You are on page 1of 11

Sustainable Energy, Grids and Networks 34 (2023) 101051

Contents lists available at ScienceDirect

Sustainable Energy, Grids and Networks


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

A data-driven stochastic energy sharing optimization and


implementation for community energy storage and PV prosumers✩

Bo Gu a,b , Chengxiong Mao a,b ,1 , Dan Wang a,b ,2 , Bin Liu a,b ,3 , Hua Fan a,b , , Rengcun Fang c,d ,
Zixia Sang c,d
a
State Key Laboratory of Advanced Electromagnetic Engineering and Technology, China
b
Huazhong University of Science and Technology, Wuhan 430074, China
c
Economic & Technology Research Institute, China
d
State Grid Hubei Electric Power Company, China

article info a b s t r a c t

Article history: The development of energy sharing community is demanded for the on-site consumption of distributed
Received 1 December 2022 generations and efficient utilization of renewable energy. This paper proposes a two-stage stochastic
Received in revised form 16 March 2023 energy sharing model considering the photovoltaic (PV) power uncertainties to minimize the social
Accepted 17 April 2023
cost of PV prosumers and community energy storage (CES) in community. The proposed optimization
Available online 22 April 2023
model derives the optimal schedule of community under the worst case in the first stage. The
Keywords: regulation model for PV prosumers is formulated in the second stage to address the PV uncertainties
Energy sharing including three regulation methods: the demand response, the charging/discharging schedule, and the
PV prosumers transactive energy schedule. Particularly, a modeling approach based on the data-driven uncertainty
Community energy storage set is introduced to find the worst probability distribution. The formulated two-stage problem can be
Stochastic optimization solved by the column-and-constraint generation algorithm, where the coupled model is decomposed
Energy hub into a main problem and a sub problem and solved by iteration. Numerical simulation results show
the effectiveness of the proposed energy sharing optimization model to decrease the energy cost.
Experimental results demonstrate the feasibility of multi-ports energy hub used in energy sharing
community.
© 2023 Elsevier Ltd. All rights reserved.

1. Introduction for local energy prosumers. The application of energy storage


systems (ESSs) can be effectively to achieve energy management
Photovoltaic (PV) generations are more widely distributed in prosumers side. In practical, some prosumers may not have
compared with other renewable energy sources. In recent years, the private energy storages (PESs) due to the high investment
the market of the distributed PV has developed rapidly [1]. For budget. The installation of community energy storage (CES), also
example, in China, regional distributed PV construction based on called shared energy storage, is an effective approach for PV
the commercial or residential building rooftop PV is becoming prosumers to share energy and realize the energy cooperation in
one of the most important trends in green sustainable develop- community [4–6].
ment, which also supports carbon peaking and carbon neutrality The coordinated operation strategies of energy community
[2,3]. However, the integration of large-scale distributed PV to have been investigated in many works [7–9]. A cooperative en-
the utility grid may have the negative impact on the operation ergy management system is introduced in [7] to reduce the
of the grid due to the volatility, dispersion, and intermittence. operational cost of multi-microgrids. In [8], a hierarchical energy
Thus, the on-site consumption and efficient utilization are urgent management framework in local energy communities is proposed
and conducted by minimizing the costs in the individual opti-
✩ This work was supported by State Grid Hubei Electric Power Company mization stage and maximizing the on-site consumption in the
Science and Technology Project, China (52153820000H).
second stage. Cost and emissions are comprehensively consid-
∗ Corresponding author at: Huazhong University of Science and Technology, ered in community by a multi-objective optimization framework
Wuhan 430074, China. in [9]. The technical and economic analysis of CES for energy
E-mail address: fanhua@hust.edu.cn (H. Fan). community is also presented in [10–14]. However, the proposed
1 Fellow, IEEE schemes in [8,10] lacks the adequate energy interaction between
2 Senior Member, IEEE CES and each prosumer, where CES may not be fully utilized and
3 Student Member, IEEE is not flexible. Similar research has been done on game-based

https://doi.org/10.1016/j.segan.2023.101051
2352-4677/© 2023 Elsevier Ltd. All rights reserved.
B. Gu, C. Mao, D. Wang et al. Sustainable Energy, Grids and Networks 34 (2023) 101051

prosumers in the second stage. Compared with the current litera-


ture about the investigations of energy community with CES, the
main differences of our work are summarized as follows. Firstly,
PESs are considered in this work, where the energy sharing strat-
egy is more flexible. Secondly, all participants in community can
directly share the surplus energy with each other instead of only
sharing part of CES. Thus, CES and PESs can be further utilized.
Thirdly, the two-stage model focuses on the social cost and reg-
ulation cost, which ensures the willingness of all participants to
join energy sharing community. Also, the formulated two-stage
model is easily solved by the column-and-constraint generation
(C&CG) algorithm. Lastly, a data-driven stochastic programming
method is introduced to assume that the probability distribution
from the historical PV data is uncertain within a confidence set in
energy sharing community. Accordingly, the main contributions
of this paper are summarized as follows:
(1) A novel two-stage stochastic energy sharing optimiza-
tion model for energy community. The model integrates the
social cost of community in the first stage and regulation cost
of PV prosumers in the second stage, which can be easily solved
Fig. 1. Energy sharing community via energy hub in this work. by iteration. The proposed optimization framework has been
implemented on a multi-ports energy hub (EH) [30].
(2) A data-driven stochastic programming method is applied,
energy sharing or energy trading strategy of community with PV where the probability distribution of the historical PV data is
prosumers and CES, and can be nearly divided into cooperative data-driven uncertainty set. Compared with the relatively con-
game-based strategy and noncooperative game-based strategy. servative results of traditional uncertainty set, the optimal re-
The Stackelberg game-based energy cooperation models are in- sults of data-driven uncertainty set improve. Moreover, some
vestigated in [13,15,16]. The designs of energy trading system of the worst-cases can be excluded in proposed energy sharing
based on CES and prosumers are introduced in [13,15], where framework.
CES acts as the leader to make first decisions and prosumers (3) A comprehensive demand response (DR) model is proposed
based on the price information and the demand preferences of
act as the followers to make second decisions. In [16], a hybrid
prosumers.
energy sharing model is formulated, which consists of day-ahead
The rest of this paper is organized as follows. In Section 2, the
scheduling and real-time pricing. Moreover, an energy sharing
structure of energy sharing community is detailed. Section 3 in-
framework in community is investigated in [17], which aims to
troduces the formulation of two-stage energy sharing model, and
improve community energy efficiency and energy economy. [18]
the profit allocation method is also included. Section 4 presents
also proposes a cooperative game model and profit allocation
the modeling for uncertainty set and corresponding algorithm.
scheme for integrated energy system with CES. For noncoop-
The numerical results are given in Section 5. The experimen-
erative game-based models, the bargaining game-based models
tal results are given in Section 6. The conclusion is drawn in
are extensively investigated in [19–22]. The benefits from the
Section 7.
energy sharing and energy cooperation are allocated to all partic-
ipants in community. A multi-objective optimization framework 2. Structure of energy sharing community
is proposed in [23] by using different types of CES. However,
the imperfections still exist in some of aforementioned literature. The structure of energy sharing community is shown in Fig. 1.
The formulated models are deterministic without including the The included PV prosumers and CES realize the energy cooper-
uncertainty. Thus, the worst-case in the uncertain scenarios has ation through EH. PV prosumers are equipped with rooftop PV
negative impact on the realistic results. panels to generate power. Battery energy storage system (BESS)
To address the uncertain variables in energy sharing models may be used as PES based the construction budget of each pro-
of PV prosumers and CES, the stochastic programming model sumer. CES is managed by the agent and cooperates with PV
or robust model is widely used [24–27]. A cost-effective en- prosumers as the community public ESS. The agent optimizes the
ergy management framework is proposed with PV power and operation of community based on the PV generation profile, load
demand uncertainties in [24]. In [25], an effective CES control profile and prices information from the power grid, and transmits
policy is introduced, which is capable to save energy cost of the schedule to PV prosumers and CES in energy sharing com-
prosumers. A two-stage robust energy sharing framework is in- munity by means of local communication. As shown in Fig. 1, EH
troduced in [26] and a bilevel stochastic optimization approach realizes the flexible interconnection mode of all participants in
is proposed in [27]. However, CES is not fully utilized in [25] community, e.g., PV prosumers can cooperate with each other in
and each prosumer only shares part of CES. Also, the modeling the form of AC or DC with different voltage through the internal
methods of uncertainty data in [26,27] are relatively conserva- multi-ports converters in EH, where the energy sharing between
tive. [28] investigates a distributed robust energy sharing opti- each participant can be easily achieved. Community is connected
mization strategy in a microgrid, but CES is not included in the to the grid through the grid-tie converter.
model to improve the flexibility of operation. The comparison of The community composed of PV and ESSs satisfies green
the aforementioned energy sharing optimization and operation energy sharing and meets the current development trend of
strategy of CES-based energy systems is provided in Table 1. emission-free microgrid [1]. However, the PV profiles of pro-
In this context, we propose a data-driven stochastic energy sumers largely depends on the local weather. Therefore, the PV
sharing optimization model for PV prosumers and CES. The two- profiles have great uncertainties and cannot be ignored by energy
stage optimization model is to minimize the social cost of com- sharing community. A data-driven two-stage distributed stochas-
munity in the first stage and minimize the regulation cost of tic energy cooperative model is proposed in this paper. In the
2
B. Gu, C. Mao, D. Wang et al. Sustainable Energy, Grids and Networks 34 (2023) 101051

Table 1
Comparison of energy sharing optimization and operation strategy.
Ref. No Energy storage Solution frame- Objective Revenue Control mode Optimization
device work/algorithm allocation horizon
[7] PES, CES MAM Min. total operation cost Hierarchical DA
[8] CES MPC First stage: Min. operation cost Hierarchical DA, RT
Second stage: Max. self-sufficiency and
self-consumption
[9] CES MO Min. electricity cost Centralized DA
Min. CO2 emission
[13] CES NCG Min. the social operation cost Centralized DA
[15] CES NCG Min. total energy cost Distributed DA
[16] CES SO, SG First stage: Min. CES economic loss Hierarchical DA, RT
Second stage: Max. total profit
[17,19] CES ADMM Min. social energy cost ✓ Distributed DA
[18] CES CG Min. total operation cost Centralized DA
[20,21] CG Max. social welfare ✓ Distributed DA
[22] PES CG Min. social energy cost ✓ Distributed DA
[23] CES MO Min. total operation cost Centralized DA
[24] PES SO Min. total operation cost Distributed DA, RT
[25] PES SO Min. the electricity cost Distributed DA, RT
[26] PES, CES RO First stage: Min. total operation cost Hierarchical DA, RT
Second stage: Min. adjustment cost
[27] PES SO Upper level: Max. total profit Distributed DA, RT
Lower level: Max. prosumers profit
[28] PES RO First stage: Min. total operation cost Hierarchical DA, RT
Second stage: Min. regulation cost
[29] PES MO Min. electricity cost ✓ Centralized DA
This paper PES, CES SO First stage: Min. total operation cost ✓ Distributed DA, RT
Second stage: Min. regulation cost

*MAM: multiagent method; MPC: model predictive control; MO: multi-objective optimization; NCG: non-cooperative game; SO: stochastic optimization; SG:
Stackelberg Game; ADMM: alternating direction method of multipliers; CG: cooperative game; RO: robust optimization; DA: day-ahead; RT: real-time

first stage of the model, the objective function is to minimize the where ηsl1 is the coefficient of price-based DR.
social cost of community, and the cooperative operation scheme
,
⎨ 0∆ 0 < ∆pr < pr ,min

of PV prosumers and CES is formulated as the day-ahead optimal p r −pr ,min
schedule. Based on the day-ahead economic dispatch schedule η sl1
= pr ,max −pr ,min
, pr ,min < ∆pr < pr ,max (3)
sl1
η , ∆pr > pr ,max

in the first stage, PESs input/output power, load profiles, and max
purchasing/selling power of PV prosumers and CES are adjusted
In Eq. (3), ∆pr is the difference of the peak price and the valley
in the second real-time dispatch stage under the uncertainties of price. Prosumers participate in DR of energy sharing community
PV profiles. Since the objective function is to maximize the overall if ∆pr is bigger than the minimum response price. ηsl1 is linear
profit of community, the reasonable allocation of profit among with ∆pr if ∆pr is between pr ,min and pr ,max . The comprehensive
participants in community is worthy consideration. Thus, a Nash cost function of DR can be formulated in (4) and (5).
bargaining-based model is formulated to solve the problem. The ⏐ ⏐
dr sl
detailed model and algorithm are depicted as follows. Cpro1,n,h = ε dr1 ⏐pn,1h ⏐ ∆H (4)
⏐ ⏐
⏐ ⏐ ⏐ ⏐
dr ⏐ sl ⏐ ⏐ sl ⏐
3. System modeling Cpro2,n,h = ε dr2 ⏐pn,2h ⏐ ∆H + ε drcon ⏐pn,2h ⏐ ∆H (5)
dr
Before the detailed mathematical formulation, it is assumed where Cpro1,n,h is the quantitative cost of the unwillingness for
that the number of PV prosumers is N, which is denoted as N . The the price-based DR of prosumer n at time slot h and ε dr1 is the
dr
dispatch horizon of energy sharing community is a scheduling cost coefficient. Cpro2,n,h is the quantitative cost of the preferences
day denoted as H with 24 time slots ∆H. of prosumers. ε dr2
is the cost coefficient of the prosumers sat-
isfaction, which is negative in (5). The community acts as an
3.1. PV prosumers modeling aggregator integrating prosumers to participate in the optimal DR
with the contracts. ε drcon is the punishment cost coefficient of the
default DR.
Before formulating the functions of PV prosumers in energy
For any prosumer n in the prosumers community N , the
sharing community, it is necessary to formulate a DR model of
objective function is to minimize the electricity cost during the
prosumers firstly. In this work, a DR model that comprehensively
scheduling day. The cost function of a PV prosumer is formulated:
considers the price and preferences of prosumers is proposed. The
load of each prosumer can be divided into the fixed load and the
ec dr bc
shiftable load (SL). Cpro,n,h = Cpro,n,h + Cpro,n,h + Cpro,n,h (6)
l l sl sl
pn0,h = pnd,h + pn,1h + pn,2h (1) where

,n,h = λh pn,h ∆H − λh pn,h ∆H


ec b b s s
sl Cpro (7)
In Eq. (1), the SL is further divided into the price-based SL pn,1h
dr dr dr
which is responding to the price information, and the preferences- Cpro,n,h = Cpro1,n,h + Cpro2,n,h (8)
sl
based SL pn,2h which is responding to the preference of each
sl
bc
Cpro,n,h =ε bc
(pchar
n,h + n,h )∆H
pdis (9)
prosumer. pn,1h can be described as
Eq. (6) is the cost function for prosumer n at time slot h includ-
l sl l
−ηsl1 pn0,h ≤ pn,1h ≤ ηsl1 pn0,h (2) ec
ing equations (7)–(9). Cpro ,n,h is the energy cost calculated by (7),
3
B. Gu, C. Mao, D. Wang et al. Sustainable Energy, Grids and Networks 34 (2023) 101051

bat ,max
ces,h , pces,h ≤ pces
0 ≤ pchar dis
which is the difference between the cost of energy purchasing (27)
from the grid at the electricity price (EP) λbh and the income of
µ char
ces,h +µ dis
ces,h ≤1 (28)
energy selling to the grid at the feed-in-tariff (FT)λsh . Cpro
dr
,n,h is the
bc
cost for DR of prosumer n at time slot h.Cpro,n,h is the quantitative ηmin
ces Sces ≤ Sces,h ≤ η max
ces Sces (29)
degradation cost of BESS charging/discharging of prosumer n. ε bc Sces,h = (1 − ξces )Sces,h−1 + (ηces pces,h − ηces
char char
pces,h )∆H
dis dis
(30)
bc
is the degradation cost coefficient of BESS. It is noted that Cpro ,n,h
Sces,0 ≤ Sces,H .
}
(31)
is not included for prosumers without BESSs.
Sets X and Y are the constraint sets for prosumer n and
3.2. Community energy storage modeling CES, respectively. Eqs. (14)–(24) are the constraints set for PV
prosumer n. Eqs. (15)–(18) are the constraints for the DR load and
For CES in the energy sharing community, the objective func- the actual load. ηdr is the proportional parameter of SL. Eq. (18)
tion is also to minimize the electricity cost during the scheduling is the DR preference constraint for different prosumers. Eq. (19)
pv
day. The cost function of CES is formulated: is to ensure the power balance at any scheduling time slot. pn,h
tr
ec
Cces,h = Cces bc is the output power of PV and pn,h is the power of prosumer
,h + Cces,h (10)
n sharing with CES. If ptr n,h >0, prosumer n transmits the power
where to CES at time slot h; If ptr n,h <0, prosumer n absorbs the power
from CES at time slot h. Eqs. (20)–(24) are the constraints for the
,h = λh pces,h ∆H − λh pces,h ∆H
ec b b s s
Cces (11)
prosumers equipped with the private BESSs. Eq. (20) indicates
bc
Cces ,h =ε ces
(pchar
ces,h + ces,h )∆H
pdis (12) the charging/discharging power limits for the private BESSs of
prosumer n. Eq. (21) is the charging/discharging binary variables
Eq. (10) is the cost function for CES at time slot h includ-
ec constraint. Eq. (22) represents the limits for the state of charge
ing equations (11) and (12). Cces ,h is the energy cost of CES (SoC) of BESSs under the normal operation. Eq. (23) is the formula
calculated by (11), which is the same as the calculation pro-
ec bc of SoC, which is calculated by the SoC of the last hour and
cess of Cpro,n,h . Cces,h is the quantitative degradation cost of CES the charging/discharging power in current hour. ξn is the scale
charging/discharging. ε ces is the degradation cost coefficient of
coefficient of the energy leakage. ηnchar and ηndis are the efficiency
CES.
parameters of charging/discharging. Eq. (24) ensures that the SoC
at the last hour of a scheduling day returns to the initial state.
3.3. Energy sharing system modeling
Note that Eqs. (20)–(24) are not included by prosumers without
BESSs.
The objective function and constraints are mathematically for-
Eqs. (25)–(31) is the constraints set for CES. In Eq. (25), ptr ces,n,h
mulated in this subsection based on the cost functions of PV
is the power of CES sharing with prosumer n. If ptr ces,n,h >0, CES
prosumers and CES. The objective function of community with
energy cooperation is to minimize the social energy cost in the transmits the power to prosumer n at time slot h; If ptr ces,n,h <0,
scheduling day H. The corresponding objective and constraints CES absorbs the power from prosumer n at time slot h. Eq. (25)
for the prosumers and CES are as follows: ensures the power balance of the energy sharing between any
prosumer and CES in community. Eq. (26) is to ensure the power
H ∑
N
∑ balance of CES at any scheduling time slot. Eq. (27) indicates the
min ( Cpro,n,h + Cces,h ) (13) charging/discharging power limits for CES. Eq. (28) is the charg-
xn ∈X
y∈Y h=1 n=1 ing/discharging binary variables constraint. Eq. (29) represents
subject to: the limits for the SoC of CES under the normal operation. Eq. (30)
is the calculation formula of SoC like in (23). ξces is the scale
X = xn |∀h ∈ H, ∀n ∈ N ,
{
coefficient of the energy leakage for CES. ηces char
and ηces
dis
are the
pbn,h , psn,h ≥ 0 (14) efficiency parameters of charging/discharging. Eq. (31) ensures
l sl sl
that the SoC of CES at the last hour of a scheduling day returns
pln,h = pn0,h + pn,1h + pn,2h (15) to the initial state like in (24).
−η sl1 l0
pn,h ≤
sl
pn,1h ≤η sl1 l0
pn,h (16) The objective function (13) with constraints set {X , Y } can be
denoted as F1. If the collected day-ahead PV profiles of prosumers
dr l0 sl sl2 dr l0
−η ≤ pn,h+ ≤η pn,1h pn,h pn,h (17) community N are accurate, the decision set of prosumer n {pbn ,
∑H ⏐⏐ sl1 psn , pdr
n , pn
char
, pdis tr b
n , pn } and the decision set of CES {pces , pces ,
s

sl2 ⏐
h=1 ⏐pn,h + pn,h ⏐ char dis tr
pces , pces , pces,n } can be the scheduling decision variables for the
1− ∑ H l0 ≥ σn,min (18)
energy sharing community. However, the optimal value of F1 will
h=1 pn,h
deteriorate due to the uncertainties of PV profiles. Thus, the reg-
pv
pn,h + pbn,h + pdis char l s tr
n,h = pn,h + pn,h + pn,h + pn,h (19) ulation cost model of prosumers for the PV power uncertainties
is necessary.
0≤ pchar
n ,h , pdis
n,h ≤ pnbat ,max (20)
µ char
n,h +µ dis
n ,h ≤1 (21) 3.4. Regulation modeling under PV uncertainties
s,min s,max
η n Sn ≤ Sn,h ≤ η n Sn (22)
The regulation objective function of PV prosumers in commu-
Sn,h = (1 − ξn )Sn,h−1 + (η char char
n pn,h −η n pn,h )∆H
dis dis
(23) nity is to minimize the regulation cost caused by the PV uncer-
tainties in the scheduling day H. The corresponding objective and
Sn,0 ≤ Sn,H .
}
(24)
constraints are as follows:
H ∑
N
Y = y |∀h ∈ H,
{ ∑ pl
min (Cntrc brc
,h + Cn,h + Cn,h ) (32)
ptr tr
n,h + pces,n,h = 0 (25) zn ∈Z
h=1 n=1
N
∑ where
pchar s
ces,h + pces,h + ptr dis b
ces,n,h = pces,h + pces,h (26) trc ,b b
n=1 ,h = ε
Cntrc λh ∆pbn,h ∆H + ε trc ,s λsh ∆psn,h ∆H (33)
4
B. Gu, C. Mao, D. Wang et al. Sustainable Energy, Grids and Networks 34 (2023) 101051

,h = ε (∆pn,h + ∆pn,h )∆H


Cnbrc brc char dis
(34) 4. Two-stage distributed optimization algorithm
pl pv pv
Cn,h =ε pl
(p̂n,h − pn,h )∆H (35)
4.1. Data-driven optimization considering uncertainties
subject to:
To address the uncertainties of PV power, a two-stage stochas-
Z = {zn |∀h ∈ H, ∀n ∈ N ,
tic optimization algorithm framework is given. Compared with
pv
n,h + ∆pn,h ) =
pn,h + (pbn,h + ∆pbn,h ) + (pdis dis
the conservative results of one-stage robust framework, the re-
sults of the adopted framework are relatively positive. Suppose
n,h + ∆pn,h ) + pn,h + (pn,h + ∆pn,h ) + pn,h
(pchar char l s s tr
(36)
that the decision problems of the first stage and the second
bat ,max
n,h + ∆pn,h ≤ pn
0 ≤ pchar char
(37) stage are linear programming, and the uncertainty set U is a
,max discrete finite point set or polyhedron. Let y denotes the first
0≤ pdis
n ,h + ∆pdis
n ,h ≤ pbat
n (38)
s,min s,max
stage decision variables set, x denotes the second stage decision
η n Sn ≤ Sn,h ≤ η n Sn (39) variables set, and U denotes the uncertainty vector. The general
form of two-stage stochastic optimization under this assumption
Sn,h = (1 − ξn )Sn,h−1 + ηnchar (pchar
n,h + ∆pn,h )
char
{
is [31]:

n,h + ∆pn,h ) ∆H
−ηndis (pdis dis
}
(40) min c T y + max min bT x (47)
y u∈U x∈F (y ,u)
Sn,0 ≤ Sn,H .
}
(41)
subject to:
Eqs. (33)–(35) are the regulation cost items. Cntrc
,h is the ad-
ditional cost from the additional buying energy ∆pbn,h and the AT y ≥ d , y ∈ S y (48)
additional selling energy ∆psn,h . Cnbrc
,h is the regulation cost for where F (y , u) = {x : Gx ≥ h − Ey − M u, x ∈ S x }, S x ⊆ R+ , S y ⊆ m
pl
private BESSs of prosumer n. Cn,h is the cost of PV curtailment Rn+ . u is the uncertainty vector. The second stage constraints
pv
calculated by the practical PV output p̂n,h and the scheduling PV F (y , u) is a linear function of u. Combining the objective functions
pv
output pn,h . ε is the loss coefficient of PV curtailment.
pl F1 and F2, the objective function of energy sharing commu-
Set Z is the constraint set for the regulation stage of pro- nity based on the social energy cost in the first stage and the
sumers. Eqs. (36)–(41) are similar to Eqs. (19)–(24), referring to regulation cost in the second stage can be expressed as
previous subsection for specific description. { H ∑
∑ N
The objective function (32) with constraints set Z can be min ( Cpro,n,h + Cces,h )+
xn ∈X
denoted as F2. y∈Y h=1 n=1
H ∑
N
}
3.5. Bargaining-based profit allocation
∑ pl
max min (Cntrc
,h + Cnbrc
,h + Cn,h ) (49)
u∈U zn ∈Z
h=1 n=1
To ensure the willingness of prosumers and CES to join energy
sharing community for energy cooperation, the profit from the subject to
energy sharing framework need to be allocated fairly. The Nash X = {xn |∀h ∈ H, ∀n ∈ N , (14) − (24)} (50)
bargaining method, which is a frequently used cooperative game
framework, is suitable to achieve the fair and Pareto optimal Y = {y|∀h ∈ H, (25) − (31)} (51)
profit allocation in energy sharing community. The standard Nash Z = {zn |∀h ∈ H, ∀n ∈ N , (36) − (41)} (52)
bargaining based profit allocation model is formulated as follows:
where constraints set {X , Y , Z } are as mentioned before. Com-
N
( N
) pared with the traditional two-stage stochastic optimization, the
second stage max–min function in (49) can be converted to the
∏ ∑
max (Cndiff − Cnadd ) Cces
diff
+ Cnadd (42)
wn ∈W expected regulation cost function with the ambiguous probability
n=1 n=1
distribution:
subject to: { H ∑
N
{ ∑
min ( Cpro,n,h + Cces,h )+
W = wn |∀h ∈ H, ∀n ∈ N , xn ∈X
y∈Y h=1 n=1
{ H ∑
N
}}

Cndiff − Cnadd ≥0 (43) max EP min (Cntrc
,h + Cnbrc
,h + pl
Cn,h ) (53)
} p∈Ω zn ∈Z
N
∑ h=1 n=1
diff
Cces + Cnadd ≥ 0. (44) In (53), the ambiguous probability distribution p represents
n=1 the probability distribution for the discrete scenarios to happen.
diff diff
where Cn and Cces are the differences between the costs of The further description of the max–min problem in (53) is as
prosumers and CES after joining energy sharing community and follow:
that before joining. Cnadd is the additional cost of prosumer n by K
{ H ∑
N
}
∑ ∑ pl
bargaining based profit allocation model. Eq. (42) can be further max pk min (Cntrc brc
,h + Cn,h + Cn,h ) (54)
p∈Ω zn ∈Z
equivalently transformed into the following form: k=1 h=1 n=1

where K is the number of the discrete scenarios and pk is the


N
( N
)
∑ ∑
max ln(Cndiff − Cnadd ) + ln diff
Cces + Cnadd (45) corresponding probability for the discrete scenery k to happen.
wn ∈W
n=1 n=1 The expected regulation cost can be maximized by searching
for the worst-case probability distribution. To make the calcu-
subject to:
lated probability distribution p close to the actual data, referring
W = {wn |∀h ∈ H, ∀n ∈ N , (43) − (44)} (46) to [32,33], a method of generating the probability distribution p
5
B. Gu, C. Mao, D. Wang et al. Sustainable Energy, Grids and Networks 34 (2023) 101051

based on the norm-1 and norm-inf is adopted to construct the Table 2


confidence constraint set Ω . Algorithm procedure.
C&CG Algorithm for Energy Sharing Community
Ω = p|p ∈ RK + ,
{
Initialize LB, UB, ε , k, and initial PV probability distribution p0 .
p − p0  ≤ θ for k=1 to K do
 
(55)
1 if k=1 do
K
∑ p ← p0
⏐pk − p0 ⏐ ≤ θ
⏐ ⏐
(56) else
k
Solve MP (1.47) with{p. Derive the optimal schedule } of energy
k=1
sharing community: pln∗ , pbn∗ , psn∗ , pchar
n

, pdis
n , pn
∗ tr ∗
. Derive the
≤θ
 
p − p0  (57) social cost of the first stage F1. Derive the optimal cost of MP and

} assign to LB.
Solve SP (1.52) with the optimal schedule. Derive the new PV
max ⏐pk − p0k ⏐ ≤ θ .
⏐ ⏐
(58) probability distribution p, the optimal regulation cost of SP and UB.
1≤k≤K
end
(55) and (57) are two confidence sets. p0 is the initial prob- Calculate convergence error εk =UB-LB
ability distribution. Supposing that K discrete scenarios is from if εk ≤ ε do
break
M observations, where M observations can be generated by else
the scenario generation method, for example the Monte Carlo k=k+1
simulation [32], and K discrete scenarios can be generated by end
the scenario reduction method, for example the commonly used end
if k>K do
K-means clustering method. Each discrete scenario includes Nk The two-stage stochastic optimization energy sharing model is
initial scenarios. Thus, p0 can be described as unsolvable. PV prosumers and CES operate without cooperation.

p0 = Nk /M (59)

The limits of the probability deviation θ can be described as subject to


Pr p − p0 1 ≤ θ ≥ 1 − 2Ke−2Mθ /K
{  }
(60) AT y ≥ d , y ∈ S y (68)
Pr p − p0 ∞ ≤ θ ≥ 1 − 2Ke−2Mθ
{  }
(61) η ≥ bT xl , ∀l ∈ O (69)
The confidence level of the probability distributions in Eqs. (60) Gx ≥ h − Ey − M ul , ∀l ≤ k
l ∗
(70)
and (61) can be set as α1 and α∞ .

Pr p − p0 1 ≤ θ1 ≥ α1
{  }
(62)
y ∈ S y , xl ∈ S x , η ∈ R, ∀l ≤ k (71)
≤ θ∞ ≥ α∞
{  }
Pr p − p0  ∞
(63)
(71) where k is the number of the iterations. The optimal solu-
The relationship between the confidence level α and the de- tions can be derived and denoted as x1∗ , . . . , xk∗ , y ∗k+1 , ηk∗+1 .
{ }
viation limit θ can be expressed by Updating LB, LB is c T y ∗k+1 + ηk∗+1 .
K 2K The SP is to obtain the worst-case probability distribution in
θ1 = ln (64) the second stage with the acquired variable y ∗k+1 and return the
2M 1 − α1
distribution to the MP for next iteration. Accordingly, the SP can
1 2K
θ∞ = ln (65) be formulated as
2M 1 − α∞
SP (y) = max min bT x (72)
In (64) and (65), the limits θ1 and θ∞ are close to zero with u∈U x∈F (y ,u)
the increase of the number of the PV historical data M.
subject to
The data-driven uncertainty set contains the historical data
and excludes the extreme scenarios. Compared with the box Gx ≥ h − Ey − M u, x ∈ S x (73)
uncertainty set or the ellipsoidal uncertainty set, the decision
variables of the data-driven stochastic optimization are less con- where SP provide an upper bound (UB) of Eq. (53).
servative [33]. UB = min UB, c T y ∗k+1 + SP (y ∗k+1 )
{ }
(74)

4.2. C&CG algorithm If the difference between LB and UB is less than the given con-
vergence error ε , the iteration ends, otherwise the next iteration
will be performed. The complete algorithm procedure is shown
The data-driven two-stage stochastic optimization model of
in Table 2.
the energy sharing community can be solved by the C&CG algo-
rithm [31]. The algorithm is used to divide the model (53) into
5. Simulation results
the main problem (MP) and the sub-problem (SP) for iteration.
The MP is to solve the optimal schedule of community that
5.1. Basic data
satisfies the constraints under the worst-case probability distri-
bution p, and provide a lower bound (LB) of Eq. (53). Assume that
The results of case study are discussed in this section to
xl is the optimal solution of (66).
evaluate the performance of the two-stage distributed stochastic
η = max min bT x (66) optimization framework in this work.
u∈U x∈F (y ,u)
In this case study, it is assumed that the energy sharing com-
Accordingly, the MP of general two-stage stochastic optimiza- munity consists of five PV prosumers equipped with the rooftop
tion model (47) can be formulated as PV panels with the capacities range from 60 kW to 80 kW. The
actual initial demand profiles of each prosumer, coming from the
MP = min c T y + η (67) realistic demand data of buildings in a city of China, are shown in
y

6
B. Gu, C. Mao, D. Wang et al. Sustainable Energy, Grids and Networks 34 (2023) 101051

Table 3
Parameters of ces and pes.

Table 4
Statistical comparison results.
Without cooperation Cooperation
Mean Variance (×103 ) Mean Variance (×103 )
Prosumer 1 15.81 1.31 −1.56 0.028
Prosumer 2 19.69 0.56 −3.92 0.039
Prosumer 3 16.47 0.51 −3.90 0.027
Prosumer 4 21.51 1.17 −1.48 0.108
Prosumer 5 21.69 0.55 −6.13 0.159
CES 0.00 0.00 109.28 25.72
Community 95.17 14.84 92.29 30.54

rapid lifetime degradation due to overwork. The initial capacities


of CES and private BESSs are set to 50% at 0:00 of a day. According
to the predicted PV power in Fig. 2(b), the output profiles are
mainly concentrated at 7:00 to 16:00 of a day. Other parameters ε
and K are 1E-4 and 50, respectively. The aforementioned problem
is solved using Matlab 2021a software with Cplex and Mosek
toolbox on Intel-i9 personal computer with 64 GB RAM.

5.2. Energy sharing profiles in the first stage

The energy performance profiles of community in the first


stage are shown in Fig. 3. Fig. 3(a) and (b) are the net de-
mands profiles of prosumers and community with the energy
non-cooperation and cooperation. In Fig. 3(a), the net demands of
prosumers are equal to the net demands of community because
CES does not work. All prosumers only depend on the utility
grid and optimize the operation according to the EP and FT
profiles. In Fig. 3(b), CES participates in the energy sharing with
prosumers. Thus, all prosumers can purchase electricity from or
sell electricity to the utility grid more flexibly. Note that with
the energy cooperation, CES will assist prosumers to achieve the
smaller fluctuation of the net demands and low dependence on
the utility. The statistical data are shown in Table 4, where the
means and the variances of prosumers without cooperation are
much bigger that those of prosumers with energy sharing. The
variances of energy sharing community is relatively bigger than
the variances of community without cooperation because of the
rapid regulation of CES for the load peak in some time slots,
which is also shown in Fig. 3(b). The energy sharing profiles of
prosumers and CES are shown in Fig. 3(c). CES shares the energy
with prosumers in the load peak periods or PV valley periods. In
the PV peak periods, CES stores the surplus renewable energy and
Fig. 2. Basic data. each prosumer can realize the energy sharing and energy balance,
for example from 8:00 to 17:00.

Fig. 2(a). The predicted PV power profiles are shown in Fig. 2(b). 5.3. Regulation profiles in the second stage and energy costs
The EP and FT profiles of the power grid are shown in Fig. 2(c)
similar with the prices in [16]. The SL of prosumers are set to The regulation profiles of PV prosumers in the second stage are
30% of the demand at any time slot. The detailed parameters of shown in Fig. 4(a), (b), (c), (d) and (e). The solid lines represent the
CES and private BESSs of prosumers are shown in Table 3, where optimal schedule in the first stage and the dotted lines represent
prosumer 1 and prosumer 4 are equipped with the household the optimal schedule considering the real-time operation in the
BESSs to realize the home energy management. The reasonable second stage under PV uncertainties. The regulation means of
capacity ranges of CES and private BESSs are set to prevent the prosumers include the load transfer, the change of net demands
7
B. Gu, C. Mao, D. Wang et al. Sustainable Energy, Grids and Networks 34 (2023) 101051

Table 5 Table 6
Cost comparison results(¥). Profit allocation results(¥).
Without cooperation Cooperation Before allocation After allocation
Cost.1 Cost.2 Total Cost.1 Cost.2 Total Total Cost Profit Total Cost Profit
Prosumer 1 179.2 3.7 182.9 −25.3 3.4 −21.9 Prosumer 1 −21.9 204.8 133.17 49.73
Prosumer 2 321.5 30.2 351.7 −43.2 28.1 −15.1 Prosumer 2 −15.1 366.8 301.97 49.73
Prosumer 3 257.1 18.5 275.6 −40.6 20.7 −19.9 Prosumer 3 −19.9 295.5 225.87 49.73
Prosumer 4 236.8 18.1 254.9 −10.6 14.4 3.8 Prosumer 4 −3.8 251.1 205.17 49.73
Prosumer 5 336.7 39.5 376.2 −71.5 25.5 −46.0 Prosumer 5 −46 422.2 326.47 49.73
CES 0 0 0 1242.0 0 1242.0 CES 1242.0 −1242 −49.73 49.73
Community 1331.3 110.0 1441.3 1050.8 92.1 1142.9 Community 1142.9 298.4 1142.9 298.4

Table 7
Results of different combinations.
α1 Total costs/¥
α∞ =0.5 α∞ =0.9 α∞ =0.99
0.2 1126.50 1130.17 1130.17
0.5 1142.92 1142.92 1142.92
0.9 1142.92 1142.92 1159.34

Table 8
Comparison of computational efficiency and cost.
Num. of Pro. Adopted method Robust
Time (s) Cost (¥) Time (s) Cost (¥)
5 33.4 1142.9 51.2 1560.7
10 61.2 2903.6 218.1 3717.4
20 324.3 4089.1 1895.5 5131.4
30 587.6 5340.0 3630.8 6058.2

(Cost.2) are shown in Table 5. The total cost of community in


the energy sharing framework is 20.7% less than the community
without cooperation. Note that the total cost of each prosumer
is negative which means that prosumers get profit from the
community because of the participation of CES in energy sharing
community. Considering the relatively large cost gap between
prosumers and CES, the aforementioned Nash bargaining-based
profit allocation is adopted in community. Table 6 shows the costs
and profits of prosumers and CES after allocation. From Table 6,
the energy cost of CES are reduced while the costs of prosumers
are increased. Thus, the individual in community has the same
profit which guarantees incentives for prosumers and CES to join
energy sharing community.

5.4. Discussion on the confidence level

In the previous case study, the confidence level of the proba-


bility distributions α1 and α∞ are set to 0.5 and 0.99 [33]. In this
subsection, the total costs of community are compared and ana-
lyzed by setting different confidence level. The comparison result
is shown in Table 7. Note that as the confidence level increases,
the total cost of energy sharing community also increases. As
shown in Eqs. (64)–(65), the reason is that the upper and lower
Fig. 3. Net demands profiles and sharing profiles. bounds of the confidence constraint set Ω are decreased which
causes the uncertainty interval of PV power larger. Thus, the
regulation costs of prosumers in the second stage increase.
and the change of charging/discharging schedule, if equipped
with the private BESS. For the load profiles, prosumer 1, 2, and 5.5. Comparison results of computational efficiency and cost
3 positively participate in the DR in load peak periods. Prosumer
4 and 5 hardly participate in the DR because of the demand The comparison results of computational efficiency and cost
preference. The solid lines and the dotted lines of the charg- is shown in Table 8, where the two-stage robust optimization is
ing/discharging profiles and net demands profiles coincide when set as the benchmark. With the number of prosumers increasing,
PV output power is zero. Fig. 4(f) shows the SoC of CES, where adopted method has the higher computational efficiency than the
the end SoC is equal to the initial SoC to realize the periodic benchmark. From the comparison results of the cost, the optimal
schedule. More specifically, the energy costs of energy sharing objectives of the benchmark are relatively conservative because
community in the first stage (Cost.1) and in the second stage of the included extreme scenarios.
8
B. Gu, C. Mao, D. Wang et al. Sustainable Energy, Grids and Networks 34 (2023) 101051

Fig. 4. The load, net demands and private BESS power profiles of (a) prosumer 1, (b) prosumer 2, (c) prosumer 3, (d) prosumer 4, (e) prosumer 5, and SoC of CES
(f). The solid lines are profiles in the first stage and the dotted lines are profiles with the regulation in the second stage.

6. Experimental results

The energy sharing community based on the multi-ports EH


is implemented on a testbed and some experimental results are
discussed in this section to validate the feasibility under different
operation conditions.
The corresponding test circuit in the laboratory is illustrated in
Fig. 5. On this testbed, the number of prosumers is set to three.
Each prosumer can be accessed in DC by a buck/boost converter
or AC by an inverter, and CES is accessed in DC by a buck/boost
converter. The rated DC bus voltage is set to 375 V. The rated DC
voltage of CES is 96 V. In Fig. 5, the blue arrows represent the
measurement signals and the pink arrows represent the switch
signals. The energy sharing EH prototype is shown in Fig. 6.
The experimental results are shown in Fig. 7. Fig. 7(a) shows Fig. 5. The test circuit.
the waveforms when multi-ports EH are starting. The whole soft
startup process consists of three stages which ensures the inrush
current in the startup process is within the acceptable range.
Fig. 7(b) and (c) shows the energy sharing waveforms, where the
waveforms of currents of prosumers are omitted and replaced
by the waveforms of current of utility grid. The former one
represents that the net demand of community increases and CES
is charging, where it occurs at the beginning of a day. The latter
one represents that the net demand of community decreases and
CES is discharging, where it occurs at the end of a day. Fig. 7(d)
shows the waveforms when EH is switching operation mode. If
utility grid works normally, the energy sharing community will
work in grid-tied mode. Otherwise, the community will work in
islanded mode, where the DC bus voltage will be supported by
CES. As shown in Fig. 7(d), CES is discharging while the current Fig. 6. Energy sharing energy hub prototype.
of utility grid is 0. The voltage of each prosumer can remain stable
and the switching processes are smooth. Fig. 7(e) and (f) shows
the current and voltage waveforms of utility grid under different waveform of grid current is unchanged due to the purely resistive
reactive power instruction values. The grid current leads the grid load. Also, the voltage of each prosumer can remain stable.
voltage in Fig. 7(e) and the grid current lags the grid voltage
in Fig. 7(f). It illustrates that multi-ports EH has the function of 7. Conclusion
reactive power compensation.
Fig. 8 shows the waveforms of hardware-in-the-loop (HIL) In this paper, we have proposed a two-stage distributed stoc-
simulation to verify that energy sharing community are capable hastic optimization model considering the PV uncertainties for
to work when utility grid is abnormal. In Fig. 8(a), the amplitude energy sharing community, including PV prosumers and CES, to
of grid voltage decreases by 7%. As shown in waveform, the increase community economy. Firstly, we have formulated the
amplitude of grid current increases to satisfy the power balance. day-ahead scheduling model in the first stage to minimize the
The voltage of each prosumer can remain stable. In Fig. 8(b), the social cost of community. Secondly, we have formulated the real-
frequency of grid voltage drops 0.2 Hz. As shown in Fig. 8(b), the time regulation model in the second stage. The confidence set for
9
B. Gu, C. Mao, D. Wang et al. Sustainable Energy, Grids and Networks 34 (2023) 101051

Fig. 7. The experimental results of (a) startup procedure, (b) and (c) energy sharing waveforms, (d) operation mode switching procedure, (e) and (f) reactive power
compensation waveforms.

compared with the cost for individually operation of participants


in community. The hardware implementation of multi-ports EH
used in community under different working conditions has been
verified on the testbed.
In this study, we do not consider the constraints of the elec-
tricity network, for example the power flow constraints in distri-
bution network [34], and the life cycle cost of ESSs [29] is also not
included, which will be further considered in our future work. In
addition, the evaluation of energy sharing community investment
will also be considered.

CRediT authorship contribution statement

Bo Gu: Methodology, Software, Data curation, Visualization,


Writing – original draft. Chengxiong Mao: Conceptualization,
Writing – review & editing, Supervision. Dan Wang: Conceptual-
ization, Writing – review & editing, Supervision. Bin Liu: Concep-
tualization, Writing – review & editing. Hua Fan: Conceptualiza-
tion, Writing – review & editing. Rengcun Fang: Project admin-
istration, Validation, Funding acquisition. Zixia Sang: Validation,
Funding acquisition.

Declaration of competing interest

The authors declare the following financial interests/personal


relationships which may be considered as potential competing
interests: Hua Fan reports financial support was provided by State
Grid Hubei Electric Power Co.

Data availability
Fig. 8. The waveforms of HIL simulation when grid is abnormal.
Data will be made available on request.

References
the probability distribution of PV uncertainties is constructed to
be the constraint set in the second stage, which is used to find the [1] J. Li, et al., How to make better use of intermittent and variable energy?
optimal regulation schedule under the worst probability distri- a review of wind and photovoltaic power consumption in China, Renew.
bution. Furthermore, we have applied the Nash bargaining based Sustain. Energy Rev. 137 (2021) Art. no. 110626.
[2] Y. Wang, J. He, W. Chen, Distributed solar photovoltaic development
profit allocation method for community to share the profit. The potential and a roadmap at the city level in China, Renew. Sustain. Energy
C&CG algorithm is introduced to solve the formulated data-driven Rev. 141 (2021) Art. no. 110772.
two-stage problems. Simulation results demonstrate that the pro- [3] T. Qiu, et al., Potential assessment of photovoltaic power generation in
posed energy sharing model can achieve the lower energy cost China, Renew. Sustain. Energy Rev. 154 (2022) Art. no. 111900.

10
B. Gu, C. Mao, D. Wang et al. Sustainable Energy, Grids and Networks 34 (2023) 101051

[4] R. Dai, R. Esmaeilbeigi, H. Charkhgard, The utilization of shared energy [19] S. Cui, Y.-W. Wang, X.-K. Liu, Z. Wang, J.-W. Xiao, Economic storage sharing
storage in energy systems: A comprehensive review, IEEE Trans. Smart framework: Asymmetric bargaining-based energy cooperation, IEEE Trans.
Grid 12 (4) (2021) 3163–3174. Ind. Inform. 17 (11) (2021) 7489–7500.
[5] D. Kalathil, C. Wu, K. Poolla, P. Varaiya, The sharing economy for the [20] S. Fan, Q. Ai, L. Piao, Bargaining-based cooperative energy trading for
electricity storage, IEEE Trans. Smart Grid 10 (1) (2019) 556–567. distribution company and demand response, Appl. Energy 226 (2018)
[6] Y. Yang, G. Hu, C.J. Spanos, Optimal sharing and fair cost allocation 469–482.
of community energy storage, IEEE Trans. Smart Grid 12 (5) (2021) [21] A. Jiang, H. Yuan, D. Li, Energy management for a community-level inte-
4185–4194. grated energy system with photovoltaic prosumers based on bargaining
[7] V. Bui, A. Hussain, H. Kim, A multiagent-based hierarchical energy man- theory, Energy 225 (2021) Art. no. 120272.
agement strategy for multi-microgrids considering adjustable power and [22] Y. Chen, et al., A comparison study on trading behavior and profit
demand response, IEEE Trans. Smart Grid 9 (2) (2018) 1323–1333. distribution in local energy transaction games, Appl. Energy 280 (2020)
[8] H. Nagpal, I.-I. Avramidis, F. Capitanescu, A.G. Madureira, Local energy Art. no. 115941.
communities in service of sustainability and grid flexibility provision: [23] T. Terlouw, T. AlSkaif, C. Bauer, W.v. Sark, Multi-objective optimization
Hierarchical management of shared energy storage, IEEE Trans. Sustain. of energy arbitrage in community energy storage systems using different
Energy http://dx.doi.org/10.1109/TSTE.2022.3157193. battery technologies, Appl. Energy 239 (2019) 356–372.
[9] W.L. Schram, T. AlSkaif, I. Lampropoulos, S. Henein, W.G.J.H.M. van Sark, [24] F. Hafiz, A.R.D. Queiroz, P. Fajri, I. Husain, Energy management and
On the trade-off between environmental and economic objectives in optimal storage sizing for a shared community: A multi-stage stochastic
programming approach, Appl. Energy 236 (2019) 42–54.
community energy storage operational optimization, IEEE Trans. Sustain.
[25] A. Walker, S. Kwon, Design of structured control policy for shared energy
Energy 11 (4) (2020) 2653–2661.
storage in residential community: A stochastic optimization approach,
[10] S. Van Der Stelt, T. AlSkaif, W. van Sark, Techno-economic analysis of
Appl. Energy 298 (2021) Art. no. 117182.
household and community energy storage for residential prosumers with
[26] S. Cui, Y. Wang, J. Xiao, N. Liu, A two-stage robust energy sharing
smart appliances, Appl. Energy 209 (2018) 266–276.
management for prosumer microgrid, IEEE Trans. Ind. Inform. 15 (5) (2019)
[11] A. Walker, S. Kwon, Analysis on impact of shared energy storage in
2741–2752.
residential community: Individual versus shared energy storage, Appl.
[27] M. Pourakbari-Kasmaei, M. Asensio, M. Lehtonen, J. Contreras, Trilateral
Energy 282 (2021) Art. no. 116172.
planning model for integrated community energy systems and PV-based
[12] P. Huang, Y. Sun, M. Lovati, X. Zhang, Solar-photovoltaic-power-sharing-
prosumers—A bilevel stochastic programming approach, IEEE Trans. Power
based design optimization of distributed energy storage systems for
Syst. 35 (1) (2020) 346–361.
performance improvements, Energy 222 (2021) Art. no. 119931. [28] B. Wang, C. Zhang, C. Li, G. Yang, Z.Y. Dong, Transactive energy sharing
[13] C.P. Mediwaththe, M. Shaw, S. Halgamuge, D.B. Smith, P. Scott, An in a microgrid via an enhanced distributed adaptive robust optimization
incentive-compatible energy trading framework for neighborhood area approach, IEEE Trans. Smart Grid 13 (3) (2022) 2279–2293.
networks with shared energy storage, IEEE Trans. Sustain. Energy 11 (1) [29] B. Gu, et al., Optimal charge/discharge scheduling for batteries in energy
(2020) 467–476. router-based microgrids of prosumers via peer-to-peer trading, IEEE Trans.
[14] H.-C. Chang, B. Ghaddar, J. Nathwani, Shared community energy storage Sustain. Energy 13 (3) (2022) 1315–1328.
allocation and optimization, Appl. Energy 318 (2022) Art. no. 119160. [30] T. Ding, W. Jia, M. Shahidehpour, O. Han, Y. Sun, Z. Zhang, Review of
[15] C.P. Mediwaththe, L. Blackhall, Network-aware demand-side management optimization methods for energy hub planning, operation, trading, and
framework with a community energy storage system considering voltage control, IEEE Trans. Sustain. Energy 13 (3) (2022) 1802–1818.
constraints, IEEE Trans. Power Syst. 36 (2) (2021) 1229–1238. [31] B. Zeng, L. Zhao, Solving two-stage robust optimization problems using a
[16] N. Liu, M. Cheng, X. Yu, J. Zhong, J. Lei, Energy-sharing provider for PV column-and-constraint generation method, Oper. Res. Lett. 41 (5) (2013)
prosumer clusters: A hybrid approach using stochastic programming and 457–461.
stackelberg game, IEEE Trans. Ind. Electron. 65 (8) (2018) 6740–6750. [32] C. Zhao, Y. Guan, Data-driven stochastic unit commitment for integrating
[17] S. Cui, Y.-W. Wang, Y. Shi, J.-W. Xiao, Community energy cooperation with wind generation, IEEE Trans. Power Syst. 31 (4) (2016) 2587–2596.
the presence of cheating behaviors, IEEE Trans. Smart Grid 12 (1) (2021) [33] T. Ding, Q. Yang, Y. Yang, C. Li, Z. Bie, F. Blaabjerg, A data-driven
561–573. stochastic reactive power optimization considering uncertainties in active
[18] H. Zhang, S. Zhou, W. Gu, C. Zhu, Optimized operation of micro-energy distribution networks and decomposition method, IEEE Trans. Smart Grid
grids considering the shared energy storage systems and balanced profit 9 (5) (2018) 4994–5004.
allocation, CSEE J. Power Energy Syst. http://dx.doi.org/10.17775/CSEEJPES. [34] J. Iria, A. Coelho, F. Soares, Network-secure bidding strategy for aggregators
2020.04760. under uncertainty, Sustain. Energy Grids Netw. 30 (2022) Art. no. 100666.

11

You might also like