to the
centralized “renewable energy and energy storage” development policy, spatiotemporal characteristics of renewable energy power stations,
which fully reflects the value of energy storage for the large-scale exploring the differences in various energy storage responses for
popularization of new energy and forms a consensus [1]. The econ­ different services, and the study on the optimization control technology
omy of the energy storage system is the key factor that restricts its large- of the shared energy storage system among multiple stakeholders can
scale application. However, conventional energy storage stations often effectively improve the stability of power grids [5] and the consumption
serve a single renewable energy power station, and its business model is capacity of renewable energy [6], and promote the healthy and orderly
simple which is not enough to achieve the economic operation of energy development of the regional clean energy.
storage [2]. Compared with a single application scenario, the shared At present, the research on the optimization control of shared energy

* Corresponding author.
E-mail address: jiarong0211@outlook.com (J. Li).

https://doi.org/10.1016/j.apenergy.2023.121801
Received 23 May 2023; Received in revised form 25 July 2023; Accepted 16 August 2023
Available online 23 August 2023
0306-2619/© 2023 Elsevier Ltd. All rights reserved.
X. Han et al. Applied Energy 350 (2023) 121801

storage systems in multiple scenarios is still in its infancy, in which response characteristics and supply-demand relationships of different
multiple scenarios mainly involve two or three scenarios among peak service entities in different scenarios, as well as the impact of seasonal
regulation, frequency regulation, energy market, power smoothing, and factors on capacity allocation and optimized operation, were not
renewable energy consumption. Kim W et al. [7] proposed an optimized considered.
scheduling strategy for shared energy storage systems based on reli­ The main purpose of this study is to propose a double-level optimi­
ability constraints, with the goal of minimizing the overall degradation zation method for shared energy storage systems based on dynamic
cost of energy storage batteries in peak regulation and energy market cooperative games, starting from the actual needs of various application
scenarios, but the profitability of energy storage systems was not scenarios in the power grid, and addressing the coupling relationship
considered; Celik et al. [8] proposed a coordinated optimization control between capacity configuration and operation strategies of energy
method for energy storage systems based on a F1PD-TI fuzzy controller storage systems. The capacity of the shared energy storage system is
with the goal of maximizing the stability of distributed photovoltaic optimized by the non-dominant sorting beluga whale optimization al­
systems in peak regulation and frequency regulation scenarios, but the gorithm (NSBWOA) in the upper level considering seasonal changes in
abandoned wind rate was not considered in the objective function; B. multiple scenarios, and the operation strategy under multiple scenarios
Cheng et al. [9] and Engels et al. [10] proposed coordinated control is optimized by the adaptive greedy search algorithm (AGSA) in the
strategies of energy storage systems in two scenarios, with the goal of lower level. With the goal of maximizing the gross annual total income
maximizing the benefits of energy storage systems, but the potential and high-value peak regulation ratio, and minimizing the cost- income
losses caused by aging of energy storage batteries were not considered in ratio, the optimal capacity configuration and operation strategy of the
the optimization control; Ippolito et al. [11] established an optimization energy storage system are obtained through interacting and iterating
scheduling model for shared energy storage systems in three application between the upper and lower level models. Through simulation analysis
scenarios of frequency regulation, power smoothing and abandoned of actual operating data, it can be seen that the method proposed in this
wind consumption, but these three scenarios operated independently paper ensures the frequency safety of the power grid, and has the ability
and lacked collaborative optimization between each scenario; C. Wang to obtain high returns in frequency regulation mode. The peak regula­
et al. [12] proposed a scheduling method of shared energy storage sys­ tion pressure under high-wind power penetration rate is alleviated by
tems based on optimal operating intervals, with the goal of maximizing consuming abandoned wind in the peak regulation mode, while the
the benefits of energy storage systems in both primary and secondary flexible electricity sales trading method in the energy market maximizes
frequency regulation scenarios, but the impact of seasonal changes on the economic benefits of the shared energy storage system.
the optimization scheduling results was not considered.
In the aforementioned literature, the capacity of energy storage 2. The relationship between energy storage demand and
systems is given. However, the larger the capacity of energy storage profitability of multiple stakeholders for power grid auxiliary
systems is, the better the effect of peak regulation and frequency regu­ services
lation is, the higher the initial investment cost of energy storage systems
is, and the worse the economic benefit of energy storage systems is. The scenarios of energy storage participating in grid auxiliary ser­
Therefore, the capacity configuration of energy storage systems is an vices mainly include peak regulation, frequency regulation and energy
important issue in the design and planning of renewable energy gener­ market. Fig. 1 shows the overall structure and operation mode of the
ation. At present, there is relatively little research on the capacity wind-thermal-shared energy storage system.
configuration of shared energy storage systems. Sardi et al. [13] estab­ In Fig. 1, the shared energy storage system assists thermal power
lished the capacity allocation model of shared energy storage systems in units in frequency regulation through rapid power response to reduce
peak regulation and energy market scenarios. However, in this model, their mechanical losses, while improving the utilization rate of renew­
the benefits and carbon reduction effects under different scenarios able energy by consuming abandoned wind power from wind farms
should be obtained through multi-objective optimization, rather than during low load periods, or selling electricity in the energy market to
simply being converted into a linear superposition of net present value; release capacity space and generate income. However, the optimal
Hafiz et al. [14] and J. Li et al. [15] established a capacity optimization operating strategy and capacity of energy storage systems vary in
configuration model for shared energy storage systems with the goal of different scenarios. In order to fully tap into their profit potential, it is
maximizing the benefits of energy storage systems. However, in the necessary to comprehensively consider the demand and benefit rela­
model, the capacity configuration of energy storage systems was only tionship between peak regulation, frequency regulation, and energy
implemented based on each season, without considering the changes in market.
different seasons throughout the year; Kargarian et al. [16] and C. Cao
et al. [17] established a capacity allocation model of shared energy 2.1. The operation cost analysis of the shared energy storage system in
storage systems, but the service life of the energy storage system was not multiple application scenarios
considered in the model; K. Zhu et al. [18] proposed a wind power
prediction method based on digital weather forecasting, which opti­ 2.1.1. Initial investment cost of energy storage systems
mized the capacity allocation of shared energy storage systems in wind The initial investment cost refers to the cost required for the con­
farm clusters under different sharing scenarios of economic and energy struction of energy storage systems, including the total expenses
storage resource. incurred in design, hardware, software, engineering, procurement,
In summary, in the process of capacity configuration of energy construction, etc. The initial investment cost is often a function that
storage systems, the operational strategy of energy storage systems is varies with the materials and market conditions. To comprehensively
optimized based on the configured energy storage capacity [19]. Due to consider the impact of energy storage service life on the revenue of the
the differential functions and benefits obtained by energy storage sys­ system, the total investment cost is converted into the annual equivalent
tems with different capacity levels by adopting different operational investment, and the calculation formula is shown in (1).
strategies, it is necessary to synergistically optimize the capacity ( )
CInv = Cp • PN + CE • EN • CRF
configuration and operational strategy of energy storage systems [20].
In [21–23], a two-layer optimization model of capacity configuration for r • (1 + r)LB
CRF =
shared energy storage systems and operation strategies of the system in (1 + r)LB − 1 (1)
two scenarios of peak regulation and energy market was established, in { }
Ncycle
which the coupling relationship between the capacity of energy storage LB = min , Ldesign
ncycle
systems and dispatching strategies was considered. However, the

2
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 1. The overall structure and operation mode of the wind-thermal-shared energy storage system.

where Cp is the investment cost per unit power; PN is the nominal power calculation formula is shown in (3).
capacity; CE is the investment cost per unit capacity; EN is the nominal CO&M = Cpo • PN + CEo • EN (3)
capacity of energy storage systems; CRF is the capital recovery coeffi­
cient; r is the discount rate; LB is the service life of energy storage sys­ where Cpo is the annual operation and maintenance cost per unit power
tems; Ldesign is the design calendar lifespan of the battery; Ncycle is the of energy storage systems; CEo is the annual operation and maintenance
actual number of cycles that energy storage systems operate annually; cost per unit capacity of energy storage systems.
ncycle is the total number of cycles of energy storage systems under
theoretical conditions. 2.1.3. Annual equivalent residual value of energy storage systems
According to [24], Ncycle is calculated by (2). The annual equivalent residual value refers to the residual value
recovered by the equipment when the energy storage system reaches its
Ncycle = β0 • DOD− β1
• eβ2 (1− DOD)
(2) service life. In order to comprehensively consider the impact of the
service life of the energy storage device on system revenue, it is con­
where β0 , β1 and β2 are the curve fitting coefficients determined based verted into an annual equivalent recovery rate, and the calculation
on the type of battery and experimental data provided by manufac­ formula is shown in (4).
turers; DOD is the rated discharge depth of the battery. ( )
CEol = σ • Cp • PN + CE • EN • SFF
2.1.2. Annual maintenance cost of energy storage systems r (4)
SFF =
Maintenance cost refers to the cost incurred during the annual LB
(1 + r) − 1
operation and maintenance of energy storage systems, mainly including
the operation and maintenance cost and the operation labor cost. where σ is the residual value rate of the equipment; SFF is the debt
Considering the discount rate and taking the time when the energy repayment factor.
storage system is put into operation as the conversion starting point, the

3
X. Han et al. Applied Energy 350 (2023) 121801

⃒ ⃒
2.2. Revenue analysis of energy storage systems in multiple application command and the output reaches the specified time, D=⃒Pref ⃒, otherwise
scenarios D = 0; BAGC is the compensation unit price for energy storage partici­
pating in the grid frequency regulation, $/MW; K is the evaluation in­
The overall revenue of energy storage participating in grid auxiliary dicator for the effectiveness of the energy storage system in tracking
services include the benefits of frequency regulation WFR , peak regula­ AGC commands, and the calculation formula is as follows:
tion WPR and energy market WEM .
vN
k1 = 1 − • Δt
Wgross = WFR + WPR + WEM (5) |PE − PS |
⃒ ⃒
⃒PE − Pref ⃒
2.2.1. Frequency regulation revenue WFR k2 = 1 − (7)
ΔPN
The output amplitude of the frequency regulation command changes
ΔPN
sharply, fluctuating up and down between high-power charging and k3 = 1 − • Δt
tst • |PE − PS |
high-power discharge [25]. Fig. 2 shows the AGC timing command
curve for energy storage participating in frequency regulation within a where PS is the output power of the battery upon receiving the AGC
typical day. command from the power grid, MW; PE is the effective threshold of the
Fig. 3 shows the frequency distribution of the power and duration for AGC command, MW; PE > 0 represents the discharge of the energy
AGC commands. storage system, and PE < 0 represents the charging of the energy storage
It can be seen from Fig. 3 that the duration of AGC commands for the system; vN = 1.5% • PN is the standard adjustment rate, MW/min; Δt is
frequency regualtion is in seconds (s), with short commands lasting tens the output power climing time; tst is the standard response time.
of seconds accounting for the majority. The distribution of charging and
discharging intervals is basically symmetrical. This is because the fre­ 2.2.2. Peak regulation revenue WPR
quency regulation command tends to be neutral over a period of time, The revenue from the peak regulation is compensated based on the
and the state of charge (SOC) of the energy storage system responsible peak and valley electricity prices according to the charging capacity of
for frequency regulation tasks does not fluctuate significantly. Taking the energy storage system. The time-of-use electricity price in the peak-
the charging process as an example, the frequency regulation process of valley refers to dividing the daily 24-h period into three periods: peak,
energy storage systems is shown in Fig. 4. flat and valley. The division of the time of use electricity prices in North
It can be seen from Fig. 4 that the initial output of the energy storage China is shown in (8).
system is PS . After the AGC command is issued, the energy storage sys­ ⎧
tem begins to charge or discharge at a uniform speed, the output of the ⎨ Peak period 8 : 00 − 12 : 00, 17 : 00 − 21 : 00 132.3 $/MWh
energy storage system increases to PE after Δt and remain unchanged Flat period 12 : 00 − 17 : 00, 21 : 00 − 24 : 00 88.2$/MWh (8)

until the next AGC command is issued. Take the deviation range (Pref − Valley period 0 : 00 − 8 : 00 36.8$/MWh
ΔPN ,Pref + ΔPN ) between the actual output and the AGC frequency The revenue of the peak regulation is calculated by (9).
regulation command as the assessment range of the frequency regula­
tion, namely PE ∈(Pref − ΔPN ,Pref + ΔPN ), in which ΔPN is the WPR = Bpr • Ppr • t (9)
maximum allowable output deviation during the frequency regulation,
where Bpr is the peak-valley compensation unit price for the energy
Pref is the target output of the frequency regulation.
storage system participating in peak regulation, $/MWh; t is the output
The revenue of the frequency regulation is calculated by (6).
time, h.
WFR = D • k1 • k2 • k3 • BAGC (6) Unlike the AGC command of the frequency regulation, the AGC
command of the peak regulation has strong seasonality. Taking the
where D is the compensation depth, when the compensation depth of the actual operating data of a 200 MW wind turbine in North China as an
energy storage system is within the assessment range of the AGC

Fig. 2. The AGC timing command curve for energy storage participating in frequency regulation within a typical day.

4
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 3. The frequency distribution histogram of the power and duration for AGC commands.

The revenue of the frequency regulation and peak regulation within


a typical day in different seasons is calculated according to (6)∼(9), and
the number of periods peak regulation revenue and frequency regulation
revenue are dominant in the four seasons separately, as shown in Fig. 6.
It can be seen from Fig. 6 that there are significant differences in the
revenue dominance of frequency regulation and peak regulation ser­
vices in different seasons throughout the year. In spring, the revenue of
the peak regulation and frequency regulation accounts for 41.7% and
58.3%, respectively; In summer and autumn, the revenue of the peak
regulation only accounts for 12.5% and 25%; In winter, the revenue of
the frequency regulation reaches 100%. The revenue curve of the peak
regulation and frequency regulation on a typical day in spring are shown
in Fig. 7.
It can be seen from Fig. 7 that in the valley period, the revenue ob­
tained by the frequency regulation is significantly higher than that ob­
tained by the peak regulation; In peak and flat periods, under the dual
influence of fluctuations in peak regulation command and differences in
peak-valley electricity prices, there is an asymmetric trend in the reve­
nue of the peak regulation and frequency regulation. From 8:00 to
Fig. 4. The working process of energy storage participating in fre­
14:30, the revenue of the peak regulation is greater than the revenue of
quency regulation.
the frequency regulation; From 14:30 to 18:45, the revenue of the fre­
quency regulation is greater than the revenue of the peak regulation.
Therefore, in order to achieve the highest revenue within each time
example (with a sampling interval of 15 min), the curve of the AGC period, the shared energy storage system needs to adaptively formulate
command for the peak regulation within a typical day in different sea­ the corresponding optimal scheduling strategy based on the specific
sons is shown in Fig. 5. AGC commands of the peak regulation and frequency regulation every
It can be seen from Fig. 5 that there are significant differences in the day.
characteristics of the AGC command for the peak regulation in different
seasons throughout the year. In spring, the overall peak regulation 2.2.3. Energy market revenue WEM
power fluctuates from 5 MW to 18 MW, but the power is low during The energy market can reserve a certain adjustable reserve capacity
16:00 and 20:00. During summer, the peak regulation power is rela­ for energy storage systems to cope with risks, while the energy market
tively low for most of the day, but the diurnal variation is significant, gains profits through electricity trading, improving the utilization rate of
with a significant increase in the peak regulation demand during 17:00 shared energy storage systems. Taking the revenue curve of the fre­
and 21:00. In autumn, the peak regulation power fluctuates the most quency regulation and peak regulation shown in Fig. 7 in spring as an
throughout the day, ranging from 3 MW to 20 MW overall, and the example, the state of charge (SOC) curve of the energy storage system
power is low during 7:00 and 17:00. In winter, the peak regulation before and after participating in energy market services is shown in
power throughout the day is relatively low, but relatively stable, fluc­ Fig. 8.
tuating from 4 MW to 8 MW. It can be seen from Fig. 8 that when the energy storage system does

5
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 5. The curve of the AGC command for the peak regulation within a typical day in different seasons.

Fig. 6. The revenue dominance of frequency regulation and peak regulation services.

not participate in energy market services, the SOC of the energy storage
WEM = Bem • Pem • t (10)
system fluctuates around 0.5 between 0:00 and 8:00, and the revenue of
frequency regulation dominates; From 8:00 to 9:45, the SOC of the en­
where Bem is the electricity price of the energy storage system partici­
ergy storage system reached 0.9 through rapid charging, and the reve­
pating in the energy market, $/MWh.
nue of peak regulation dominates. If the energy storage system continues
to consume the abandoned wind, it will lead to overcharging of the
3. A dynamic game collaborative optimization model for
energy storage system. Therefore, the energy storage system is forced to
sharing energy storage in multiple application scenarios
perform frequency regulation services at lower incomes. From 0:30 to
1:15, the energy storage system discharges 3 times participating in the
3.1. Dual layer optimization control structure of shared energy storage in
energy market, and the SOC is adjusted to 0.18 to leave sufficient ca­
multiple application scenarios based on dynamic cooperative game theory
pacity space for peak regulation services; During the period from 12:00
to 14:30, the energy storage system discharges 4 times participating in
The game theory refers to the discipline that studies how multiple
the energy market to maintain the soc. within the range of (0.62, 0.83) to
individuals or teams use the strategies of related parties in the game
avoid overcharging. Therefore, the participation of the energy storage
under specific conditions to implement corresponding strategies [26].
system in the energy market not only enhances the ability to consume
Game modeling theory can be divided into non-cooperative games and
abandoned wind power, but also generates certain benefits through the
cooperative games. Compared to non-cooperative games, the overall
sale of electricity in the energy market, maximizing the benefits of the
benefits of cooperative alliances are generally greater than the sum of
energy storage system.
individual non-cooperative benefits. Therefore, the above three sce­
The revenue of the energy market is calculated by (10).
narios share the same energy storage equipment, with the goal of

6
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 7. The revenue curve of the frequency regulation and peak regulation for a typical day in spring.

Fig. 8. The SOC curve of the energy storage system before and after participating in energy market services.

maximizing the annual total revenue and high-value peak regulation year is used as the decision set, which is sent back to the upper level. The
ratio, and minimizing the annual total cost rate of the energy storage capacity configuration of the shared energy storage system is the basis of
system, a dual-level optimization scheduling model for the shared en­ the daily optimal scheduling, and the economic benefits obtained by the
ergy storage system based on dynamic cooperative games is established. daily optimal scheduling determine the investment cost of the shared
By coordinating the capacity configuration of the energy storage system energy storage system, thus affecting the decision-making of the ca­
with the optimal scheduling strategy, the comprehensive economic pacity configuration. The optimal capacity configuration and scheduling
benefits of the energy storage system in various application scenarios are schemes of the energy storage system are determined by dynamic
maximized, as shown in Fig. 9. cooperative games between the upper and lower layers.
It can be seen from Fig. 9 that in the upper level model, the investor
of the energy storage system is taken as the leader, the power and ca­
3.2. Decision sets
pacity of the energy storage system are regarded as the decision set,
which is sent to the lower level; In the lower level model, the three
3.2.1. Leader strategy
auxiliary service scenarios of frequency regulation (A), peak regulation
Take the energy storage builder as the upper level decision-maker,
(B), and energy market (C) are regarded as three game participants (A,
known as the leader (Z). The decision values of energy storage in­
B, C). Each participant forms an alliance through cooperation, and the
vestors are the rated power PN and rated capacity EN of the energy
optimal daily scheduling plan of the energy storage system within the
storage system, and the strategy set is recorded as:

7
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 9. The dual-layer optimization control fdiagram for shared energy storage in multiple application scenarios based on dynamic cooperative games.

SLeader = SZ = {PN, EN} (11) { }


SdA = Sd,1
A A
, Sd,2 A
, …, Sd,96
{ }
where PN is the optimal rated power of the energy storage system; EN is SdB = Sd,1
B B
, Sd,2 B
, …, Sd,96 (13)
the optimal rated capacity of the energy storage system, PN, EN ≥ 0. { }
SdC = Sd,1
C C
, Sd,2 C
, …, Sd,96
3.2.2. Follower strategy
Take the three auxiliary service scenarios of frequency regulation where SAd,i ,SBd,i ,SCd,i respectively represent the strategy sets of each follower
(A), peak regulation (B), and energy market (C) as three game partici­
in the ith period of the dth day; SAd 、SBd 、SCd are the complete strategy
pants (A, B, C). Due to the fact that the duration of the frequency fr
regulation is in seconds and the duration of the peak regulation is in sets for each follower on the dth day, respectively; Pd,i,j represents the
hours (as shown in Fig. 3), according to [10], each day is discretised into response output power of the jth AGC frequency regulation command in
time steps of 15 min to meet the requirements of the peak regulation and the ith period on the dth day; Ppr
d,i represents the response output power of
frequency regulation as much as possible. During each time period, the the peak regulation command in the ith period of the dth day; Pem
d,i rep­
energy storage system can freely choose one auxiliary service to perform resents the output power of electricity sold in the energy market during
based on revenue, SOC, and capacity space, but cannot freely switch to the ith period of the dth day.
other auxiliary services. The output power of the energy storage system For a cooperative game involving three participants, which can be
in each period is used as a follower strategy in three scenarios, and the arranged and combined to form up to seven alliances, namely u1 =
strategy set is recorded as: {A},u2 = {B},u3 = {C},u4 = {A, B},u5 = {A, C},u6 = {B, C} and u7 =
{( )} {A, B, C}. Assume that the alliance chosen by the lower level partici­
A
Sd,i = Pfrd,i,1 , Pfrd,i,2 …, Pfrd,i,n
{ } pants of the energy storage system on the dth day is Ld , the complete
B
Sd,i = Ppr d,i (12) decision of the followers on a natural day is:
{ } { }
C
Sd,i = Pd,i em SdFollower = SdA , SdB , SdC , Ld (14)

8
X. Han et al. Applied Energy 350 (2023) 121801

The complete decision of the participants throughout the year is:


⎧ PN ≥0
{ } ⎪
S Follower
= S1Follower , S2Follower , …, S365
Follower
(15) ⎪



⎪ EN ≥0





⎪ 0.5•PN ≤EN ≤4•PN

3.3. Objective functions ⎪



⎨ n ⃒ ⃒ ∑ D ∑
I ⃒ ⃒ ∑ D ∑
I

D ∑
I ∑
⃒ΔSOCj (i) ⃒ + ⃒ΔSOCpr (i) ⃒ +
s.t. |ΔSOCem (i)|
The first objective function f1 is to maximize the gross annual in­
fr

⎪ d=1 i=1 j=1 d=1 i=1 d=1 i=1
⎪ ncycle =
come. It can be written as (16). ⎪


⎪ 2



⎪ Ncycle
D ∑
∑ I ∑
n D ∑
∑ I ⎪
⎪ LB =
Maximize f1 = D • k1 • k2 • k3 • Pfrd,i,j + Bpr • Ppr ⎪

d,i ⎪
⎪ ncycle
d=1 i=1 j=1 d=1 i=1 ⎩
LB ≤Ldesign
D ∑
∑ I
•t+ Bem • Pem
d,i • t (16) (19)
d=1 i=1

The net profit of the energy storage system is directly proportional to 3.4.2. Lower-level constraints
the cost of the energy storage system, and the service life of the energy The constraints of the lower level include the output constraints of
storage system is closely related to the scheduling strategy. Therefore, it frequency regulation, peak regulation and energy market, the effective
is unscientific to simply pursue the maximum net profit of the energy threshold range constraint of the AGC frequency regulation command
storage system to achieve optimal economic performance.The increase PE , and Soc constraint of the energy storage system. In actual scheduling,
in net profit of energy storage systems is often accompanied by a huge due to the significant differences in the time scales of frequency regu­
increase in costs. Also, the service life of energy storage system is closely lation commands, peak regulation commands, and energy market elec­
related to the annual scheduling strategy of energy storage systems. tricity sales, 15 min are selected as a scheduling period, with a total of 96
Therefore, it is not scientific to achieve economic optimization by simply scheduling periods throughout a day.
pursuing the absolute maximum net profit of energy storage. 1) The output constraints of frequency regulation, peak regulation
The cost-income ratio is the percentage of the total cost expense to and energy market.
the total operating income, which can fully reflect the ability of all costs The three types of auxiliary services are not compatible with each
and expenses incurred in the current period to bring benefits. The second other during the same scheduling period. The energy storage system
objective function f2 is to minimize the annual cost-income ratio of the only executes one output strategy during each period, and the output
energy storage system. It can be written as (17). power cannot exceed the maximum charging and discharging power

( ) ( ) ( )
CInv SLeader , SFollower + CO&M SLeader − CEol SLeader , SFollower
Minimize f2 = (17)

D ∑
I ∑
n ∑∑
D I ∑∑
D I
D • k1 • k2 • k3 • BAGC + Bpr • Ppr
d,i • t + Bem • Pem
d,i • t
d=1 i=1 j=1 d=1 i=1 d=1 i=1

where CInv and CEol respectively represent the annualized initial invest­ limit of the energy storage system. The constraint conditions are shown
ment cost and recovery residual value of energy storage systems, they in (20) and (21).
are both functions of the upper and lower decision sets. ⎧
The comprehensive economic benefits of the energy storage system ⎪ min j max
⎨ Pb ≤ Pfr (i) ≤ Pb

are not limited to its own cost and income. The energy storage system Pb ≤ Ppr (i) ≤ Pmax
min (20)
⎪ b
participating in the peak regulation service can well alleviate the peak ⎪
⎩ Pmin ≤ P (i) ≤ Pmax
em
regulation pressure of the grid under high permeability of wind power,
b b

and reduce the equipment loss cost of traditional thermal power units ⎧
due to the energy storage system participating in the peak regulation. ⎪
⎪ j
⎨ Pfr (i) • Ppr (i) = 0

The third objective function is to maximize the high-value peak j
Pfr (i) • Pem (i) = 0 (21)
regulation ratio which measures how much high-value abandoned wind ⎪


⎩ Ppr (i) • Pem (i) = 0
power is absorbed by the shared energy storage system f3 . It can be
written as (18).
where Pmin
b and Pmax
b represent the lower and upper limits of the oper­
∑ ating power for the energy storage system, respectively; i is the ith
D ∑
I
Ppr
d,i •t
j
Maximize f3 = d=1 i=1
(18) scheduling period of the day, i = 1,2…, 96; Pfr (i) is the output power of

D ∑I
Pref pr
d,i •t the jth frequency regulations command in the ith scheduling period,
with a total of n instructions, j = 1,2…, n; Ppr (i) and Pem (i) are the output
d=1 i=1

where Pref pr power of the peak regulation and energy market during the ith sched­
d,i is the reference power of the abandoned wind in the ith
uling period, respectively.
period on the dth day, MW.
2) The effective threshold range constraints of the AGC frequency
regulation command PE.
3.4. Constraint conditions To ensure the accuracy of the frequency regulation, the output power
of the energy storage system in response to each frequency regulation
3.4.1. Upper-level constraints command must reach the specified effective threshold range, while
The constraints of the upper level include the power capacity con­ maintaining it for at least 2 s. The constraint condition is shown in (22).
straints and the lifespan constraints of the energy storage system. It can
be given in (19).

9
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 10. The interactive and iterative solution processe based on NSBWOA and AGSA.

10
X. Han et al. Applied Energy 350 (2023) 121801

0 ≤ Δt ≤ T − 2 adaptive and play an important role in controlling exploration and


(22)
Pref − ΔPN ≤ PE ≤ Pref + ΔPN development capabilities. This algorithm utilizes the Levy flight func­
tion to enhance the convergence ability during the algorithm develop­
3) SOC constraints of the energy storage system
ment phase and has the whale fall phase to get out of local optimum. In
In the operation process of the energy storage system, the SOC at
addition, the BWOA only has two initial parameters: the number of it­
each instantaneous moment shall not exceed the specified maximum
erations and the number of population, which does not depend on
and minimum values of the SOC. The constraint condition is shown in
parameter debugging and therefore is quite suitable for the joint nested
(23).
solution proposd in this paper.
SOCmin ≤ SOC(∀i) ≤ SOCmax (23) Multi objective optimization problems typically require solving
various conflicting objectives, some of which may consist of maximizing
The update formula of the SOC at the next moment is shown in (24)
functions while others consist of minimizing functions [29]. To solve the
and (25).
multi-objective optimization problem, a naturally inspired hybrid al­
SOC(i + 1) = SOC(i) + ΔSOCfr (i) + ΔSOCpr (i) + ΔSOCem (i) (24) gorithm NSBWOA is developed based on the non-dominated sorting
optimization technology [30]. The algorithm implementation process of

⎪ Pjs (i) + Pjfr (i) ( ) the NSBWOA is as follows:




⎪ 2
• Δt + Pjfr (i) • T j (i) − Δt Step 1: Initialize algorithm parameters, generate initialization pop­

⎨η •
⎪ Pjfr (i) ≥ 0 ulation and calculate fitness;
j
EN
ΔSOCfr (i) =
⎪ Step 2: Perform non-dominated sorting on the initial population and
⎪ Pjs (i) + Pjfr (i) ( )


⎪ • Δt + Pjfr (i) • T j (i) − Δt calculate the crowding distance and Pareto sorting of the initial




2 Pjfr (i) < 0 population;
η • EN
Step 3: Start iteration and calculate the balance factor Bf and the
n [
∑ ] probability of whale landing Wf based on the number of iterations.
SOCjfr (i)
j=1
ΔSOCfr (i) = Bf = B0 • (1 − T/2T
/ max )
EN (26)
Wf = 0.1 − 0.05T Tmax
η• Ppr
d,i •t
ΔSOCpr (i) =
EN where T is the current iteration; Tmax is the maximum number of itera­
Pem •t tions; B0 randomly changes between 0 and 1 during each iteration.
Step 4: Update the location of each beluga whale, where each po­
d,i
ΔSOCem (i) =
η • EN
sition can be regarded as a set of leader strategy{ PN, EN }. This process
(25)
includs three stages: exploration, development, and whale landing. The
where SOC(i) is the SOC of the energy storage system at the end of the ith mathematical descriptions are as follows:
j
scheduling period; SOCfr (i) is the SOC of the energy storage system
① Exploration phase
when the jth frequency regulation command during the ith scheduling ⎧ ( )
period is completed;ΔSOCfr (i), ΔSOCpr (i), ΔSOCem (i) respectively repre­ ⎪
⎨ Xi,jT+1 = Xi,p
T T
+ Xr,p T
− Xi,p • (1 + r1 ) • sin(2π r2 ), j = even
sent the changes in the SOC generated by the three auxiliary services of (27)
j 1 j
( )

frequency regulation, peak regulation and energy market during the ith
T+1 T T T
⎩ Xi,j = Xi,p + Xr,p − Xi,p • (1 + r1 ) • cos(2πr2 ), j = odd
j 1 j

scheduling period; SOCmin , SOCmax represent the lower and upper limits
of the SOC of the energy storage system; EN is the capacity of the energy T
where T is the current number of iterations; Xi,p is the new position of
j
storage system.
the ith beluga whale in the jth dimension; pj is a random number
T
randomly selected from the d-dimension; Xi,p is the current position of
4. Model solving j
T
the ith beluga whale at the p1 th dimension; Xr,p 1
is the current position of
A joint nested scheme of the NSBWOA and AGSA is adopted to solve the rth beluga whale at the p1 th dimension; r1 and r2 are a random
the dual-level dynamic game optimization model. In the upper level number between (0,1). The updated position reflects the synchronized
model, the NSBWOA is used to determine the initial capacity of the or mirrored behavior of beluga whales during swimming or diving based
shared energy storage system which is input to the optimization on the dimensions of odd and even numbers. Two random numbers r1
scheduling model in the lower level, and the scheduling result obtained and r2 are two random operators used to enhance the exploration order.
by the AGSA is returned to the upper level model to further optimize the
capacity of the energy storage system. The upper and lower levels ② Development phase
continuously interact and iterate until the maximum number of itera­ XiT+1 = r3 • Xbest
T
(
− r4 • XiT + C1 • LF • XrT − XiT
)
(28)
tions is reached, the multi attribute boundary approximation region
comparison (MABAC) method [27] is used to evaluate the obtained T
where Xbest is the best position among all white whales; r3 and r4 are a
pareto solution set, and finally determine the optimal capacity config­
random number between (0,1); C1 = 2r4 • (1 − T/Tmax ) is the random
uration and the corresponding optimal scheduling strategy. The specific
jump intensity that measures Levy flight intensity.
interactive iteration process is shown in Fig. 10.
The calculation formula of the flight function for Levy LF is as fol­
lows:
4.1. The solution of the upper level model based on the NSBWOA μ•ε
LF = 0.05 •
|v|1/δ
The beluga whale optimization algorithm (BWOA) was proposed by ( )1/δ (29)
Z. Chang et al. [28] in 2022, which is a novel meta heuristic algorithm Γ • (1 + δ) • sin(π • δ/2)
ε=
based on population, inspired by the behavior of beluga whales. The Γ • ((1 + δ)/2 ) • δ • 2(1+δ)/2
BWOA includes three stages: exploration, development, and whale fall,
corresponding to paired swimming, predation, and whale fall behaviors. where μ and v are random numbers of the normal distribution; Γ is a
The balance factor of the BWOA and the probability of whale fall are

11
X. Han et al. Applied Energy 350 (2023) 121801

gamma function; δ is the default constant, δ=1.5. optimal solutions of all sub-processes to generate the final global
optimal solution. The AGSA can be divided into three parts: initializa­
③ Whale fall tion, destruction, and reconstruction. The specific process is as follows:

XiT+1 = r5 • XiT − r6 • XrT + r7 • Xstep (30)


4.2.1. Initialization
Step 1: Calculate the revenue of the frequency regulation and the
where r5 , r6 and r7 are a random number between (0,1); Xstep is the step
peak regulation during 96 time periods throughout the day according to
size of a white fall, it can be calculated by (31).
(6)∼(9) (as shown in Fig. 11), and determine the auxiliary service types
Xstep = (ub − lb ) • exp( − C2 • T/Tmax ) (31) corresponding to high revenue.
It can be seen from Fig. 11 that when the revenue of the peak
where C2 is the step size factor, which is related to the probability of regulation is higher than that of the frequency regulation, it is defined as
whale falling and population size C2 = Wf • n; ub and lb are the upper the peak period of the peak regulation, denoted as Tpr ; Similarly, when
and lower boundaries of the variable, respectively. It can be seen that the revenue of the frequency regulation is higher than that of the peak
the step size is jointly influenced by the variable boundary, the number regulation, it is defined as the peak period of the frequency regulation,
of iterations, and the maximum number of iterations. denoted as Tfr .
Step 5: Integrate the parent population with the newly generated Step 2: Here, an auxiliary variable “virtual SOC” is defined, mainly
offspring population; used to calculate the capacity space required for the energy storage
( )
Step 6: Calculate the three objective values f1 , f2 , f3 of the fusion system to perform the high-value peak regualtion. The virtual SOC
population through the AGSA and conduct non-dominated sorting on all (denoted as SOCvirtual ) is different from the actual SOC and is not limited
individuals and use the elite strategy to obtain the new generation of by the range of (0,1). Take the high-value auxiliary services corre­
population; sponding to Tpr and Tfr in Step 1 as the virtual scheduling strategy, the
Step 7: When the number of public iterations is limited, the algo­ value of SOCvirtual is calculated by (32).
rithm ends. Otherwise, proceed to Step 3.
SOCvirtual (i + 1) = SOCvirtual (i) + ΔSOCfr (i) + ΔSOCpr (i) (32)

The corresponding curve of the virtual SOC is shown in Fig. 12.


4.2. The solution of the lower level model based on the AGSA It can be seen from Fig. 12 that according to the virtual scheduling
strategy, the energy storage system has been in a charging state during
Due to the large number of optimization strategy combinations in the the peak regulation periods from 8:00 to 14:30, and the SOCvirtual of the
lower level model, traditional optimization algorithms are difficult to energy storage system around 11:30 exceeds the SOCmax (0.9). The en­
meet the requirements of model solving. The adaptive greedy search ergy corresponding to the portion exceeding the SOCmax is regarded as
algorithm adopts a top-down iterative method for greedy search, the amount of the electricity should be sold when participating in the
continuously approaching a given target, and quickly converging, which energy market, it can be calculated by (33).
is suitable for the optimization operation requirements of the shared ⎧( ( ) ) ( )〉
energy storage system in multiple application scenarios. For the problem ⎪
⎨ SOCvirtual Tprmax
− SOCmax • EN SOCvirtual Tpr max
SOCmax
to be solved, the process of processing the problem is divided into E= ( ) (33)

multiple sub-processes. By gradually establishing the optimal solution, ⎩ max
0 SOCvirtual Tpr ≤ SOCmax
corresponding greedy strategies are determined in different sub-
processes, and adaptive adjustments are made during the search pro­ max
where Tpr is the final moment of the last peak regulation period of the
cess to make optimal decisions for the sub-processes and synthesize the

Fig. 11. The revenue curve of the frequency regulation and the peak regulation throughout the day.

12
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 12. The corresponding curve of the virtual SOC.

day; SOCmax is the upper limit of the SOC for the energy storage system. for participating in energy market electricity sales, and obtain the initial
Step 3: Calculate the number of the periods N required to participate solution Ω.
in the energy market to sell electricity and complete the reserved
discharge capacity. 4.2.2. Destruction
Step 1: Based on the initial solution Ω, the SOC optimal operating
E
N=⌈ ⌉ (34) curve of the energy storage system during the destruction phase ob­
Pem • t
tained through N iterations is shown in Fig. 13.
Step 4: Rank the income values of each time period obtained in Step It can be seen from Fig. 13 that during the period from 1:15 to 8:00, if
1 in ascending order, taking the first N time periods as the time periods

Fig. 13. The SOC optimal operating curve of the energy storage system during the destruction phase obtained through N iterations.

13
X. Han et al. Applied Energy 350 (2023) 121801

the energy storage system continues to discharge in the electricity sales the optimization interval of the discharge points continues to expand
period to participate in the energy market, it will cause the SOC to be until the number of discharges reaches N′, covering the entire optimi­
lower than SOCmin . Under the constraint of the SOC, the electricity sales zation range of the reconstruction stage and ensuring the optimality of
fail
to participate in the energy market fail, recorded as tem . During the the model solution to the greatest extent possible.
period from 11:15 to 14:30, if the energy storage system continues to Step 4: The optimization scheduling results Ωbest obtained by the
charge in the peak regulation period, it will cause the SOC to exceed AGSA in the lower level model is output to the upper level model to
SOCmax . Under the constraint of the SOC, the peak regulation fails, achieve the optimal capacity configuration of the shared energy storage
fail power station.
recorded as tpr .
Step 2: Perform a destructive operation, and ΩD is obtained by
5. Results and discussion
removing the electricity sales periods participating in the energy market
fail
tem from Ω.
5.1. Simulation description

4.2.3. Reconstruction
The actual operation data are from wind farms with an installed
Step 1: On the basis of the destruction stage, the optimization scope
( [ ] [ ]) capacity of 300 MW in North China and thermal power plants with an
fail fail
of the reconstruction stage is further reduced to max tem , max tpr . installed capacity of 200 MW, and the sampling interval is 15 min.
Unlike the frequency regulation service, during the reconstruction Lithium iron phosphate battery is selected as the energy storage system,
phase, a new round of discharge is selected during the peak regulation the minimum charging time is 0.5 h, and the maximum charging time is
period when the SOC gradually increases. Therefore, the number of the 4 h. The parameter settings of three auxiliary services, Lithium iron
discharge periods N′ during the reconstruction phase is calculated by phosphate battery and optimization algorithms are shown in Table A.1,
(35). Table A.2 and Table A.3 respectively. The three application scenarios are
divided as follows:
∑ pr
Pd,tfail − N′ • Ppr fail Scenario I: Energy storage system only participates in frequency
(35)
d,tpr
regulation auxiliary service;
′ pr
N =
Pem
Scenario II: Energy storage system participates in peak regulation
Step 2: Before the SOC of the energy storage system exceeds SOCmax , and energy market auxiliary services;
the search interval for each iteration needs to be adaptively adjusted Scenario III: Energy storage system participates frequency regula­
(known as the adaptive optimization interval τR ) and find the optimal tion, peak regulation and energy market auxiliary services.
discharge time point within the adaptive optimization interval. The
adaptive optimization interval τR is calculated by (36).
( [ fail ] [ ]) 5.2. Capacity configuration of the shared energy storage system obtained
τR = max tem fail
, min tpr (36) by different optimization algrithums in different scenarios

Select the period with the lowest return within τR as the new For comparative analysis, the capacity of the energy storage system
discharge point to generate Ω′, calculate the SOC of the energy storage in different scenarios is respectively configured by the NSBWOA, NSGA-
system. If the SOC value meets the constraint limit of the SOC, Ωbest is II (Non-dominated Sorting Genetic Algorithm-II) and MOPSO (Multi-
updated by Ω′, otherwise replace Ω with Ω′ to search for the optimal Objective Particle Swarm Optimization) [31]. The comparison results of
discharge point. After each iteration, the optimization interval τR is evaluation indicators and capacity configuration under three optimiza­
adaptively corrected. The SOC curve and the adaptive optimization in­ tion algorithms are shown in Table 1.
terval division of the virtual SOC during reconstruction iteration process It can be seen from Table 1 that among the three algorithms, the
are shown in Fig. 14. NSBWOA has the least optimization parameters (only 2) and the
It can be seen from Fig. 14 that as the number of iterations increases, calculation time is the shortest compared to NSGA-II and MOPSO. Not
only the cast-income ratio is lower, but also the gross annual income and

Fig. 14. The SOC curve and adaptive optimization interval of the energy storage system during reconstruction iteration process.

14
X. Han et al. Applied Energy 350 (2023) 121801

Table 1
The comparison results of evaluation indicators and capacity configuration under three optimization algorithms.
Algorithms NSBWOA NSGA-II MOPSO

Number of parameters to be set 2 4 6


Scenario I Calculation Time 3 min 25 s 5 min 05 s 10 min 39 s
Capacity configuration 30MW/36MW • h 30MW/41MW • h 31MW/30MW • h
Cost-income ratio 0.3288 0.3328 0.3387
Gross annual income $10.3756 Million $10.3749 Million $10.3735 Million
Scenario II Calculation Time 10 min 23 s 15 min 21 s 32 min 12 s
Capacity configuration 50MW/152MW • h 48MW/163MW • h 46MW/170MW • h
Cost-income ratio 0.5502 0.5610 0.5701
Gross annual income $9.9373 Million $9.9117 Million $9.8676 Million
High-value PR ratio 0.8060 0.8010 0.7953
Scenario III Calculation Time 36 min 19 s 50 min 06 s 110 min 08 s
Capacity configuration 49MW/101MW • h 46MW/110MW • h 47MW/116MW • h
Cost-income ratio 0.4051 0.4067 0.4136
Gross annual income $12.6865 Million $12.6317 Million $12.6618 Million
High-value PR ratio 0.9395 0.9286 0.9278

Fig. 15. The Pareto solution set and corresponding evaluation results in Scenario I.

Fig. 16. The Pareto solution set and corresponding evaluation results in Scenario II.

15
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 17. The Pareto solution set and corresponding evaluation results in Scenario III.

high-value PR ratio obtained by the NSBWOA are higher than those multiple application scenarios, the configured capacity is 49 MW/101
obtained by the NSGA-II and MOPSO. Therefore, the NSBWOA is used to MW•h, located between Scenario I and Scenario II. It can not only
obtain the optimal Pareto solution set of the capacity configuration in achieve high returns while ensuring the frequency safety of the power
the upper model. grid, but also alleviate the pressure of the peak regulation under high
The Pareto solution set of the upper model obtained by the NSBWOA wind penetration rate. By participating in energy market electricity sales
and its corresponding evaluation results in different scenarios are transactions, the economic benefit of the energy storage system is
respectively shown in Figs. 15–17. maximized. Compared with Scenario I, the cost-income ratio increased
For Scenario I, it can be seen from Fig. 15 that the cost-income ratio by 7.63%, but the gross annual income increased by 2.31 million dollars.
has an approximate proportional relationship with the gross annual Compared with Scenario II, the cost-income ratio decreased by 14.51%,
income within the range of (0.32, 0.35) and (0.37, 0.40). The cost the gross annual income increased by 2.749 million dollars, and most
expense ratio increases with the increase of the annual total income. importantly, the high value peak regulation ratio increased by 13.35%.
Because of the inherent characteristics of the frequency regulation ser­ Among the three scenarios, the comprehensive economic efficiency of
vices that do not cause significant fluctuations in the SOC of the energy Scenario III is the best.
storage system, a small capacity configuration of 30 MW/36 MW•h can
meet the requirements and therefore its initial investment cost is the 5.3. Optimized operation of the shared energy storage system in different
lowest. Due to the high unit profit of frequency regulation services, the scenarios
gross annual income is 10.376 million dollars, and the small capacity
results in a cost- income ratio of only 0.3288, which is the lowest among The annual operating curve and typtical daily operating curve of the
the three scenarios and has good economic efficiency. However, the SOC obtained by the AGSA in different scenarios are respectively shown
capacity of the energy storage system configured by Scenario I is too in Fig. 18~Fig. 20.
small to have the potential to participate in peak regulation services and For Scenario I, it can be seen From Fig. 18(a) that since the energy
cannot alleviate the pressure of the abandoned wind power in the new storage system only performs frequency regulation services, 97% of the
power system. SOC fall within the range of 0.7 to 0.9 during the whole year, and do not
For Scenario II, it can be seen from Fig. 16 that as the high-value change with seasons. And from Fig. 18(b) that The operating curve of the
peak regulation ratio increased from 0.5914 to 0.8113, and the cost- SOC on a typical daily is relatively stable due to the inherent charac­
income ratio increased from 0.4026 to 0.6405. At the same time, the teristics of the frequency regulation services, and the fluctuation range
gross annual income spiraled up with the increase of the cost-income of the SOC is within ±0.12.
ratio and the high-value peak regulation ratio, increased from 7.479 For Scenario II, it can be seen from Fig. 19 that the SOC operation
million dollars to 10.032 million dollars. Due to the insufficient profit­ curve of the energy storage system has obvious seasonality. Based on the
ability of the peak regulation services during the valley and flat periods characteristics of different seasons and the peak-valley electricity prices
of electricity prices compared to frequency regulation, the gross annual of the typical day, the energy storage system make the optimal decision
income of the energy storage system is only 9.937 million dollars. using the AGSA in advance to reserve capacity space for peak regulation
However, the configured capacity is 50 MW/152 MW•h, which is the services. And because of the long-term one-way charging required for
largest among the three scenarios. Therefore, the initial investment cost peak regulation services, when the energy storage system participates in
is relatively high, and the cost- income ratio reached 0.5502, which is peak regulation and energy market auxiliary services, the typical daily
the highest among the three scenarios and the economy efficiency is the operating curves of the SOC in four seasons all showed significant
worst. fluctuations, frequently approaching the maximum(0.9) and minimum
For Scenario III, it can be seen from Fig. 17 that as the high-value (0.1) limit values of the SOC.
peak regulation ratio increased from 0.8438 to 0.9395, the cost- For Scenario III, it can be seen from Fig. 20(a) that the alliances
income ratio increased from 0.3400 to 0.4144. At the same time, the between different application scenarios are closely related to seasonal
gross annual income spiraled up with the increase of the cost-income changes. During December and January, the shared energy storage
ratio and the high-value peak regulation ratio, increased from 12 system mainly adopts an alternating operation mode between frequency
million dollars to 12.724 million dollars. But it is much more tortuous regulation (A) and frequency regulation-energy market (AC), which is to
compared to Scenario II. Through a cooperative alliance between execute alliance A for 2–5 days and AC alliance for one day during this

16
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 18. The annual and typtical daily operating curve of the SOC in Scenario I

period. In April, due to the extra large amount of abandoned wind power the energy storage system consumed as much abandoned wind power as
in spring, the energy storage system chose to execute BC alliance for 3 possible.
days. The rest of the year adopts the operation mode of ABC alliance. In winter, due to the small amount of wind power generation in
The detailed analysis of different seasons is shown in Fig. 20(b). winter, the amount of abandoned-wind power is the lowest throughout
In spring, the amount of the abandoned-wind in spring is the highest the year, and the benefit of the peak regulation is much lower than
throughout the year, the optimal periods for the peak regulation range frequency regulation. All the time belongs to the frequency regulation
from 8:00 to 15:00 and from 18:00 to 24:00. According to multiple dominant periods. The energy storage system adopts the operation
refactoring operations of the AGSA, the optimal discharge point for the strategy of combining the A alliance and the AC alliance, which not only
energy storage system is obtained from 8:00 to 14:00, which can reserve ensures stable revenue throughout the day, but also sells the accumu­
sufficient energy space for the peak regulation and effectively avoid the lated excess electricity after executing frequency regulation commands
SOC exceeding SOCmax during operation. for a long time, maximizing the economic benefits of the energy storage
In summer, the amount of the abandoned-wind in summer is system.
significantly smaller than that in spring, and the benefits of the fre­
quency regulation are basically dominant. The peak regulation domi­ 6. Conclusions
nant periods are only concentrated between 17:00 and 20:00 every day,
making it easier to determine the discharge point for the energy storage Aiming at the problems of peak regulation pressure and severe
system. abandoned wind faced by the power grid under high penetration rate of
In autumn, the amount of the abandoned-wind power is located wind power, with the research background of peak regulation and fre­
between summer and spring, and the peak regulation dominant periods quency regulation in the wind power-thermal-shared energy storage
last almost from 16:00 until 24:00. The AGSA requires more recon­ coupling system, a dual-level optimization control method of the shared
struction operations than in summer to determine the discharge point of energy storage system in multiple application scenarios considering
the energy storage system, mostly concentrated in areas near the energy storage economy is proposed in this paper. By analyzing the
SOCmax . Under the premise of avoiding the SOC from exceeding SOCmax needs of multiple stakeholders involved in power grid auxiliary services,

17
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 19. The annual and typtical daily operating curve of the SOC in Scenario II.

18
X. Han et al. Applied Energy 350 (2023) 121801

Fig. 20. The annual and typtical daily operating curve of the SOC in Scenario III.

19
X. Han et al. Applied Energy 350 (2023) 121801

a dual-level dynamic cooperative game model for the shared energy energy storage system fluctuates frequently within the range of (0.1,
storage in multiple application scenarios is established. The main con­ 0.9) and is greatly affected by seasonality; In Scenario III, the major
clusions are as follows: alliance (ABC) operation mode is adopted in Spring, Summer, and
Autumn, and the (BC) alliance operation mode is only selected
1) The demand and profit relationships between multiple stakeholders during severe wind abandonment period (such as April); In winter,
involved in grid auxiliary services and the shared energy storage due to the low amount of abandoned wind power, the operation
system are explored. The frequency regulation and peak regulation modes of the frequency regulation (A) and the frequency regulation
commands are related to different time scales. The frequency regu­ energy market (AC) are alternately adopted. By using the AGSA to
lation commands belong to the level of seconds (s), and the SOC adaptively implement the collaborative operation of alliances and
fluctuation of the energy storage system is relatively small, resulting scheduling strategies according to different seasons and application
in less demand for the capacity of the energy storage system; The scenarios, the economy benefit of the shared energy storage system is
peak regulation commands belong to the level of hours (h), and the maximized.
SOC of the energy storage system fluctuates greatly, resulting in a
high demand for the capacity of the energy storage system; In CRediT authorship contribution statement
addition, the benefits of the frequency regulation and peak regula­
tion are related to seasonality. In spring, the proportion of peak Han Xiaojuan: Writing – review & editing, Project administration.
regulation dominant periods accounts for 41.7%; In summer and Li Jiarong: Writing – original draft, Software, Methodology, Formal
autumn, the proportion of frequency regulation dominant 7periods analysis, Conceptualization. Zhang Zhewen: Writing – review &
accounts for 87.5% and 75%, respectively; In winter, the proportion editing.
of frequency regulation dominant periods reaches 100%.
2) On the basis of cooperating with the daily optimal scheduling Declaration of Competing Interest
strategy of the lower model, the NSBWOA is used to obtain the Pareto
solution set of the capacity configuration for the shared energy We declare that we have no financial and personal relationships with
storage system in the upper level model, and the optimal capacity of other people organizations that can inappropriately influence our work,
the shared energy storage system in different scenarios is determined there is no professional or other personal interest of any nature or kind in
by the MABAC evaluation method. Through comparative analysis, any product, service and/or company that could be construed as influ­
the capacity of the shared energy storage system configured by encing the position presented in, or the review of, the manuscript
Scenario III is 49 MW/101 MW•h, located between Scenario I and entitled.
Scenario II. The cost- income ratio increased by 7.63% compared to
Scenario I, but decreased by 14.51% compared to Scenario II. But Data availability
the gross annual income increased by 2.749 million dollars, and the
high- value peak regulation ratio increased by 13.35%. The data that has been used is confidential.
3) The daily optimal scheduling strategy in the lower level model is
closely related to application scenarios and seasonal changes. In Acknowledgments
Scenario I, the SOC of the energy storage system operates very
smoothly, with a box operating within the range of (0.7, 0.9) for 352 This paper is supported by the National Natural Science Foundation
days, unaffected by seasonal changes; In Scenario II, the SOC of the of China (52277216).

Appendix A
Table A.1
The parameter settings in different application scenarios.

Parameters Value Unit

Simulation time interval 15 min


Peak price period 8 : 00 ∼ 12 : 00/17 : 00 ∼ 21 : 00 –
Normal price period 12 : 00 ∼ 17 : 00/21 : 00 ∼ 24 : 00 –
Valley price period 0 : 00 ∼ 8 : 00 –
FR compensation price 0.97 $/MW
PR compensation price 36.8/88.2/132.3 $/MWh
EM compensation price 57.1 $/MWh

Table A.2
The relevant parameters of the energy storage system.

Parameters Value Parameters Value

Charge and discharge efficiency η 95% Investment cost per unit power CP 323.5$/(kw)
Fitting coefficient β0 3832 Investment cost per unit capacity CE 273.5$/(kwh)
Fitting coefficient β1 0.68 Annual operation and maintenance cost per unit power CPO 14.06$/(kw)
Fitting coefficient β2 1.64 Annual operation and maintenance cost per unit capacity CEO 1.56$/(kwh)
Rated depth of discharge DODrated 0.8 SOCup 0.9
Discount rate γ 0.08 SOClow 0.1
Residual value rate σ 5% Standard response time tst 30s
Charge rate [0.25C, 2C] Lifetime 15/year

20
X. Han et al. Applied Energy 350 (2023) 121801

Table A.3
The parameter settings for different algorithms.

Algorithms NSBWOA NSGA-II MOPSO

Parameters

Population size 30 30 30
Number of iterations 5000 5000 5000
Cross probability – 0.8 –
Mutation probability – 0.1 –
Individual learning factor – – 1.7
Global learning factor – – 1.7
Inertial weight – – 0.9
Maximum speed – – 20

References [17] Chao C, et al. Capacity alocation method of hybrid energy storage system based on
empirical mode decomposition and Fuzy chance constrained programming. Distrib
Energy 2016;1(03):43–8.
[1] Guney MS, Tepe Y. Classification and assessment of energy storage systems. Renew
[18] Zhu KG, et al. Grid optimization of shared energy storage among wind farms based
Sustain Energy Rev 2017;75:1187–97.
on wind forecasting. In: 2018 IEEE/PES Transmission and Distribution Conference
[2] Benjaafar S, Hu M. Operations management in the age of the sharing economy:
and Exposition (T&D); 2018.
what is old and what is new? Manuf Serv Oper Manag 2019;22(1):93–101.
[19] Li R, et al. Cooperative planning model of renewable energy sources and energy
[3] Subburaj AS, Pushpakaran BN, Bayne SB. Overview of grid connected renewable
storage units in active distribution systems: a bi-level model and Pareto analysis.
energy based battery projects in USA. Renew Sustain Energy Rev 2015;45:219–34.
Energy 2019;168:30–42.
[4] Ho WS, et al. Optimal scheduling of energy storage for renewable energy
[20] Zhao B, et al. Optimal sizing, operating strategy and operational experience of a
distributed energy generation system. Renew Sustain Energy Rev 2016;58:1100–7.
stand-alone microgrid on Dongfushan Island. Appl Energy 2014;113:1656–66.
[5] Thompson CC, et al. Optimization of data center battery storage Investments for
[21] Li S, et al. Double-layer energy management system based on energy sharing cloud
Microgrid Cost Savings, emissions reduction, and reliability enhancement. IEEE
for virtual residential microgrid. Appl Energy 2021;282:116089.
Trans Indust Appl 2016;52(3):2053–60.
[22] Chen C, et al. Two-stage robust planning-operation co-optimization of energy hub
[6] Zakeri B, Syri S. Electrical energy storage systems: a comparative life cycle cost
considering precise energy storage economic model. Appl Energy 2019;252:
analysis. Renew Sustain Energy Rev 2015;42:569–96.
113372.
[7] Kim W-W, et al. Operation scheduling for an energy storage system considering
[23] Liu J, et al. Optimal planning and investment benefit analysis of shared energy
reliability and aging. Energy 2017;141:389–97.
storage for electricity retailers. Int J Electr Power Energy Syst 2021;126:106561.
[8] Celik E, Ozturk N. Novel fuzzy 1PD-TI controller for AGC of interconnected electric
[24] Lee J-O, Kim Y-S. Novel battery degradation cost formulation for optimal
power systems with renewable power generation and energy storage devices.
scheduling of battery energy storage systems. Int J Electr Power Energy Syst 2022;
Engineering. Sci Technol 2022:35.
137:107795.
[9] Cheng B, Powell WB. Co-optimizing battery storage for the frequency regulation
[25] Maluenda M, et al. Optimal operation scheduling of a PV-BESS-Electrolyzer system
and energy arbitrage using multi-scale dynamic programming. IEEE Trans Smart
for hydrogen production and frequency regulation. Appl Energy 2023;344:121243.
Grid 2018;9(3):1997–2005.
[26] Moafi M, et al. Optimal coalition formation and maximum profit allocation for
[10] Engels J, Claessens B, Deconinck G. Optimal combination of frequency control and
distributed energy resources in smart grids based on cooperative game theory. Int J
peak shaving with battery storage systems. IEEE Trans Smart Grid 2020;11(4):
Electr Power Energy Syst 2023;144:108492.
3270–9.
[27] Pamučar D, Ćirović G. The selection of transport and handling resources in logistics
[11] Ippolito MG, et al. Multi-objective optimized management of electrical energy
centers using multi-attributive border approximation area comparison (MABAC).
storage systems in an islanded network with renewable energy sources under
Expert Syst Appl 2015;42(6):3016–28.
different design scenarios. Energy 2014;64:648–62.
[28] Zhong C, Li G, Meng Z. Beluga whale optimization: a novel nature-inspired
[12] Wang C, et al. Distributed shared energy storage scheduling based on optimal
metaheuristic algorithm. Knowl Based Syst 2022;251:109215.
operating interval in generation-side. Sustain Energy Grids Netw 2023;34:101026.
[29] Jangir P, Jangir N. A new non-dominated sorting grey wolf optimizer (NS-GWO)
[13] Sardi J, et al. Multiple community energy storage planning in distribution networks
algorithm: development and application to solve engineering designs and
using a cost-benefit analysis. Appl Energy 2017;190:453–63.
economic constrained emission dispatch problem with integration of wind power.
[14] Hafiz F, et al. Energy management and optimal storage sizing for a shared
Eng Appl Artif Intel 2018;72:449–67.
community: a multi-stage stochastic programming approach. Appl Energy 2019;
[30] Deb K, et al. A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE
236:42–54.
Trans Evol Comput 2002;6(2):182–97.
[15] Jianlin L, et al. Optimal configuration strategy of energy storage capacity in wind/
[31] Taghavifar H, Mazari F. 1D diesel engine cycle modeling integrated with MOPSO
PV/storage hybrid system. Trans China Electrotech Soc 2018;33(06):1189–96.
optimization for improved NOx control and pressure boost. Energy 2022;247:
[16] Kargarian A, Hug G, Mohammadi J. A multi-time scale co-optimization method for
123517.
sizing of energy storage and fast-ramping generation. IEEE Trans Sustain Energy
2016;7(4):1351–61.

21

You might also like