Energy 239 (2022) 122124

Contents lists available at ScienceDirect

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

Multi-objective optimization design and multi-attribute decision-

making method of a distributed energy system based on nearly zero-
energy community load forecasting
Jiacheng Guo a, Peiwen Zhang a, Di Wu a, Zhijian Liu a, *, Xuan Liu a, Shicong Zhang b,
Xinyan Yang b, Hua Ge c
a
Department of Power Engineering, School of Energy, Power and Mechanical Engineering, North China Electric Power University, Baoding, 071003, China
b
Institute of Building Environment and Energy, China Academy of Building Research, Beijing, 100013, China
c
Concordia University, Montreal, Quebec, Canada

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

Article history: Currently, many scholars have researched distributed energy systems (DES) from different dimensions.
Received 25 May 2021 However, the load forecasting on both source-side and load-side, the optimal design of DES, and the
Received in revised form multi-attribute decision making need further research. Therefore, a novel DES combining solar photo-
28 July 2021
voltaic and hybrid energy storage is proposed. Combining the Monte Carlo method and improved K-
Accepted 18 September 2021
Available online 28 September 2021
means clustering is presented to predict the load on the source-side and load-side. Then, an optimal
design method considering system independence and solar energy utilization scale is proposed, and the
novel system is optimized. The entropy method and TOPSIS method are combined to make the multi-
Keywords:
Distributed energy system
attribute decision on the optimization results of the system. The novel system is used to supply en-
Load forecasting ergy to nearly zero-energy communities (NZEC) in different scenarios. The research results show that the
Solar photovoltaic proposed load forecasting method can more accurately reflect the actual load demand of users. When
Solar fraction “cost - external energy dependence - solar fraction” is taken as the target, the comprehensive energy
External energy dependence import rate with nearly zero-energy office communities is only 36.7%, showing excellent independence.
Hybrid energy storage Finally, this paper can provide a method for load forecasting of NZECs, and provide certain theories for
the optimization and decision-making of DESs.
© 2021 Elsevier Ltd. All rights reserved.

1. Introduction mismatch problem in the time scale of source-load in the energy

supply process of the energy system and meet various energy de-
A series of global issues, such as climate warming and air mands on the demand side [5]. Therefore, combining the DES with
pollution, have attracted growing attention [1]. Reducing green- renewable energy technology and energy storage technology is an
house gas emissions and achieving “carbon neutrality” and “zero effective way to improve the DES comprehensive energy efficiency,
emissions” has become the consensus of many countries [2]. decrease carbon emissions, and the cost of system operation.
Distributed energy systems (DES) can improve energy utilization Combining the DES with wind power or photovoltaic (PV) to
efficiency through energy cascade utilization, combine with form a DES with renewable energy technology is the most common
renewable energy, energy storage, etc., to output multiple energy combination at present [6]. Ref. [7] combines PV, photothermal,
products [3]. However, the traditional DES still has large carbon and water tanks with the DES and provides heat, cold, and elec-
emissions and relatively high investment operating costs. Utilizing tricity load to the community. Ref. [8] couples the DES with PV,
renewable energy technology is greatly important for improving wind power, and lithium batteries (Li-ion) to form a DES with en-
the energy structure and realizing sustainable economic and social ergy storage. In Ref. [9], considering the electric vehicles (EV)
development [4]. Energy storage technology can further solve the charging load of the demand side, batteries store the excess power
generated by the DES. In Ref. [10], biomass energy and PV are
combined with the DES, and the storage of heat and electricity are
* Corresponding author. considered, forming a new DES, and the system is used to commit
E-mail address: zhijianliu@ncepu.edu.cn (Z. Liu). the cold, heat, and electricity load of users. However, the research

https://doi.org/10.1016/j.energy.2021.122124
0360-5442/© 2021 Elsevier Ltd. All rights reserved.
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

Abbreviations t Time/period of time

r Normalized value in the solar irradiance
AHP Absorption heat pump a, b Shape parameters in the Beta distribution
ASHP Air-source heat pump m Mean value
CEIR Comprehensive energy import rate h Efficiency
CO2 Carbon dioxide s Loss coefficient/variance
CO2$ERR Carbon dioxide emission reduction rate Dt Time interval
DES Distributed energy system
EV Electric vehicle Subscript
GB Gas boiler AHP Absorption heat pump
HE Heat exchanger ASHP Air-source heat pump
HES Hybrid energy storage arr Aarrival time
ICE Internal combustion engine buy Purchasing electricity
Li-ion Lithium battery ca Thermal power plant
NSGA-II Non-dominated sorting genetic algorithm II cold Cold
NZEC Nearly zero-energy community DES Distributed energy system
NZEOC Nearly zero-energy office community dep Trip time
NZERC Nearly zero-energy residential community EV Electric vehicle
PESR Primary energy-saving rate env Environment
PER Primary energy ratio GB Gas boiler
PLR Partial load rate g Side
PV Photovoltaic grid Municipal grid
SF Solar fraction heat Heat
SP Separated production ICE Internal combustion engine
in/out Import/export
Variables inv Annual investment cost
A Area Li-ion Lithium battery
AC Annual value cost load Load
B(a, b) Beta function main Maintenance expense
C Cooling load/cooling quantity/cooling power/cost max/min Maximum/minimum
CEIR Comprehensive energy import rate mean Mean
COP Coefficient of performance ng Natural gas
CO2$ERR Carbon dioxide emission reduction ratio nom Nominal temperature
CO2$qeuiv Amount of carbon dioxide emitted ope Annual maintenance cost
E Electricity load/electric quantity/electric power PV Photovoltaic
F Natural gas consumption pun Punishment
f heat recovery efficiency real Actual operation
f(r, a, b) Probability distributio SP Separate production
G Rated capacity sell Electricity selling
L Real-time scene load value solar Solar energy
m Number in the PV panels installed tank Water tank
P Power user User
PESR Primary energy-saving rate waste Waste
PER Primary energy ratio
PLR Partial load rate Superscript
Q Heat load/heat quantity/heat power DES Distributed energy system
Rad Solar irradiance SP Separate production
SOC State of charge 1/2/3 Correlation coefficient
T Temperature

on the novel DES combining PV and hybrid energy storage (HES) cope with extreme situations such as significant fluctuation of user
technologies and considering the charging load of EVs on the de- loads and transmission congestion. Therefore, Ref. [13] focuses on
mand side is still insufficient. the influence of user loads uncertainty on system performance.
DES coupled with renewable energy, energy storage, end EV Ref. [14] proposes the Monte Carlo method and considers the
technologies has the advantages of energy conservation, emission impact of uncertain EV charging load on the energy supply of DESs.
reduction, and environmental optimization, but there are disad- In addition, Ref. [15] considers the influence of source-side uncer-
vantages such as uncertainty, volatility, and intermittence [11]. tainty on the energy supply characteristics of energy systems.
However, most current load studies on communities or regions However, most of the literature only considers the source-side or
superimpose the cold, heat, and electricity loads of a single building load-side uncertainties, and does not fully consider the impact of
to form a load of a community or region, without considering the new loads such as EVs and traffic signal lights on system
difference, uncertainty, and intermittence of loads among different performance.
buildings [12]. When the DES supplies energy to users, it may not In optimizing the DES, considering the interaction between the

2
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

configuration mode of power, renewable energy, and energy stor- 2. System description
age equipment and the operation mode of the energy supply sys-
tem, the comprehensive benefits of the system are jointly The DES is mainly composed of PV, internal combustion engine
determined. System operation problems and system configuration (ICE), air-source heat pump (ASHP), absorption heat pump (AHP),
problems are generally considered a whole in the research, and gas boiler (GB), water tank, Li-ion, and heat exchanger, etc. It can
there are mainly two solutions involved. One is to carry out the produce and provide users with cold, heat, and electricity load. The
collaborative optimization of configuration parameters and oper- system flow chart of DES is shown in Fig. 1. Among them, PV is the
ation parameters; the other is to carry out hierarchical optimization most common renewable energy utilization technology, converting
according to different parameter types. The equipment type and solar radiation energy into electric energy to meet the electricity
capacity of the energy supply system, energy storage/release size, load of users. The ICE by combustion natural gas and combine
etc., of collaborative optimization. However, the collaborative electric motors and generators to produce electricity for users. At
optimization method has a large number of optimization variables the same time, the high-temperature flue gas and jacket water heat
at one time, and it is prone to problems that take a long time to can be recovered through the heat exchanger (HE) to provide heat
solve and fall into local optimal solutions. Therefore, some refer- energy for users. ASHPs can generate cold or heat by consuming
ences have carried out optimization studies at two levels [16] and high-grade electrical power and are usually selected as peak
three levels [17]. However, most of the current studies only shifting equipment to supplement insufficient cold and heat de-
consider system configuration and operation optimization, and mand. The system uses PV, ICE, and municipal grids to provide
there are relatively few studies that consider the influence of electricity loads to community users, ASHP, and EV. Among them,
different optimization objectives on system optimization. priority uses PV supply the user electricity load, and Li-ions store
In terms of the decision-making of DESs, many references adopt excess PV power generation. If there is excess power after storage
the subjective weighting method to determine the weight of each by Li-ion, they will be sold to the municipal grid. The DES electricity
optimization objective, which is subjective of relatively strong [18]. load supply strategy is shown in Fig. 2. The system uses an AHP and
Some references to eliminate the influence of individual subjec- an ASHP to supply cold to users, and the start-up sequence is the
tivity on system decision-making adopt the equal weight method AHP and the ASHP. In addition, the system uses an ICE to recover
to make decisions on the optimization results of the system [19]. On waste heat, water tank, GB, and an ASHP to provide heat load to
this basis, some references use the method of minimum square users.
summation, analytic hierarchy process, entropy method, and
optimal weight method to determine the weight of each optimi-
zation objective [20]. On this basis, most references use the grey
relational analysis method, data envelopment method, and TOPSIS 3. Methods
method, etc., to evaluate the system [21]. However, most of the
decision-making methods in the references only apply to the 3.1. Mathematical model
traditional energy system, and the research on the multi-attribute
decision-making methods of DESs combining renewable energy 3.1.1. Internal combustion engine
technology and energy storage technology is not enough. The ICE has many advantages, such as long service life, high fuel
In summary, the researches on DESs mainly focuse on load cleanliness, energy-saving, and environmental protection. It is
prediction of source-side and load-side, optimization design, and often used as the core power equipment of DESs [27]. The main
multi-attribute decision making, and there are relatively few technical parameters of an ICE include power generation (EICE, kW),
studies on the compromise of three aspects. In addition, the energy power generation efficiency (hICE), natural gas consumption (FICE,
supply objects of the DES are mainly concentrated in single build- kW), and amount of waste heat recovery (QICE, kW). The partial load
ings such as schools, hospitals, office buildings, and residential rate (PLRICE) of the ICE is determined by Equation (1).
buildings [22,23] or communities or regions composed of ordinary
buildings [24]. However, relatively few studies are applied in nearly EICE
PLRICE ¼ (1)
zero-energy communities (NZEC) [25,26]. However, the operational GICE
characteristics of DESs in NZECs have not been fully considered.
Therefore, considering the load prediction on both source-side and
hICE ¼  0:0278,PLR3ICE  0:0975,PLR2ICE þ 0:2233,PLRICE
load-side, optimization design of DES, multi-attribute decision
making, and its application in NZEC need to be further studied. þ 0:2661 (2)
Therefore, this main contributions and innovations of this paper
are as follows: (1) Proposes a novel DES with coupled PV and HES
and adopts the operation strategy of prioritizing solar energy. (2) A
load prediction method combining the Monte Carlo method and
improved K-means clustering method is proposed and compared
with the simple load superposition method. (3) Using the non-
dominated sorting genetic algorithm II (NSGA-II) as a tool, estab-
lish five dimensions of the multi-objective optimization model,
including economy, energy-saving, environmental protection,
renewable energy utilization scale, and system independence are
considered. (4) Combining the entropy method and the TOPSIS
method, a multiple attribute decision-making method for DES is
proposed. (5) Taking nearly zero-energy residential community
(NZERC) and office community (NZEOC) as cases, the optimization
results under the decision are studied and analyzed.

Fig. 1. Flow chart of DES.

3
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

3.1.3. Lithium battery

Li-ion is a relatively mature technology of small power storage
equipment, with long service life, high storage efficiency, and
environmentally friendly advantages [25]. Therefore, in this paper,
the Li-ion is selected as the electric storage equipment of the sys-
tem, and its electric quantity (ELi-ion(t), kWh) stored at time t is
determined by Equation (7).

ELiion ðtÞ ¼ ELiion ðt  1Þ þ PLiion;in ðtÞ,Dt,hLiion;in

 (7)
 PLiion;out ðtÞ,Dt hLiion;out

ELiion ðtÞ
SOCLiion ðtÞ ¼ (8)
ELiion; max

SOCLiion;min  SOCLiion ðtÞ  SOCLiion;max (9)

where ELi-ion(t-1) is the battery stored at the last moment, kWh; Dt

is the time interval, h; PLi-ion,in(t) and PLi-ion,out(t) are the charging
and discharging power of Li-ion at time t, respectively, kW; hLi-ion,in
and hLi-ion,out are the charging and discharging efficiency of Li-ion,
respectively; SOC(t) is the state of charge of Li-ion at time t; ELi-
ion,max is the Li-ion rated capacity, kWh; SOCLi-ion,min and SOCLi-
ion,max are the minimum and maximum state of charge of Li-ion,
respectively.

3.1.4. Water tank

The water tank is the current technology is relatively mature
and widely used a kind of heat storage equipment, with long ser-
vice life, low unit cost and environmentally friendly advantages.
Therefore, the water tank is selected as the heat storage device of
the system in this paper, and the heat (Qtank(t), kWh) stored at time
t is determined by Equation (10).

Qtank ðtÞ ¼ ð1  stank Þ,Qtank ðt  1Þ þ Ptank;in ðtÞ,Dt htank;in

 Ptank;out ðtÞ,Dt htank;out
(10)

Fig. 2. Operation strategy of electricity load supply in DES. 0  Qtank ðtÞ  Qtank;max (11)

where Qtank(t-1) is the battery stored at the last moment, kWh;

stank is the heat loss coefficient of the water tank; Ptank,in(t) and
EICE Ptank,out(t) are the heat storage and heat release power of the water
FICE ¼ (3)
hICE tank at time t, respectively, kW; htank,in and htank,out are the heat
storage and heat release efficiency of the water tank, respectively;
QICE ¼ FICE ,ð1  hICE Þ,fICE (4) Qtank,max is the rated capacity of the water tank, kWh.

where GICE is the rated capacity of the ICE, kW; fICE is the efficiency
3.1.5. Gas boiler
of waste heat recovery of the ICE.
When the heat load of the user is large, the GB is used as the
peaking equipment to supplement the insufficient heat demand.
3.1.2. Photovoltaic panel Assuming that the thermal efficiency (hGB) of a GB is a definite
PVs can convert solar radiant energy into electric energy to meet value, its energy balance is determined by Equation (12).
the users electricity load [28]. The corresponding PV generation
(PPV, kW) is determined by Equation (5). QGB ¼ FGB ,hGB (12)

  where QGB is the heat output of GB, kW; FGB is the natural gas
PPV ¼ 0:15 , APV , 1  0:0045 , TPV;real  25 (5)
consumption of GB, kW.

ðTnom  20Þ,Rad
TPV;real ¼ Tenv þ (6) 3.1.6. Air-source heat pump
800
ASHPs can generate cold or heat by consuming high-grade
where APV is the area of a single PV panel, m2; TPV,real is the actual electric energy, which can be used as peak-regulating equipment
working temperature of a PV panel,  C; Tenv is the around ambient to supplement the insufficient cold or heat demand of the system
temperature,  C; Tnom is the nominal temperature of PV panels,  C; [29]. The cold (COPASHP,cold) and heat (COPASHP,heat) efficiency of the
Rad is the local actual solar irradiance, W/m2. ASHP are assumed to be definite values.
4
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

QASHP ¼ COPASHP; heat ,EASHP; heat (13) " #

m ð1  m Þ
b ¼ ð1  msolar Þ solar 2 solar  1 (18)
ssolar
CASHP ¼ COPASHP; cold ,EASHP; cold (14)
where B(a, b) is a Beta function; r is the normalized value of solar
where QASHP and CASHP are the corresponding heat and cold ca- irradiance in summer at time t; a and b represent the shape pa-
pacity generated by ASHP, respectively, kW; EASHP,heat and EASHP,cold rameters of Beta distribution, respectively; msolar is the mean value
are the electric energy consumed by an ASHP in heat and cooling, of solar irradiance at time t; ssolar is the variance of solar irradiance
respectively, kW. at time t.

3.1.7. Absorption heat pump

AHP is a kind of equipment with high-grade heat energy as a
driving heat source and producing cold capacity [30]. Assuming 3.2.2. Electric vehicle
that the energy efficiency ratio (COPAHP) of the AHP is a setting The daily driving distance of EVs can be modeled as a lognormal
value, its energy balance is determined by Equation (15). distribution function, and its probability density function is defined
as Equation (19) [33]. The travel time and home time of users can be
CAHP modeled as normal distribution functions, which are determined
QAHP ¼ (15) by Equations (20) and (21), respectively.
COPAHP
!
where QAHP is the heat consumed of AHP, kW; CAHP is the corre- 1 ðln x  ms Þ2
fs ðxÞ ¼ pffiffiffiffiffiffi exp  (19)
sponding cold capacity of AHP, kW. 2pss x 2s2s

3.2. User load prediction model 8 0  2 1

>
>
> 1 B t  mdep C

>
> p ffiffiffiffiffiffi exp@  A 0<t  m þ12
The load prediction method combining the Monte Carlo method >
> dep
> 2psdep
< 2s2dep
and the improved K-means method is used to forecast the local
hourly solar irradiance building heat and cold loads and EV fdep ðtÞ¼ 0  2 1
>
>
>
> t 24 m 
charging loads. The process of load prediction using the Monte >
> 1 B dep C
Carlo method is shown in Ref. [31]. The clustering process of the
> pffiffiffiffiffiffi
: 2ps exp@ 
> A mdep þ12<t  24
dep 2sdep
2
improved K-means method is shown in Fig. 3, and the complete
process of load forecasting is shown in Step 2 of Fig. 4. (20)

8 !
3.2.1. Solar irradiance 1 ðt þ 24  marr Þ2
>
> pffiffiffiffiffiffi  ð0 < t  marr  12Þ
The randomness of solar irradiation intensity is a crucial factor >
> exp
< 2psarr 2s2arr
of PV power generation. Although the solar irradiation intensity has
farr ðtÞ ¼ !
random intermittence in the short and long term, it follows the >
>
>
> 1 ðt  marr Þ2
Beta distribution [32], and the data need to be standardized before : pffiffiffiffiffiffi exp  ðmarr  12 < t  24Þ
2psarr 2s2arr
fitting. Its probability distribution (f(r, a, b)) is determined by
Equation (16). (21)

1 where ms ¼ 3.2, ss ¼ 0.88; mdep ¼ 8.92, sdep ¼ 3.24; marr ¼ 17.6,

f ðr; a; bÞ ¼ r a1 ,ð1  rÞb1
Bða; bÞ (16) sarr ¼ 3.4.
ð0  r  1; a  0; b  0Þ

m2solar ð1  msolar Þ 3.2.3. User load

a¼  msolar (17) User loads usually include cold, heat, and electricity loads,
s2solar
which are regarded as following their normal distribution, and its
probability density (f(L)) is determined by Equation (22) [34].

Fig. 3. Improved K-means clustering method flow.

5
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

Fig. 4. Flow chart of load forecasting-multi-objective optimization-multi-attribute decision making.

" #
1 ðL  mload Þ2 m , PPV ðtÞ þ EICE ðtÞ þ Ebuy ðtÞ  PLiion;in ðtÞ
f ðLÞ ¼ pffiffiffiffiffiffi exp  (22) (23)
2psload 2,s2load þPLiion;out ðtÞ ¼ Euser ðtÞ þ EASHP ðtÞ þ EEV ðtÞ þ Esell ðtÞ

where L is the real-time scene load value; mload is the mean of load where m is the number of PV panels installed, pcs; Ebuy is electricity
demand at time t; sload is the standard deviation of the load de- purchased from the municipal grid, kW; EEV is the charging load of
mand at time t. EV, kW; Esell is the sells electricity to the municipal grid, kW.
Heat balance constraint:

3.3. Optimized design model QICE ðtÞ þ QGB ðtÞ þ QASHP ðtÞ þ Ptank;in ðtÞ  Ptank;out ðtÞ
(24)
¼ Quser ðtÞ þ QAHP ðtÞ þ Qwaste ðtÞ
3.3.1. Decision variables
PV is the main power generation equipment in the DES, ICE is where Quser is the user heat load, kW; Qwaste is the heat wasted by
the critical power supplement equipment in the system, and ASHP the system, kW.
is the primary cold and heat supplement equipment in the system. Cold balance constraint:
Energy storage equipment has an important impact on the energy
CAHP ðtÞ þ CASHP ðtÞ ¼ Cuser ðtÞ (25)
efficiency and economy of DES. Therefore, PV number (m), ICE rated
capacity (GICE), ASHP rated heat (PASHP,heat) and cold (PASHP,cold) where Cuser is user cold load, kW.
power, water tank rated capacity (QHTS,max), and Li-ion rated ca-
pacity (ELi-ion,max) are selected as decision variables.
3.3.3. Objective function

3.3.2. Work characteristic constraints (1) Primary energy-saving ratio

In addition to the equipment capacity and operational con-
straints in Section 3.1, the constraint conditions also need to meet Primary energy-saving rate (PESR) is a commonly used index to
the hourly supply and demand balance of cold, heat, and electricity evaluate the energy-saving performance of DES, which can better
energy between DES and users, determined by Equations 23e25. reflect the energy-saving effect of the system and is determined by
Electricity balance constraint: Equation (26). The calculation process is detailed in Ref. [25].
6
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

  performance of the system, determined by Equation (31).

FgSP þ Fca
SP  F DES þ F DES
g ca
PESR ¼ (26) CO2 ,equivSP  CO2 ,equivDES
FgSP þ Fca
SP CO2 , ERR ¼ (31)
CO2 ,equivSP
where FSP g and FSP ca is the primary energy consumed at the
where CO2$qeuivSP and CO2$qeuivDES are the amount of CO2 emitted
on-site of the separated production (SP) system, and by power
by the SP system and DES, respectively, g.
purchase from the municipal grid, respectively, kWh; FDES g, and
FDES ca is the primary energy consumed on-site of DES system and
(5) Comprehensive energy import rate
by electricity purchase from the municipal grid, respectively, kWh.

The energy that a system imports from the outside world can
(2) Primary energy ratio
reflect the degree of independence of the system. When the system
imports less energy from the outside world, the lower the depen-
Primary energy ratio (PER) is another index to evaluate the
dence of the system, the higher the degree of autonomy. Therefore,
energy-saving performance of DES, which is determined by
this study proposes the comprehensive energy import rate (CEIR)
Equation (27). When considering the PER of DES, most literature
to evaluate the independence of the proposed system, which is
only considers the consumption of fossil energy and exaggerates
determined by Equation (32).
the role of solar energy in DES. Some references use solar energy as
fossil energy to reduce the contribution of solar energy to DES. Fbuy þ Ebuy
Therefore, solar energy is converted into a certain proportion of CEIR ¼ (32)
Euser þ EEV þ Quser þ Cuser
fossil energy (Fsolar, kWh) in this paper, which is determined by
Equation (28) [25].
Fbuy ¼ FGB þ FICE (33)
Euser þ Quser þ Cuser
PER ¼ DES (27) where FGB and FICE are GB and ICE consumption of natural gas,
Fg þ Fca DES þ F
solar
respectively, kWh.

fPV (6) Solar fraction

Fsolar ¼ Rad, (28)
fSP ,fgrid
Solar fraction (SF) is the proportion of the total solar power
where Rad is solar irradiance, W/m2; fPV, fSP, and fgrid is PV power generation (EPV, kWh) in the whole power generation of the sys-
generation efficiency, power generation efficiency of thermal po- tem. It reflects the scale of renewable energy utilization of a system
wer plants, and municipal grid transmission efficiency, and also reflects the dependence of the system on the municipal
respectively. grid and the effect of energy-saving and emission reduction to a
certain extent, as determined by Equation (34).
(3) Cost per unit supply area
EPV
SF ¼ (34)
The cost per unit supply area of the system (ACmean, yuan/m2) is EASHP þ Euser þ EEV þ Esell
a commonly used index to evaluate the economy of DES. It is equal
to the annual value cost of the system (ACDES, yuan) divided by the
total energy supply area (Auser, m2), which is determined by
Equation (29). The annual value cost (ACDES) of the system is 3.3.4. Optimization algorithm
determined by Equation (30), and the calculation process is detailed The mathematical programming method represented by the
in Ref. [35]. Levenberg-Marquardt method has the advantages of relatively
strong stability and fast convergence and is suitable for solving
ACDES nonlinear least-squares problems [36]. However, the Levenberg-
ACmean ¼ (29)
Auser Marquardt method has disadvantages such as sensitivity to initial
values, which can easily lead to large randomness in the optimi-
ACDES ¼ Cinv þ Cope þ Cmain þ Cng þ Cgrid þ Cpun  Csell (30) zation process and reduce greatly the reliability of the optimization
results [37]. In addition, this study solves the multi-objective
where Cinv is the initial investment of the system, yuan; Cope is the optimization problem of the energy system rather than the
system labor costs, yuan; Cmain is the system maintenance cost, nonlinear least-squares problem, and the Levenberg-Marquardt
yuan; Cng is the cost of natural gas purchase for the system, yuan; method is not exceptionally applicable here. NSGA-II [38] is one
Cgrid is the power purchase cost of the system from the municipal of the most popular intelligent algorithms, which can effectively
grid, yuan; Cpun is the penalty for the sale of excess electricity to reduce the complexity of solving multi-objective optimization
municipal grids, yuan; Csell is the revenue from electricity sales of problems, and has the advantages of fast running speed, good
the system, yuan. convergence of solution set, and strong robustness. It has been
widely used in solving multi-objective optimization problems of
(4) Carbon dioxide emission reduction ratio energy systems. In addition, NSGA-II is compared with intelligent
algorithms such as genetic algorithm, ant colony algorithm, and
With environmental pollution, greenhouse gas emissions, and multi-objective particle swarm optimization algorithm [39]. A fast
other global problems becoming more serious, countries have non dominated sorting algorithm is proposed to reduce the
proposed a “carbon neutral” schedule, and environmental issues computational complexity, merge the parent population and the
have been more widely concerned. Carbon dioxide (CO2) is a proxy child population, and retain all the best individuals. The elite
for greenhouse gas emissions, so this paper takes the CO2 emission strategy is introduced to retain excellent population individuals
reduction rate (CO2$ERR) as an index to evaluate the environmental and improve the accuracy of the optimal structure. The crowding
7
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

degree and the crowding degree comparison operator are adopted electricity purchased from the grid, equipment operation and
to evenly extend the population to the whole Pareto domain and maintenance costs, carbon emissions, and the output of each
ensure the diversity of the population [40]. equipment are accumulated throughout the year, and the fitness of
all individuals is calculated according to the fitness function. The
3.4. Multi-objective decision making next generation is generated through selection, crossover, muta-
tion, and non-dominated sorting based on parent genes. The fitness
In the decision-making stage, the entropy method and TOPSIS of each new individual is recalculated, and the best individual is
method are combined to make the system multi-attribute decision. selected from the new individual. Until the last generation
The TOPSIS method is a mainly valuable decision-making method, (Gen ¼ 500), the Pareto front solution considering multiple objec-
which is simple, direct, and effective. According to the relative tives is obtained.
approximation of each solution in the Pareto solution set, the Step 4: Multi-criteria decision making
optimal solution is obtained by calculating the distance from the By inputting the Pareto front solution, the eigenmatrix
positive ideal solution and the distance from the negative ideal composed of the Pareto front solution is normalized, and the
solution. The entropy method determines the index weight ac- normalized vector is obtained (rij). The normalized matrix about
cording to the variation degree of each index value, which is an the normalized vector (rij) is established. The weight (wj) of each
objective weighting method and avoids the deviation caused by objective is obtained by using the entropy method, wj is the weight
human factors. Compared with those subjective assignment of the j objective, the ideal solution (A*), and the anti-ideal solution
methods, it is more accurate and objective and can better explain (A) are determined according to the weight normalized value (vij).
the results obtained. For the mathematical models of entropy Calculate the distance scale, that is, calculate the distance from each
method and TOPSIS method, please refer to Ref. [41]. The decision objective to the ideal solution and the anti-ideal solution. Calculate
process is shown in Step 4 of Fig. 4. the ideal closeness degree (C*), when C* i ¼ 0, Ai ¼ A, indicating
that this objective is the worst objective; when C* i ¼ 1, Ai ¼ A*,
3.5. Optimization scheme indicating that the objective is the optimal objective. According to
the value of C*, the evaluation objectives are ranked in order from
Load prediction on both source-side and load-side, optimized the smallest to the largest. The larger the closeness C* value of the
design of DES, and multi-attribute decision-making process is sorting results is, the better the objective is, and the maximum C*
shown in Fig. 4: value is the optimal objective. It is the optimization result of DES
Step 1: Basic parameter setting under this decision.
To ensure the numerical values adopted can ensure that the Step 5: Output optimization results
NSGA-II has the advantages of excellent global search ability, not By analogy, after completing Step 1, Step 2, Step 3, and Step 4, the
easy to fall into the optimal local solution, and inherent implicit optimal objective value considering multiple optimization objec-
parallelism. The crossing probability and mutation probability of tives is output. The corresponding optimization variables m, GICE,
NSGA-II adopted in this paper are 0.8 and 0.2 [40,42], respectively, PASHP,heat, PASHP,cold, ELi-ion,max, and Qtank,max, i.e., the best combina-
the population size is 200, and the number of generations is 500. tion of the system under the compromise multiple objectives. In
Set the technical, economic, and environmental parameters of each the meantime, the optimal hour-by-hour operation state of DES in a
equipment system. The optimization range of decision variables typical year can be determined.
such as m, ELi-ion,max, and PASHP, heat is set, and the initial running
time of the system t is 1 h.
Step 2: Load forecasting
First, the building load simulation software DeST is used to
obtain the hour-by-hour cold, heat, and electricity loads of build- 3.6. Research case
ings. Then, the hour-by-hour solar irradiance of typical years in
Beijing is obtained from the China National Meteorological Science Hour-by-hour outdoor parameters of Beijing in typical years are
Data Center. The historical data of community EV travel is obtained shown in Fig. 5. The maximum temperature in a typical year is
from Ref. [33]. The fitting tool in Matlab is used to fit the above data, 37.2  C, which appears on June 21. The lowest temperature
respectively, and the Monte Carlo method is used to extract the M1 is 14.2  C, which appeares on January 19. The difference between
building heat, cold and electricity load scenario, M2 EV charging them is 51.4  C. This region is a typical cold climate zone, where
load scenario, and M3 solar irradiance scenario. The improved K- buildings need to be cooled in summer and heated in winter. The
means clustering method is used to reduce the scenes and cluster annual solar irradiation amount in this area is 1405 kWh/m2, which
into N1 typical scenario of building cold, heat, and electricity load, belongs to the region with abundant solar energy resources. Among
N2 typical scenario of EV charging load, and N3 typical scenario of them, the maximum instantaneous solar irradiance appeares on
solar irradiance. Finally, these typical scenarios are composed of the April 26, which is 1200 W/m2.
base data of the community. The hour-by-hour cold, heat, and electricity loads of the building
Step 3: Multi-objective optimization are shown in Fig. 6. The indoor environmental parameters, tech-
Import the system parameters of Step 1. Import Step 2 to predict nical indicators, energy efficiency indicators, and technical mea-
hour-by-hour solar irradiance, user cold, heat, electricity, and EV sures of the building align with the technical standard for nearly zero
charging load. Under the given decision variables m, GICE, PASHP,heat, energy buildings (GBT 51350e2019) [43]. The electricity load of
PASHP,cold, ELi-ion,max, and Qtank,max, the energy supply and demand residential buildings and office buildings vary in a cycle of days and
constraints between the system and the user are taken into ac- weeks, respectively. The maximum cold and heat load of residential
count. According to the thermodynamic model of the DES estab- buildings are 78.3 kW and 19.0 kW, respectively, and the former is
lished in Section 3.1, the optimization objective is determined. 4.1 times higher than the latter. The maximum values of the two in
According to the operation scheme shown in Fig. 2, From 0 to office buildings are respectively 287.5 kW and 641.9 kW, and the
8760 h, the power output, operation and maintenance costs, and former is only 0.45 times of the latter. The fluctuation of the cold
carbon emissions of each DES equipment are calculated hour-by- load of residential buildings is much higher than that of heat load,
hour. The hourly natural gas consumed by the system, hourly while the variation trend of office buildings is opposite.
8
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

Fig. 5. Local outdoor meteorological parameters.

Fig. 6. Hour-by-hour cold, heat, and electricity load of the building.

9
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

Fig. 7. Hourly electricity price and gas price in cities.

3.7. System parameter setting Table 2

Technical parameters of equipment [25,27e30].

The price of electricity purchased from and sold to the municipal Item Parameters Value
grid and the price of natural gas purchased from the municipal gas ICE Waste heat recovery efficiency in the ICE (fICE) 0.8
network by Beijing hourly are shown in Fig. 7. The economic and PV A single PV panel area (APV, m2) 2.0
technical parameters of each piece of equipment are shown in Nominal temperature in the PV panel (Tnom,  C) 25.0
Tables 1 and 2. The amount of CO2 emitted by burning natural gas is Li-ion Charging efficiency in the Li-ion (hLi-ion,in) 0.945
Discharging efficiency in the Li-ion (hLi-ion,out) 0.945
203.7 g/kWh, and the amount of CO2 emitted by purchasing elec-
Water tank Heat storage efficiency in the water tank (htank,in) 0.90
tricity from the municipal grid is 886.9 g/kWh [44]. Heat release efficiency in the water tank (htank,out) 0.90
Heat loss coefficient in the water tank (stank) 0.015
GB Thermal efficiency in the GB (hGB) 0.85
4. Results and discussion
ASHP Heat efficiency in the ASHP (COPASHP,heat) 4.0
Cooling efficiency in the ASHP (COPASHP,cold) 3.5
4.1. Comparison of optimization of uncertain and superposition AHP Energy efficiency ratioin the AHP (COPAHP) 0.70
load

The annual hourly solar irradiance, EV charging load, commu- The predicted results of solar irradiance are demonstrated in Fig. 8
nity heat, cold, and electricity load are predicted in this section. (a) and (b). It can be seen from the figure that the local solar irra-
Then, the system studied in this paper is multi-objective optimized diance on typical days varies in a particular range, and the greater
with PESR, ACmean, CO2$ERR, CEIR, and SF as optimization objectives, the solar irradiance at a certain time, the greater the fluctuation. In
and the final optimization results are determined by multi- this paper, the average value of these 30 typical scenes is taken as
attribute decision-making method. the final predicted value of solar irradiance on typical days. Pre-
dicted values of cold, heat, and electricity load of residential
4.1.1. Results of load forecasting buildings are shown in Fig. 8 (c), (d), (e), and (f). The cold, heat, and
The forecast results of residential building cold, heat and elec- electricity loads vary in a particular range, and the change is more
tricity load, and solar irradiance on typical days are shown in Fig. 8. obvious at the peak of energy use, which is characterized by un-
certainty, fluctuation, and intermittence. Therefore, the cold, heat,
and electricity load of a community or region cannot be simply
Table 1
superimposed by a single building load.
Economic parameters of equipment [25,27e30].
In building cold, heat, and electricity load prediction, this paper
Equipment Initial investment Maintenance cost carries out load prediction for residential buildings and office
ICE (-76.21$lnGICEþ1303.3)$F (yuan/kW) 0.0358 (yuan/kWh) buildings, respectively, and on this basis, compositions NZERC and
PV 2160 (yuan/pcs) 0.0385 (yuan/kWh) NZEOC, respectively. The NZERC comprises 40 residential buildings
Li-ion 1650 (yuan/kWh) 0.0189 (yuan/kWh) with six floors, and the NZEOC includes 16 office buildings with six
Water tank 310 (yuan/kWh) 0.0165 (yuan/kWh)
AHP 1500 (yuan/kW) 0.0132 (yuan/kWh)
floors, with an overall floorage of 11600 m2. In addition, EVs in
HE 200 (yuan/kW) 0.0167 (yuan/kWh) NZEOCs are charged at night, and EVs in NZEOCs are charged during
ASHP 970 (yuan/kW) 0.0173 (yuan/kWh) the day. The number of EVs participating in the dispatcher is 200,
GB 620 (yuan/kW) 0.0160 (yuan/kWh) among which EVs are charged every day, and the maximum driving
EV 3000 (yuan/kW) 0.0236 (yuan/kWh)
distance of each EV is 200 km. The rated capacity of the EV battery
Notes: F stands for the exchange rate of US dollar to yuan, take 6.84.

10
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

Fig. 8. Forecast results of typical daily load of residential buildings.

is 30 kWh, and the state of charge of the battery is between 0.2 and cold load in an NZERC is much higher than that of heat load, while
0.98. The hourly prediction results of cold, heat, electricity, and EV the variation trend of NZEOC is opposite. This phenomenon occurs
charging loads in the community are shown in Fig. 9. As seen from because the window-wall ratio of office buildings is greater than
the figure, the electricity load and charging load of EVs in NZERCs that of residential buildings. The activity degree of office building
change in a daily cycle, while in NZEOCs, it changes in a weekly personnel is greater; the thermal insulation requirements of the
process. Because NZERCs need electricity throughout the year, outer envelope of the two types of buildings are also different,
while NZEOCs need electricity only on weekdays. The fluctuation of causing this phenomenon to the occurrence.

11
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

Fig. 9. Community load prediction results.

Table 3 can meet the electricity demand of the community for a day at
Value ranges of decision variables. most.
Item Symbol Value 1 Value 2 The optimization results of various DES devices are shown in
Table 4. “Case 100, “Case 200 , “Case 300 and “Case 400 indicate that the
ICE GICE (kW) [50, 500] [50, 500]
PV m (pcs) [0, 8000] [0, 8000]
load of NZERC and NZEOC is respectively composed of single
ASHP PASHP,heat (kW) [0, 600] [0, 2800] building superposition and load prediction (Monte Carlo method
PASHP,heat (kW) [0, 4500] [0, 10000] combined with improved K-means clustering). Table 4 that the
Li-ion ELi-ion,max (kWh) [0, 6000] [0,10000] rated capacity of the ICE in Case 3 is 124 kW smaller than that in
Water tank Qtank,max (kWh) [0, 3000] [0,3000]
Case 1, while the number of PV is 1104 pcs more. Because the load of
the office community is mainly concentrated during the daytime, it
is synchronized with the solar energy utilization curve. It can
4.1.2. Results of optimization effectively reduce the ICE capacity of the system and improve the
The optimization range of each decision variable is shown in solar energy utilization scale. In addition, under the influence of
Table 3. ‘Value 10 and ‘Value 20 represent NZERC and NZEOC, this factor, the Li-ion rated capacity in Case 3 is reduced by
respectively. Among them, the optimization range of the rated ca- 3433 kWh compared with that in Case 1. The ICE rated power and
pacity of ICE is between 50 kW and 500 kW. The area of each PV is the PV number of Case 2 are 16 kW and 332 pcs more than that of
2 m2. The roofs of each residential building and office building can Case 1, respectively. Because the load predicted by the load is more
be arranged with PV areas of 400 m2 and 1000 m2, respectively. fluctuant and intermittent than that by simple superposition, it is
Two types of the community can be arranged PVs area is 16000 m2 more in line with the actual needs of users. In turn, this has led to
(8000 pcs). So the optimization range of PV panel numbers is greater ICE capacity and an increase in the size of PV installations.
0e8000 pcs. The optimization ranges of ASHP rated cooling and In addition, the system uses ASHPs for heat and cold as much as
heating power are 0 to the maximum community load. The Li-ion possible. Because the ASHP is not only more efficient than the AHP,

12
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

Table 4
Optimization results of each piece of equipment.

GICE (kW) m (pcs) PASHP, cold (kW) PASHP, heat (kW) Qtank, max (kWh) ELi-ion, max (kWh)

Case 1 306 6879 2800 600 2549 5674

Case 2 322 7211 2800 600 3000 5693
Case 3 182 7983 4500 9988 624 2241
Case 4 237 7977 4328 9591 1670 4250

but also can consume the excess power generation of the system 4.2.1. Problem 1: Cost-energy-saving/CO2 emissions
and reduce the amount of primary energy used by the system. In The Pareto solution set of DES for cost-energy-saving/CO2
addition, due to the complex structure of the system, if all the emissions is shown in Fig. 10. As seen from the figure, same con-
equipment of the system is finely modeled, there are still some dition, the NZEOC ACmean is always higher than that of the NZERC
difficulties. As a result, some accessories in the system parameters
are calculated according to the fixed value. The network loss in the
system is directly converted into the corresponding coefficient
calculation. The design of the system operating mode is relatively
ideal, there are few errors with the experimental verification and
actual operating conditions in the system, but the error is
controllable and allowed.
The optimization results of various DES objectives are shown in
Table 5. As seen from the table, the PESR, CO2$ERR, and SF of Case 2
and Case 4 are higher than those of Case 1 and Case 3, respectively.
However, the changing trend of CEIR and ACmean is the opposite.
Because the community load predicted by the load is closer to the
actual load of users, it has better potential to save energy, reduce
emissions, increase the scale of renewable energy utilization, and
reduce the external dependence of the system. For example, the
Case 2 CEIR is decreased by 3.1% compared with that of Case 1, and
the Case 4 CO2$ERR is increased by 6.5% compared with that of Case
3. However, due to the significant fluctuation and the load-
predicted uncertainty, the ACmean system expenditure also in-
creases. For example, the ACmean cost of Case 4 is 5 yuan/m2 more
than that of Case 3. However, in the multi-parameter collaborative
optimization of DES, the initial parameters such as population size,
iteration times, crossover probability, and mutation probability of
NSGA-II have a certain relationship with the system optimization
results, convergence speed, reliability, and accuracy. But these pa-
rameters are considered to be set, so they have certain limitations,
and the calculation results are also slightly error.

4.2. Multi-objective optimization

In this section, a load of the NZERC and NZEOC obtained by the

Monte Carlo method combined with improved K-means clustering
prediction is taken as the case. And multi-objective optimization is
carried out on the proposed system with different objectives,
respectively. The multi-attribute decision-making method is used
to make decisions on the Pareto frontier solution obtained by
optimization.

Table 5
Optimization results of each objective.

PESR (%) CO2$ERR (%) CEIR (%) SF (%) ACmean (yuan/m2)

Case 1 68.4 75.7 47.6 79.8 46.0

Case 2 70.3 77.1 44.5 81.1 46.8
Case 3 62.9 63.6 31.9 68.9 54.9
Case 4 65.7 70.1 30.6 72.5 59.9 Fig. 10. Pareto solution set for “cost-energy-saving/CO2 emissions” as objectives.

13
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

Table 6
Decision-making results of each piece of equipment.

GICE (kW) m (pcs) PASHP, cold (kW) PASHP, heat (kW) Qtank, max (kWh) ELi-ion, max (kWh)

Case 5 402 4944 2800 575 1771 3764

Case 6 458 4740 2800 576 1782 3462
Case 7 483 6886 4413 9728 1052 1220
Case 8 498 6748 4239 1031 1034 1119

Table 7 the objective of “cost-energy-saving” is used more than the result

Decision results of each objective value. of the objective of “cost-CO2 emissions”. In contrast, the trend of
PESR (%) CO2$ERR (%) CEIR (%) SF (%) PER (%) ACmean (yuan/m2) the ICE rated capacity is the opposite. When the objective is
Case 5 56.8 66.1 63.6 60.1 63.0 37.0
economy and energy-saving, the system will increase the scale of
Case 6 55.1 68.1 66.9 58.4 63.2 36.1 solar energy utilization on the one hand and reduce the annual
Case 7 58.0 63.0 56.9 62.9 70.3 50.4 value cost of the system on the other hand, resulting in the
Case 8 52.4 63.1 65.8 70.0 65.6 48.4 occurrence of the above phenomenon. In addition, “cost-energy-
saving” ACmean is higher than “cost-CO2 emissions”, of which Case 7
is 2 yuan/m2 higher than Case 8. As the objective of “cost-energy-
because the NZEOC cold, heat, and electricity load demand are saving”, the system will install more PV and Li-ion, which will in-
higher than that of the NZERC. Furthermore, the maximum load of crease the initial investment of the system to a certain extent.
cold, heat, and electricity in the NZEOC is much larger than that in
the NZERC; the investment of the system is relatively large. In
4.2.2. Problem 2: Cost-energy-saving-solar fraction
addition, the electricity load of the NZEOC is mainly concentrated
The Pareto solution set of DES with the target of “cost-energy-
during 8:00e18:00 of the working day, which is at the peak elec-
saving-solar fraction” is shown in Fig. 11. The variations of PESR, SF
tricity price of the municipal grid, which dramatically increases the
and ACmean in NZERC ranged from 30.3% to 73.0%, 25.6%e73.1% and
electricity purchase cost of the system. In identical conditions, the
26.6 yuan/m2-55.6 yuan/m2, respectively. The variation range of
energy conservation and environmental protection of the NZERC
PESR, SF and ACmean in the NZEOC are 37.3%e76.0%, 38.0%e73.6%
are better than that of the NZEOC. The load of the NZERC is more
and 24.6 yuan/m2-91.0 yuan/m2, respectively. It is seen that the
dispersed. It can play the advantages of energy storage in the DES,
energy-saving effect of the NZEOC is better than that of the NZERC,
improve the energy conservation of the system and environmental
but the cost per unit supply area of the NZEOC is higher. The reason
protection, and effectively reduce the amount of energy waste of
is that the cold and heat load of the NZEOC is more consistent with
the system and CO2 emissions.
the solar energy utilization curve of the system, which improves
The decision results of DES are shown in Tables 6 and 7. Among
the solar energy utilization scale of the system and enhances the
them, “Case 500 , “Case 600 , “Case 700, and “Case 800 represent the de-
energy conservation of the system. However, due to the high unit
cision results of the NZERC and NZEOC in “cost-energy-saving” and
cost of energy storage and PV panels, the cost per unit supply area
“cost-CO2 emissions”, respectively. As seen from Table 6, the
of NZEOC is relatively high. In addition, due to space constraints,
number of PV and the capacity of Li-ion installations determined by
the maximum SF value of the two Cases differs by 0.5%, but the
minimum SF value of the two Cases differs by 12.4%.
The decision results of DES under the objective of “cost-energy-
saving-solar fraction” are shown in Tables 8 and 9. “Case 900 and
“Case 1000 respectively represent the decision results of the NZERC
and the NZEOC. As seen from Table 9, in Case 10, the ICE rated
capacity and the PV number are 257 kW and 1465 pcs more than in
Case 10, respectively, but the Li-ion capacity is 2828 kWh fewer.
Because the cold, heat, and electricity load in the NZEOC is higher
than that in the NZERC, mainly concentrated in the working day
8:00e18:00 period. Increasing the solar scale of the system reduces
Li-ion capacity, leading to an increase in the ICE capacity. PESR,
CO2$ERR, and SF in Case 9 are 3.2%, 5.1%, and 5.7% higher than those
in Case 10, respectively. They indicate that DES has a better effect on
energy conservation and emission reduction when applied in
NZERC. The CEIR of Case 10 is 7.0% lower than that of Case 9, indi-
cating that the system is less dependent on the outside world and
has strong reliability when supplying energy to the NZEOC.

4.2.3. Problem 3: Cost-external energy dependence-solar fraction

The Pareto solution set of DES with the target of “cost-external
energy dependence-solar fraction” is shown in Fig. 12. The variation
range of CEIR, SF and ACmean in NZERC are 35.1%e99.7%, 25.2%e
93.3% and 26.6 yuan/m2-55.0 yuan/m2, respectively. The variation
Fig. 11. Pareto solution set for “cost-energy-saving-solar fraction” as objectives.

14
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

Table 8
Decision results of each piece of equipment.

GICE (kW) m (pcs) PASHP, cold (kW) PASHP, heat (kW) Qtank, max (kW) ELi-ion, max (kW)

Case 9 235 6489 2787 581 2381 5282

Case 10 492 7954 4500 9952 474 2454

Table 9
Decision results of each objective value.

PESR (%) CO2$ERR (%) CEIR (%) SF (%) PER (%) ACmean (yuan/m2)

Case 9 66.6 73.3 47.3 76.2 59.8 43.5

Case 10 63.4 68.2 40.3 70.5 70.1 55.6

relatively small and the distribution is more uniform, the inde-

pendence of DES application in the NZERC is relatively superior. In
addition, in identical conditions, the ACmean of the residential
community is always smaller than that of the NZEOC, which in-
dicates that the economy of the system is better when DES supplies
energy to the NZERC.
The decision results of DES under the objective of “Cost-external
energy dependence-solar fraction” are shown in Tables 10 and 11.
“Case 1100 and “Case 1200 represent the decision result of the NZERC
and NZEOC, respectively. Under this decision, the system will
reduce the ICE capacity as much as possible and increase the PV
number. In Case 11, the rated capacity of the ICE is only 53 kW,
while the PV number in Case 12 is as high as 8000 pcs. The Li-ion
capacity of Case 12 is 3363 kWh less than that of Case 11. When
DES is applied in NZEOC, most of the PV power generation is
directly used by users, and the excess power generation is incor-
porated into the municipal grid as far as possible to reduce the Li-
ion capacity. In addition, the differences of PESR, CO2$ERR, CEIR, SF,
and PER in the two cases are a little different, within 5.0%. However,
the ACmean of Case 2 is 12.1 yuan/m2 more than that of Case 11.
Because the initial investment of NZEOC is higher than that of the
NZERC, the initial investment of DES is relatively significant.
Fig. 12. Pareto solution set for “Cost-external energy dependence-solar fraction” as
objectives.

Table 10
Decision-making results of each piece of equipment.

GICE (kW) m (pcs) PASHP, cold (kW) PASHP, heat (kW) Qtank, max (kWh) ELi-ion, max (kWh)

Case 11 53 6293 2794 600 1296 4979

Case 12 187 8000 4448 10000 582 1616

Table 11 5. Conclusions
Decision results of each objective value.

PESR (%) CO2$ERR (%) CEIR (%) SF (%) PER (%) ACmean (yuan/m2) This paper proposes a novel DES combining solar energy utili-
Case 11 61.4 64.9 42.5 73.8 60.9 42.4
zation, HES, and EV. A load forecasting method combining the
Case 12 61.0 63.0 36.7 69.7 71.5 54.5 Monte Carlo method and improved K-means clustering method is
proposed to forecast solar irradiance, EV charging load, and com-
munity cold, heat, and electricity load are predicted hour-by-hour.
range of PESR, SF and ACmean in NZEOC is 29.0%e86.6%, 21.7%e90.7% On this basis, a new collaborative optimization method is proposed
and 37.9 yuan/m2-75.9 yuan/m2, respectively. The difference be- to optimize the novel system according to different optimization
tween the maximum and minimum CEIR in the NZERC is 64.6%, and objectives. Then, a multi-attribute decision-making method
that in the NZEOC is 57.6%. Because the load of the NZERC is combining the entropy method and TOPSIS method is proposed to
make decisions on the optimization results. Finally, the novel

15
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

system is applied to NZERC and NZEOC, respectively, and the References

comprehensive performance of the system is evaluated. The rele-
vant conclusions are as follows: [1] Wang J, He B-J, Wang H, Santamouris M. Towards higher quality green
building agenda e an overview of the application of green building techniques
in China. Sol Energy 2019;193:473e93.
(1) Compared with the superposition load, the load prediction [2] Tian X, You F. Carbon-neutral hybrid energy systems with deep water source
method combining the Monte Carlo method and the cooling, biomass heating, and geothermal heat and power. Appl Energy
2019;250:413e32.
improved K-means clustering method is closer to the com- [3] Fan Y, Zhao X, Han Z, Li J, Badiei A, Akhlaghi YG, et al. Scientific and techno-
munity or actual cold, heat, and electricity loads of the re- logical progress and future perspectives of the solar assisted heat pump
gion. In addition, the prediction of EV charging load and local (SAHP) system. Energy 2021;229:120719.
[4] Murphy CA, Schleifer A, Eurek K. A taxonomy of systems that combine utility-
solar irradiation intensity increases the diversity of load scale renewable energy and energy storage technologies. Renew Sustain En-
prediction on the source-side and load-side. ergy Rev 2021;139:110711.
(2) Taking CEIR and SF as optimization objectives can more [5] Chen S, Li Z, Li W. Integrating high share of renewable energy into power
system using customer-sited energy storage. Renew Sustain Energy Rev
accurately reflect the independence of the system and the
2021;143:110893.
scale of solar energy utilization. Taking the PESR, ACmean, [6] Tang X, Li G, Zhao X. Effect of air gap on a novel hybrid photovoltaic/thermal
CO2$ERR, CEIR, and SF as optimization objectives, the and thermally regenerative electrochemical cycle system. Appl Energy 2021:
293.
comprehensive performance of the system can be accurately
[7] Chen H, Li Z, Xu Y. Assessment and parametric analysis of solar trigeneration
evaluated from five dimensions of energy-saving, economy, system integrating photovoltaic thermal collectors with thermal energy
environmental protection, independence, and renewable storage under time-of-use electricity pricing. Sol Energy 2020;206:875e99.
energy utilization scale. [8] Al-Ghussain L, Darwish Ahmad A, Abubaker AM, Mohamed MA. An integrated
photovoltaic/wind/biomass and hybrid energy storage systems towards 100%
(3) When taking “cost-energy-saving/CO2 emissions” as the renewable energy microgrids in university campuses. Sustainable Energy
objective, it can improve the energy-saving and environ- Technologies and Assessments 2021;46:101273.
mental protection of the system within a reasonable eco- [9] Bostro € m T, Babar B, Hansen JB, Good C. The pure PV-EV energy system e a
conceptual study of a nationwide energy system based solely on photovoltaics
nomic range. When the objective is “cost-energy-saving- and electric vehicles. Smart Energy 2021;1:100001.
solar fraction”, the system can reduce the annual value cost [10] Ji L, Liang X, Xie Y, Huang G, Wang B. Optimal design and sensitivity analysis
of the system, improve system energy efficiency, and in- of the stand-alone hybrid energy system with PV and biomass-CHP for remote
villages. Energy 2021;225:120323.
crease the solar energy utilization scale of the system. When [11] Kang W, Chen M, Li Q, Lai W, Luo Y, Tavner PJ. Distributed optimization model
taking “cost-external energy dependence-solar fraction” as and algorithms for virtual energy storage systems using dynamic price. J Clean
the objective to ensure the system economy and energy Prod 2021;289:125440.
[12] Mariano-Hern andez D, Hern andez-Callejo L, Zorita-Lamadrid A, Duque-
saving, improve the external dependence tolerance of the rez O, Santos García F. A review of strategies for building energy manage-
Pe
system. ment system: model predictive control, demand side management, optimi-
(4) Using the multi-attribute comprehensive evaluation method zation, and fault detect & diagnosis. Journal of Building Engineering 2021;33:
101692.
combining the entropy method and TOPSIS method, the
[13] Wang J, Zeng P, Liu J, Li Y. Coordinated planning of multi-energy systems
system configuration and operation scheme, which is more considering demand side response. Energy Rep 2020;6:745e51.
suitable for the needs of users, can be determined according [14] Ahn H, Rim D, Pavlak GS, Freihaut JD. Uncertainty analysis of energy and
to the own conditions of users. Namely, the system is the best economic performances of hybrid solar photovoltaic and combined cooling,
heating, and power (CCHP þ PV) systems using a Monte-Carlo method. Appl
combination under the consideration of multiple objectives Energy 2019;255:113753.
and can determine the best hourly running state of DES in a [15] Zhong J, Cao Y, Li Y, Tan Y, Peng Y, Cao L, et al. Distributed modeling
typical year. considering uncertainties for robust operation of integrated energy system.
Energy 2021;224:120179.
[16] Li K, Wei X, Yan Y, Zhang C. Bi-level optimization design strategy for com-
The novel DES proposed in this paper is feasible and reliable pressed air energy storage of a combined cooling, heating, and power system.
when it provides multiple energy sources for different users. The Journal of Energy Storage 2020:31.
[17] Yan Y, Zhang C, Li K, Wang Z. An integrated design for hybrid combined
proposed load forecasting method can provide a method for the cooling, heating and power system with compressed air energy storage. Appl
load forecasting of NZEC. The proposed optimization and decision Energy 2018;210:1151e66.
method can provide some theories for the optimization and deci- [18] Zhang G, Wang J, Ren F, Liu Y, Dong F. Collaborative optimization for multiple
energy stations in distributed energy network based on electricity and heat
sion of the novel DES.
interchanges. Energy 2021;222:119987.
[19] Wu D, Han Z, Liu Z, Zhang H. Study on configuration optimization and eco-
Author contributions nomic feasibility analysis for combined cooling, heating and power system.
Energy Convers Manag 2019;190:91e104.
[20] Solangi YA, Longsheng C, Shah SAA. Assessing and overcoming the renewable
Jiacheng Guo: Software, Writing - Original Draft, Writing - Re- energy barriers for sustainable development in Pakistan: an integrated AHP
view & Editing; Peiwen Zhang: Software, Writing - Original Draft; and fuzzy TOPSIS approach. Renew Energy 2021;173:209e22.
Di Wu: Writing - Review & Editing, Project administration; Zhijian [21] Alzahrani AM, Zohdy M, Yan B. An overview of optimization approaches for
operation of hybrid distributed energy systems with photovoltaic and diesel
Liu: Conceptualization, Methodology, Project administration; Xuan turbine generator. Elec Power Syst Res 2021;191:106877.
Liu: Software; Shicong Zhang: Writing - Review & Editing; Xinyan [22] Wang X, Yang C, Huang M, Ma X. Off-design performances of gas turbine-
Yang: Funding acquisition; Hua Ge: Writing - Review & Editing. based CCHP combined with solar and compressed air energy storage with
organic Rankine cycle. Energy Convers Manag 2018;156:626e38.
[23] Li H, Sun B, Zhang C. Capacity design of a distributed energy system based on
Declaration of competing interest integrated optimization and operation strategy of exergy loss reduction. En-
ergy Convers Manag 2021;231:113648.
[24] Ridha HM, Gomes C, Hizam H, Ahmadipour M, Heidari AA, Chen H. Multi-
The authors declare that they have no known competing objective optimization and multi-criteria decision-making methods for
financial interests or personal relationships that could have optimal design of standalone photovoltaic system: a comprehensive review.
appeared to influence the work reported in this paper. Renew Sustain Energy Rev 2021;135.
[25] Liu Z, Guo J, Wu D, Fan G, Zhang S, Yang X, et al. Two-phase collaborative
optimization and operation strategy for a new distributed energy system that
Acknowledgments combines multi-energy storage for a nearly zero energy community. Energy
Convers Manag 2021;230:113800.
[26] Heendeniya CB, Sumper A, Eicker U. The multi-energy system co-planning of
This work is supported by the National Key R&D Program of
nearly zero-energy districts e status-quo and future research potential. Appl
China (No. 2019YFE0193100).
16
J. Guo, P. Zhang, D. Wu et al. Energy 239 (2022) 122124

Energy 2020:267. nanocatalyst for oxidative desulfurization of sour gas condensate: process
[27] Wu D, Liu Z, Han Z, Zhang H, Ma F, Zhang S, et al. Thermodynamic perfor- parameters optimization based on the LevenbergeMarquardt algorithm.
mance analyses and collaborative optimization for a novel integrated energy Journal of Environmental Chemical Engineering 2021;9(3):105200.
system coupled with organic Rankine cycle. Energy Convers Manag 2020:225. [37] Xia B, Cui D, Sun Z, Lao Z, Zhang R, Wang W, et al. State of charge estimation of
[28] Aghamolaei R, Shamsi MH, O'Donnell J. Feasibility analysis of community- lithium-ion batteries using optimized Levenberg-Marquardt wavelet neural
based PV systems for residential districts: a comparison of on-site central- network. Energy 2018;153:694e705.
ized and distributed PV installations. Renew Energy 2020;157:793e808. [38] Ogbonnaya C, Turan A, Abeykoon C. Energy and exergy efficiencies
[29] Wu W, Zhai C, Sui Y, Zhang H. A novel distributed energy system using high- enhancement analysis of integrated photovoltaic-based energy systems.
temperature proton exchange membrane fuel cell integrated with hybrid- Journal of Energy Storage 2019;26:101029.
energy heat pump. Energy Convers Manag 2021:235. [39] Zhou Y, Cao S, Kosonen R, Hamdy M. Multi-objective optimisation of an
[30] Cui Q, Ma P, Huang L, Shu J, Luv J, Lu L. Effect of device models on the mul- interactive buildings-vehicles energy sharing network with high energy
tiobjective optimal operation of CCHP microgrids considering shiftable loads. flexibility using the Pareto archive NSGA-II algorithm. Energy Convers Manag
Appl Energy 2020:275. 2020;218:113017.
[31] Yan R, Lu Z, Wang J, Chen H, Wang J, Yang Y, et al. Stochastic multi-scenario [40] Ren F, Wang J, Zhu S, Chen Y. Multi-objective optimization of combined
optimization for a hybrid combined cooling, heating and power system cooling, heating and power system integrated with solar and geothermal
considering multi-criteria. Energy Convers Manag 2021;233:113911. energies. Energy Convers Manag 2019;197:111866.
[32] Ju L, Tan Z, Yuan J, Tan Q, Li H, Dong F. A bi-level stochastic scheduling [41] Li B, Miao H, Li J. Multiple hydrogen-based hybrid storage systems operation
optimization model for a virtual power plant connected to a wind- for microgrids: a combined TOPSIS and model predictive control methodol-
ephotovoltaiceenergy storage system considering the uncertainty and de- ogy. Appl Energy 2021;283:116303.
mand response. Appl Energy 2016;171:184e99. [42] Xu C, Ke Y, Li Y, Chu H, Wu Y. Data-driven configuration optimization of an
[33] Lu X, Zhou K, Yang S, Liu H. Multi-objective optimal load dispatch of microgrid off-grid wind/PV/hydrogen system based on modified NSGA-II and CRITIC-
with stochastic access of electric vehicles. J Clean Prod 2018;195:187e99. TOPSIS. Energy Convers Manag 2020;215:112892.
[34] Lindberg KB, Bakker SJ, Sartori I. Modelling electric and heat load profiles of [43] Liu Z, Zhou Q, Tian Z, He B-j, Jin G. A comprehensive analysis on definitions,
non-residential buildings for use in long-term aggregate load forecasts. Util development, and policies of nearly zero energy buildings in China. Renew
Pol 2019;58:63e88. Sustain Energy Rev 2019;114.
[35] Wu D, Zuo J, Liu Z, Han Z, Zhang Y, Wang Q, et al. Thermodynamic analyses [44] Yan J, Broesicke OA, Tong X, Wang D, Li D, Crittenden JC. Multidisciplinary
and optimization of a novel CCHP system integrated organic Rankine cycle design optimization of distributed energy generation systems: the trade-offs
and solar thermal utilization. Energy Convers Manag 2019;196:453e66. between life cycle environmental and economic impacts. Appl Energy
[36] Mahmoudabadi ZS, Rashidi A, Yousefi M. Synthesis of 2D-porous MoS2 as a 2021;284:116197.

17