Professional Documents
Culture Documents
A R T I C LE I N FO A B S T R A C T
Keywords: This study proposes a two-stage stochastic p-robust optimal energy trading management for microgrid, including
Optimal energy trading management photovoltaic, wind turbine, diesel engine, and micro turbine. To achieve optimal energy management for an
Stochastic p-robust optimization microgrid, a hybrid demand response, which combines improved incentive-based and price-based demand re-
Multi-scenario tree method sponses, is incorporated to reduce peak period load while ensuring the reliability of the microgrid. A multi-
Hybrid demand response
scenario tree method is used to generate scenarios for uncertain parameters such as wind turbine, photovoltaic,
Gaussian-based regularized particle swarm
optimization
loads, and market-clearing prices, where each probability density function has been discretized by certain in-
tervals. Then, using a scenario reduction technique, a differential evolution clustering, a set of reduced scenarios
can be obtained. The proposed energy management combines a Gaussian-based regularized particle swarm
optimization with a fuzzy clustering technique to solve the optimization problem and determine the best
compromise solution according to cost-effectiveness and reliability. The effectiveness of the proposed approach
has been analyzed for a typical microgrid test system, and then the results demonstrate that the robustness can
be improved substantially while guaranteeing the economical operation of microgrid. Therefore, the proposed
energy trading management determines the most reasonable solution in terms of economic and reliability issues
for the microgrid operator.
⁎
Corresponding author.
E-mail addresses: glen625@cau.ac.kr (H.J. Kim), mkim@cau.ac.kr (M.K. Kim), jiwon7542@cau.ac.kr (J.W. Lee).
https://doi.org/10.1016/j.ijepes.2020.106422
Received 22 March 2020; Received in revised form 22 July 2020; Accepted 2 August 2020
Available online 12 August 2020
0142-0615/ © 2020 Elsevier Ltd. All rights reserved.
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
Nomenclature Variables
limitations of general clustering algorithm, k-means clustering, and values. To overcome the limitations of pure stochastic optimization, a
reduce the burden. A stochastic economic-environmental optimization risk measurement index, CVaR, has been utilized to guarantee the ro-
for MG energy management was validated by using a modified PSO bustness of stochastic programming [6–9]. A risk-aware stochastic op-
algorithm that considered an IBDR [4]. The authors demonstrated that timization method was proposed in [6] for MG energy trading by in-
considering uncertain parameters gives an MG operator reasonable tegrating the uncertainties into the optimization problem. The risks of
criterion during decision making to account for unprecedented sce- day-ahead and real-time markets of MG were also considered by
narios. Additionally, a DB validity index was utilized to reduce initial adopting the CVaR. A combination of two different scenario reduction
scenarios when solving optimal energy management problems. In [5], a methods, backward reduction and forward selection, was addressed to
scenario-based optimal scheduling of MG was created by considering downsize the initial scenarios. In [7], stochastic-based optimal energy
the uncertainties of PV, WT, and MCP via mixed integer linear pro- management of an energy hub considering the uncertainties of market
gramming. The adopted approach combined a backward method to price and wind speed was presented. Decisions were obtained under a
decrease the initial scenario sets and revealed that considering the risk-aversion strategy by applying CVaR. Moreover, a stochastic-CVaR
uncertain parameters in objective functions can lead to an enhancement optimization solution was provided in [8], where the uncertainties of
in these functions. Although a proper scenario generation and reduction prices and wind power were considered while the CVaR was used to
method was successfully employed, still the economic and security control the potential risk of operating cost increase. The study in [9]
constraints were not fully guaranteed in the optimization problem. As a presented the CVaR-based optimal energy management of battery en-
result, the robustness of this method could deteriorate if obtained ergy storage in a building MG for increasing its resilience while redu-
forecasting errors and samples are widely different from the actual cing operating cost. Although the CVaR-based stochastic optimization
2
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
problem can effectively deal with the risk of uncertain variables and ahead optimal MG energy management is more robust than pure sto-
costs, it requires complete information on the distribution of uncertain chastic programming case with minimal increase in EOC. By in-
variables. In addition, because the results of CVaR vary with respect to corporating a G-RPSO together with the fuzzy clustering method, the
confidence level and risk aversion parameters, these parameters could proposed OETM becomes capable of obtaining the best optimal solution
be suitably determined in advance. between economic and robustness issues through the proper selection
Robust optimization is the latest strategy that optimizes the energy of robustness value p.
trading management problem of MG. This strategy ensures that any The major contributions of this work can be summarized as follows:
nominal sub-optimal solutions remain feasible for any realization of
pre-defined volatility [10-12]. In [10], a day-ahead robust energy • An efficient scenario generation and reduction method is adopted to
scheduling for an MG was adopted by considering uncertainties through model the uncertain variables to ensure that the accuracy and re-
mixed integer non-linear optimization programming. The authors re- liability of the final solutions are close to the optimal real-world
vealed that an adaptive robust strategy led to a lower expected cost and solution. A MSTM is used to generate primary scenarios; then, the
cost variance. A mixed integer robust energy management approach differential evolution algorithm (DEC) is applied for clustering.
was also presented in [11], in which a worst-case scenario for renew- • A robustness factor is adopted in the proposed OETM model, which
able energy and loads was developed to show the robustness of this bounds the relative regret in each scenario when minimizing the
strategy in eliminating model uncertainty. In another study, a multi- EOC of MG.
stage microgrid energy trading management was implemented by ap- • The proposed two-stage stochastic p-robust OETM of MG, which
plying a robust optimization method [12]. The authors demonstrated considers the uncertainties of PV, WT, loads, and MCPs, can present
that this optimization method bolstered the robustness of operating the more robust solution with a minimal increase in EOC.
conditions and increased energy efficiency. However, as the robust • A HDR, which combines I-IBDR and PBDR is implemented to lessen
method optimizes worst-case scenarios, its results may tend to be overly and shift the peak period load of MG, thus reducing the EOC while
pessimistic compared to the stochastic optimization method when sol- ensuring reliability of the MG peak period load.
ving OETM problems of MG. Although dynamic uncertainty sets and • Comparison analyses with respect to pure and CVaR-based sto-
budgets can be implemented to control the level of energy conservation, chastic optimizations are presented, to demonstrate the advantage
systematic procedures are still needed for determining proper con- of the proposed p-robust stochastic OETM.
servation levels.
Another technique for optimizing the day-ahead OETM of MG is The rest of this paper is organized as follows: Section 2 describes the
chance-constrained optimization. This is a competitive tool for solving scenario generation and reduction method to increase the accuracy of
optimization problems and ensures that the likelihood of satisfying a system modeling of MG. In Section 3, the proposed two-stage stochastic
particular constraint is maintained above a certain level [13,14]. A p-robust OETM for MG with mathematical formulations is illustrated.
confidence-based chance-constrained optimization method was em- Section 4 presents the proposed algorithm and overall solution proce-
ployed for day-ahead energy trading of MG by evaluating wind speed dure. Section 5 shows and analyzes the simulation results and, finally,
uncertainty [13]. The chance-constrained method was adjusted to a Section 6 concludes the paper.
stochastic optimization and sample average approximation method and
they were compared to show the effectiveness and robustness of each
2. Uncertainty modelling and evaluation
approach. In [14], an OETM for an islanded MG was represented by
using chance-constrained mixed integer linear programming to mini-
2.1. Uncertainty modelling
mize its operating cost. The authors insisted that this method improved
the optimization results in terms of stability and calculation times by
2.1.1. PV generation modelling
handling multiple uncertainties. Since the chance-constrained method
In most studies, Beta PDF has generally been selected as an appro-
uses a non-Gaussian PDF and tends to be non-convex, its ability to solve
priate model to express the behavior of hourly SI [17]. The relationship
optimal OETM problems may be inadequate for MG. Interval optimi-
between the SI and the PV output power can be obtained as follows:
zation was also employed for day-ahead MG scheduling where un-
certain variables were represented as interval numbers [15,16]. The Γ(α + β )
× SI (α − 1) × (1 − SI )(β − 1) 0 ⩽ SI ⩽ 1, α ⩾ 0, β ⩾ 0
interval optimization method was applied in [15] to address the un- fB (SI )={ Γ(α )Γ(β )
certainty of WT and solve a multi-objective OETM problem of MG. The 0 otherwise
authors demonstrated the benefits of this method to profit intervals (1)
through evaluating WT uncertainty levels. A multi-objective optimal The mean (μ) and the standard deviation (σ) are used to calculate
energy management was also considered in [16] by using the interval the values of the parameters (α and β). These parameters of Beta PDF
optimization method to fulfill the quality, security, and economics of can be obtained through the following equations:
MG. Notably, the interval optimization method is more conservative
but has a lower computational burden than the stochastic optimization μ (1 + μ)
β = (1 − μ) × ( − 1)
method. Meanwhile, compared with the robust optimization method, σ2 (2)
the interval optimization method is less accurate in evaluating OETM of
MG, as robust optimization accounts for worst-case scenarios. μβ
α=
(1 − μ) (3)
1.3. Contribution and paper organization
Once the parameters of Beta PDF are obtained for a specific time
The above-mentioned studies have their own advantages when interval, the output power of PV can be defined as
computing OETM of an MG with uncertainties. However, developing an PPV = η PV × SPV × SI (4)
advanced method in terms of optimality, cost, robustness, and addres-
sing computational challenges is a nontrivial task when applied to the The power outputs from PV are dependent on the efficiency, plane
day-ahead OETM problem. To this end, this study proposes a two-stage area, and SI of the PV module. Since the hourly power output of PV is a
stochastic p-robust OETM method for MG that is financially economical, stochastic variable in the MG energy trading problem, the continuous
adjustably robust, and computationally viable, and considers un- Beta PDF parameters can be divided into several discrete intervals to
certainties for PV, WT, load, and MCP. The proposed approach for day- model the SI and its associated probability in different states.
3
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
4
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
the obtained results can be easily extended to more universal cases. The ⎧ γs (t ) × Ptr , s (t ) Ptr , s (t ) > 0
p-robustness measurement can be combined with the minimization of Ctr , s (t ) = 0 Ptr , s (t ) = 0 ∀ s, t
the EOC function as an objective, which can be described as ⎨
⎩ − γs (t ) × Ptr , s (t ) Ptr , s (t ) < 0 (24)
min ∑ ρs Zs (X )
s ∈ NS (16) Cp, s (t ) = ∑ (Cs, e δs, i, e) × ps,i (t )+ ∑ (Cs, e δs, j, e) × ps,j (t )
e e
which is subject to + ∑ (Cs, grid δs, grid, e) × ps,grid (t ) ∀ s, t
e (25)
Zs (X ) ⩽ (1 + p) Zs∗ s ∈ NS, X ∈ χ (17)
y k
N N
Equation (16) refers to the minimization of EOC while equation (17) CDR, s (t ) = ∑ ∑ λsDR k
, y, k PDR, s, y (t ) ∀ s, t
guarantees the stochastic p-robust condition. For any solution, the ob- y=1 k=1 (26)
jective value of each scenario s must be smaller than (1 + p)Zs*. The
constraints, as described in equations (14), (15), and (17), force the The total EOCs for time t can be computed as the sum of operating
objective function to search for the solution that is feasible under any costs for DE, MT, procurement costs, pollutant treatment costs, and
existing scenario while obtaining the minimum EOC of MG. costs from I-IBDR as described in equation (21). Equations (22) and
To ensure the existence of a feasible solution for the day-ahead (23) refer to the operating costs of DE and MT, respectively. The
OETM problem, the limit of the relative regret p cannot be infinitesimal. transactional power cost between the main grid and MG is explained in
Consequently, determining a LB for the relative regret limit is essential. equation (24). The power trading coefficient of scenario s, γs(t), is as-
When p reaches its lower bound, the relative regret for the worst-case sumed to be the value of MCP [24]. The pollutant treatment costs of DE,
scenario reaches the lowest level, thus achieving the day-ahead OETM MT, and the main grid for e type (CO2, NOx, and SO2) is given in
in the worst case. The constraint (17) must be satisfied, even in a worst- equation (25) [25]. According to equation (26), the cost for I-IBDR can
case scenario. The LB of p is the smallest value that guarantees the be described as a multiplication of the bid price and the amount of
existence of a feasible solution to the day-ahead OETM of MG. In this power curtailed.
regard, the LB of the relative regret limit p can be obtained as:
3.2.2. Constraints
LB = min p
s. t Zs (X ) ⩽ (1 + p) Zs∗ ∀ s • p-robustness constraint
p⩾0 (18)
As mentioned in equation (17), the p-robustness measure should
When the relative regret limit p is large enough, constraint (17) meet the following equation:
relaxes, resulting in a pure stochastic optimization problem. Thus, it is
also necessary to set an upper bound for the relative regret limit that EOCs ⩽ (1 + p) EOCs∗ , ∀ s (27)
enables the EOC of MG to reach its minimum value as closely as pos- Since in this case, p is equal to infinity, equation (27) is relaxed and
sible without constraint (17) in the day-ahead OETM problem. Equation the p-robustness optimization problem (equation (20)-(26)) becomes a
(17) ensures that the EOC of each scenario s is less than or equal to conventional stochastic optimization problem. Noted that EOCs* is the
(1 + p)Zs*, and the UB of the objective function can then be defined minimum operating cost for each s scenario. By reducing p, we can
with an extra constraint, Z = Zmin: obtain a single solution that demonstrates a low regret and a high op-
UB = min p erating cost.
Zs (X ) ⩽ (1 + p) Zs∗
• Power balance constraint
s. t
Z = Zmin
p⩾0 (19)
The sum of the total output power of DERs (DE, MT, PV and WT)
and transferred power between the main grid and MG should be equal
to the estimated load with DR.
3.2. Problem formulation
Ni Nj Nq Nu
5
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
Psmin max
, MT , j (t ) δs, MT , j (t ) ⩽ Ps, MT , j (t ) + ΔPs, MT , j (t ) ⩽ Ps, MT , j (t ) δs, MT , j (t ) ∀ s , j , t customers is also included, in that the incentive paid to the customers
(34) has to be at least equal to their cost of the participating DR. Here, the
loads are considered as an aggregate of individual loads that could be
Equations (29)-(32) ensure that the power generation from DE, MT, classified into different types based on their willingness for DR parti-
PV, and WT, respectively, should be more than their minimum outputs cipation. Customer types can range from 0 to 1, where 0 refers to the
but smaller than their maximum outputs. Equations (33) and (34) de- least willing customer whereas 1 refers the most willing customer for
note that the DE and MT ramp rate limits are not violated. DR participation [27].
Meanwhile, during load reduction, the cost occurs to the customers
• Power exchange constraint for their level of power consumption PkDR can be described as follows:
In the current work, the HDR, combining PBDR and I-IBDR, is im- CRE , y (t ) = λsDR k k
, y, k PDR, s, y (t ) − c (θy , PDR, s, y (t )) ⩾ 0 ∀ s , k , y , t (44)
plemented primarily to reduce the peak period load when solving
OETM of MG. The cost of generating and purchasing energy is higher in
Nt
the peak period, compared with the off-peak period. Therefore, the
application of HDR can lead to EOC reduction, while securing the re-
∑ (λsDR k
, y, k PDR, s, y (t )
k
− c (θy , PDR, s, y (t ))) ⩾ 0 ∀ s , k , y , t
t=1 (45)
liability of the MG peak period load.
The main purpose of PBDR, based on time-of-use rates, is to shift
Nt
some percentage of demand from the peak period to off-peak periods to
∑ (λsDR k k
, y, k PDR, s, y (t ) − c (θy , PDR, s, y (t ))) ⩾
flatten the load profile and lessen the EOC of MG [26]. The mathe- t=1
matical model and its corresponding constraints of time-of-use can be Nt
expressed as ∑ (λsDR k−1
, y, k − 1 PDR, s, y (t )
k−1
− c (θy , PDR, s, y (t ))) k ≠ 1, ∀ s , k , y , t
t=1 (46)
Ls (t ) = (1 − DRs (t )) × LsBase (t ) + PDR, s (t ) (36)
N t
N t ∑ ∑ λsDR k
, y, k PDR, s, y (t ) ⩽ CMy ∀s
(48)
∑ PDR,s (t )= ∑ DRs (t ) × LsBase (t ) t=1 k=1
6
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
7
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
4.1. Gaussian-based regularized PSO Equations (49) and (50) represent the velocity and position up-
dating formula, respectively. For solving the stochastic p-robust OETM
The PSO [28] is a meta-heuristic optimization algorithm that is of MG via the PSO algorithm, improvements are needed as a conven-
based on the swarm intelligence of particles that share their explora- tional PSO tends to fall into local optima or converge prematurely when
tions among themselves. We assume that the z-th particle in N dimen- the dimension of the problem increases. Thus, the velocity of the par-
sional space xP,z=(xP,z1, xP,z2, ⋯, xP,zN) has a position vector pP,z = ticles should be adjusted to overcome the limitations of the conven-
(pP,z1, pP,z2,⋯, pP,zN) and velocity vector vP,z = (vP,z1, vP,z2, ⋯, vP,zN). tional PSO, ensure global optima, and maintain the speed and simpli-
The local and global optimal solutions for iteration b can be denoted as city of the algorithm. In this work, we propose the G-RPSO with the
pP,best,z,n(b) and gP,best,z,n(b), respectively. According to pP,best,z,n(b) and following equations:
gP,best,z,n(b), the particles’ velocity and position will change and are
1 ⎛ ⎛ ⎛ b ⎞l ⎞
updated via the following expressions: wP = ⎜cos ⎜ π ⎟ + 1⎞⎟
2 ⎝ ⎝⎝ Nb ⎠ ⎠ ⎠ (51)
vP, z, n (b + 1) = wP × vP, z, n (b) + c1 × r1 × (pP, best , z, n (b) − xP, z, n (b))
b
+ c2 × r2 × (gP, best , z, n (b) − xP, z, n (b)) (49) c1 (b) = (c1, final − c1, initial ) + c1, initial
Nb (52)
8
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
vP, z, n (b + 1) = wP × vP, z, n (b) + c1 (b) × N (0, σz2) 5.1. Data and description
× (pP, best , z, n (b) − xP, z, n (b)) + c2 (b) × N (0, εz2)
To demonstrate the superior performance and validity of our work,
× (gP, best , z, n (b) − xP, z, n (b)) (56) the grid-connected MG test system has been used as a case study. Fig. 3
shows the structure of the MG test system, which consists of 10PV,
10 WT, 5DE, 2MT, and five responsive loads. The characteristics of PV
4.2. Fuzzy-clustering method
and WT are listed in Table 1. The hourly forecasted wind speed and
corresponding standard deviation are taken from [29,30] and the pre-
In the proposed stochastic p-robust OETM problem, the best solution
dicted mean and standard variance of SI are listed in Table 2 [17]. The
that consists of trade-off between MRR and EOC is chosen. In this re-
techno-economic parameters of DE and MT are presented in Table 3
gard, a fuzzy-clustering method [29] is applied in this study to choose
[29,31]. Table 4 presents the cost coefficient and daily DR limit of each
the final optimum solution among a Pareto optimal set. An appropriate
customer [29] as per five responsive loads. The characteristics of k-step
membership function is defined for both EOC and MRR as follows:
wise price-quantity data for I-IBDR customers are illustrated in Table 5
⎧ 1 fτ (ϕ) ⩽ f τmin [32]. To implement the PBDR, the maximum rate increase and decrease
⎪ f min − f are set to 20 percent. Discharge coefficients of pollutants from DE and
τ (ϕ)
μτ (ϕ) = τ
f τmin ⩽ fτ (ϕ) ⩽ f τmax MT were adopted from [28]. We assumed that the MG operator’s TB is
⎨ fτmax − fτmin
⎪ $2000.
0 fτ (ϕ) ⩾ f τmax (57)
⎩ The G-RPSO algorithm was applied to solve the proposed OETM
In considering both EOC and MRR to obtain the best solution, the problem, whose performance changed according to the particles’
overall satisfying value can be described as number of iterations and sizes. Hence, 100 and 20,000 for the number
of maximum iterations and particles, respectively, were adopted. The
∑ wF , τ μτ (ϕ)
τ
initial and final values of learning factors were set to
μ (ϕ) = c1,initial = c2,final = 2.5 and c1, final = c2,initial = 0.5. The fuzzy-clustering
∑ ∑ wF , τ μτ (ϕ)
ϕ τ (58) method was also utilized to select the best compromise solution against
the trade-off between the economic and robustness issues. The con-
The weight value of wτ can be determined based on the importance
ventional PSO and regularized PSO (RPSO) [28] were compared to
of cost and robustness issues. Finally, the highest value of membership
show the effectiveness and robustness of our proposed approach. In
function μ(ϕ) with respect to others can be the best operating solution.
addition, five cases have been considered to reveal the beneficial of
proposed OETM in terms of peak load management and EOC. The si-
4.3. Solution procedure
mulation was performed in Matlab R2019b on a 3.4 GHz CPU with
16 GB RAM and a Windows 10 Pro 64-bit operating system.
The overall procedure of the proposed stochastic p-robust OETM of
MG can be summarized as follows:
9
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
Table 2 EOC and MRR has been then obtained. When the robustness becomes
Mean and standard deviation of SI. smaller than 0.0175, in this case constraint (27) bounds the relative
Hour μ σ regret of each scenario. By reducing the value of p with 0.0005, the EOC
and the corresponding MRR can be updated. When p reaches 0.001, we
8 0.307 0.182 modified the decrement from 0.005 to 0.001. Furthermore, new EOC
9 0.406 0.217
and MRR values are achieved until p reaches 0.0006. As can be seen in
10 0.522 0.253
11 0.575 0.273
the second and fourth columns, the MRR is decreased while EOC is
12 0.576 0.284 increased when the robustness factor decreases. This means that the
13 0.567 0.282 robustness of MG shows a negative correlation with p. Meanwhile, the
14 0.561 0.235 third and fifth columns represent the reduction and increase ratios of
15 0.520 0.237
MRR and EOC, respectively. By setting p = 0.0006, for example, a
16 0.454 0.204
17 0.391 0.164 substantial reduction have been occurred for the MRR (96.971%) with
18 0.294 0.152 only a minimal increase (1.7098%) with EOC value. These results show
that by adjusting the robustness factor, the proposed approach can
present the more robust solution with small increase in EOC for MG
Table 3 operator to assist in a decision making with respect to economic and
Technical characteristics of DE and MT. robustness issues. No feasible solution could be found for p ≤ 0.0005,
Parameters Unit
Pmin(kW) 30 33 27 20 37 10 10
Pmax(kW) 150 143 120 80 60 55 30
UR(kW/h) 80 60 60 40 40 25 20
DR(kW/h) 80 60 60 40 40 25 20
a($/h) 0.6 0.5 0.8 0.07 0.4 2.62 2.55
($/kWh) 0.05 0.053 0.068 0.25 0.5 0.015 0.02
c($/kW2h) 0.02 0.12 0.112 0.03 0.06 0.015 0.02
Table 4
Customer cost function and daily DR limit.
y K1,y K2,y θy CMy(kWh)
Table 5
Bid price quantity for I-IBDR participants.
Off-peak period Peak period
10
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
(a) PV
(b) WT
(c) Load
Fig. 5. Hourly forecasted results.
which means that existing scenarios could not satisfy the MRR con- robustness factor p on the results of the proposed OETM for grid-con-
straints. nected MG. In the stochastic programming case (p = ∞), the results
Fig. 6 represents the comparative analysis of the effect of the show that the EOC is widely distributed from $ 14,120 to $ 14,480.
11
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
(d) MCP
Fig. 5. (continued)
12
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
Table 7
The HDR program results.
HDR results
Customer (y) DR volume (kWh) Incentives ($) Customer cost ($) Customer benefit ($) Amount of load shifting (kWh)
13
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
running time is obtained by G-RPSO (7 m 50 s), followed by G-RPSO To elucidate the advantageous nature of the proposed OETM ap-
(11 m 30 s) and PSO (15 m 01 s). Because of the effect of the modified proach against CVaR-based stochastic optimization, the obtained re-
inertia weight, acceleration parameters, and added Gaussian-random sults are compared in Table 9. In the stochastic CVaR optimization,
variables, the proposed G-RPSO can provide the lowest convergence confidence level α is set to 0.9 and risk-averse solutions are obtained by
iteration and CPU time, among others. Certainly, these characteristics varying the risk aversion parameter (β) from 0 to 1. Details of the CVaR-
reveal the benefits of the proposed algorithm in solving the day-ahead based stochastic optimization are described in Appendix A. The sto-
OETM problem of MG. chastic CVaR results illustrate that as the level of β increases, the CVaR
In order to show the effects of the HDR on the proposed approach decreases, while the EOC of MG increases. Both stochastic optimiza-
and to reduce the EOC of grid-connected MG, five cases have been tions can sufficiently compute the loss of tail and appropriately deal
considered: with numerous samples along with convenient operations. Never-
theless, the stochastic variables, which are related to the risk mea-
Case 1: The OETM without considering DR surement CVaR are highly dependent on the accuracy of the distribu-
Case 2: The OETM with conventional incentive-based DR tion data of uncertain parameters. Due to limited data on renewable
Case 3: The OETM with PBDR sources, end users, and MCP in MG system, CVaR has difficult to pro-
Case 4: The OETM with I-IBDR vide precise presentation on risk in most practice when applied to real
Case 5: The OETM with HDR and large MG system. Moreover, the results of CVaR vary according to
the risk parameters; the MG operator must estimate the confidence level
Fig. 9 shows the result and performance of the proposed OETM in as well as the risk aversion parameter when considering the risk aver-
the day-ahead MG operation for five different cases. As mentioned in sion level. In this regard, a suitable risk parameter must be chosen for
Table 6, it can also be also shown in Fig. 9 that the minimal operating the MG operator for day-ahead OETM. On the other hand, the proposed
cost was obtained at p = ∞, but the robustness of MG was not fully stochastic p-robust OETM is beneficial in terms of easy implementation,
guaranteed. Meanwhile, the maximum robustness of MG was observed i.e., by simply bounding the relative regret limit (p) in each scenario,
at p = 0.0006, and no feasible solution could be found at a value the robustness of MG is guaranteed, while ensuring economic opera-
smaller than 0.0005. The best operating solution is represented at the tion. Additionally, the increasing computational complexity of the
11th point which is indicated in the red square; the EOC and MRR added CVaR measurement is avoided, thus enhancing the effectiveness
values are $ 14,211 and 0.002292, respectively. Note that the weight of the proposed approach to solve the day-ahead OETM.
values of wτ for cost and robustness were set to 0.5. As shown in Fig. 9,
the superiority of our proposed OETM is evident in terms of the EOC.
6. Conclusion
The EOC values of the optimal operating point obtained from fuzzy
clustering for cases one to five are $16,002, $15,773, $15,301, $15,072,
In this work, an advanced framework was proposed based on the
and $14,211, respectively. Although Case 2 to 4 present a decreased
two-stage stochastic p-robust OETM model considering HDR. Uncertain
EOC with respect to Case 1, the benefit of implementing HDR is realized
variables, such as PV, WT, loads, and MCP, were considered in the
in Case 5. Certainly, by varying incentives according to the period and
proposed OETM of MG. The continuous PDF of each uncertain variable
demand shifting, the proposed OETM can flatten the demand when the
was divided into certain segments to generate primary scenarios by
cost of energy is higher. In other words, when the generating and
implementing MSTM. Then, the DEC algorithm was applied to reduce
transactional costs are high in the peak period, Case 5 can reduce the
the scenario sets and obtain the typical scenarios. The problem was
EOC significantly with respect to cases 1 to 4 by reducing the peak
formulated as minimizing the value of EOC while ensuring the p-ro-
period load as well as selling the remaining power to the main grid, as
bustness constraints for each scenario. The proposed OETM problem
already depicted in Fig. 7.
was solved by the G-RPSO, and, the fuzzy clustering was then employed
Fig. 10 illustrates the new load profile of MG considering each case.
to obtain the best compromise solution by a proper choice of p that
In Case 1, the base load of MG is given. Because the incentive rate is the
reduced the EOC while simultaneously improving the robustness. The
same regardless of the period, Case 2 shows a similar amount of load
comparative results between pure stochastic programming and the
curtailment at all hours, rather than a greater peak period reduction. In
proposed stochastic p-robustness method confirmed that the robustness
Case 3, 218.97 kWh of load has been shifted from peak period to off-
peak period. Case 4 decreases the demand mainly in the peak period
since the incentive payment differs according to the period. It is evident
that the most flattened profile of MG is achieved in Case 5, owing to the
implementation of both parts of the HDR. These results demonstrate the
superiority of the proposed approach because Case 5 can reduce the
EOC while improving the reliability of MG by effectively flattening the
demand of MG.
Table 8 summarizes the effectiveness of the proposed OETM in
terms of economic and peak load management issues. It can be seen
from Case 5 that it shows a beneficial result in terms of the EOC because
substantial reduction in the peak period load is implemented. The
quantitative results of the peak to average ratio also demonstrate the
superiority of our proposed approach. Note that the peak to average
ratio was calculated as the total demand divided by the peak load. In
Case 5, the peak to average ratio increased up to 22.3, which represents
a better performance in terms of peak load reduction implemented by
HDR. This lies in fact that although Case 2 to 4 have also reduced the
EOC and ensured the reliability of the peak period demand, Case 5
provides the most reasonable OETM solution for the MG operator.
Therefore, it should be noted that the proposed OETM provides a more
reliable operating condition by peak load reduction at the same time,
enables operating MG with much reduced cost. Fig. 9. Trade-off between MRR and EOC.
14
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
Appendix
To assess the risk, two risk control measurements, VaR and CVaR, are realized. VaR indicates the probability of loss excess, while CVaR reflects
the expectation of loss excess. In this regard, CVaR is widely used for risk control measurement that utilizes fat tailed loss distribution functions. By
combining risk measurement with CVaR, the objective function of CVaR-based OETM can be expressed as
Nt Ns
min ∑ ∑ ρs EOCs (t ) + βCVaR α
t=1 s=1 (A.1)
where
Ns
1
CVaR α = ς −
1−α
∑ ρs ηs
s=1 (A.2)
Table 9
Comparison of stochastic results for OETM.
Stochastic p-robustness solution Stochastic CVaR solution
Min Max Min Max Average β=0 β = 0.5 β=1 β=0 β = 0.5 β=1 Average
0.000131 0.019116 14,142 14,388 7 m 50 s 14,142 14,232 14,382 15,174 13,718 13,122 8 m 21 s
15
H.J. Kim, et al. Electrical Power and Energy Systems 124 (2021) 106422
Here, ζ and ηs are auxiliary variables. In equation (A.1), the objective of the function is to minimize the EOC, while maintaining the risk at an
acceptable level. For the risk measurement expressed in equation (A.2), the CVaRα at the confidence level (α) is utilized. β is a risk aversion
parameter that represents the MG operator’s risk-aversion attitude. Moreover, the objective function (A.1) is subject to constraints (28)–(47).
References 2014;63:523–33.
[18] Wang MQ, Gooi HB. Spinning reserve estimation in microgrids. IEEE Trans Power
Syst 2011;26:1164–74.
[1] Mahmoud MS, Hussain SA, Abido M. Modeling and control of microgrid: an over- [19] Liu C, Wang X, Guo J, Huang M, Wu X. Chance-constrained scheduling model of
view. J Franklin Inst 2014;351:2822–59. grid-connected microgrid based on probabilistic and robust optimisation. IET
[2] Khodaei A, Bahramirad S, Shahidehpour M. Microgrid planning under uncertainty. Gener, Transm Distrib 2018;12:2499–509.
IEEE Trans Power Syst 2015;30:2417–25. [20] Jadid-Bonab M, Dolatabadi A, Mohammadi-Ivatloo B, Abapour M, Asadi S. Risk-
[3] Zareipour H, Bhattacharya K, Canizares CA. Electricity market price volatility: the constrained energy management of PV integrated smart energy hub in the presence
case of Ontario. Energy Pol 2007;35:4739-48. of demand response program and compressed air energy storage. IET Renew Power
[4] Mostafa S, Masoud E, Atefeh J, Mohammad-H G. Stochastic multi-objective eco- Gener 2019.
nomic-environmental energy and reserve scheduling of microgrids considering [21] Sachs J, Sawodny O. Multi-objective three stage design optimization for island
battery energy storage system. Electr Power Energy Syst 2019;106:1–16. microgrids. Appl Energy 2016;165:789–800.
[5] Mosayeb B, Rahmat-A H, Amin K, Moein R. Optimal stochastic scheduling of CHP- [22] Das S, Abraham A. Automatic clustering using an improved differential evolution
PEMFC, WT, PV units and hydrogen storage in reconfigurable micro grids con- algorithm. IEEE Trans Syst 38:218-237.
sidering reliability enhancement. Energy Convers Manage 2017;150:725–41. [23] Snyder LV, Daskin MS. Stochastic p-robust location problems. IIE Trans
[6] Fazlalipour P, Ehsan M, Mohammadi-Ivatloo B. Risk-aware stochastic bidding 2006;38:971–85.
strategy of renewable micro-grids in day-ahead and real-time markets. Energy [24] Kim MK, Nam YW, Park JW. Market-clearing for pricing system security based on
2019;171:689–700. voltage stability criteria. Energy 2011;36:1255–64.
[7] Najafi A, Falaghi H, Contreras J, Ramezani M. Medium-term energy hub manage- [25] Hongbin W, Wingyue L, Ming D. Dynamic economic dispatch of a microgrid:
ment subject to electricity price and wind uncertainty. Appl Energy Mathematical models and solution algorithm. Int J Electr Power Energy Syst
2016;168:418–33. 2014;63:336–46.
[8] Wang Y, Yang Y, Tang L, Sun W, Zhao H. A stochastic-CVaR optimization model for [26] Nojavan S, Qesmati H, Zare K, Seyyadi H. Large consumer electricity acquisition
CCHP micro-grid operation with consideration of electricity market, wind power considering time-of-use rates demand response programs. Arab J Sci Eng
accommodation and multiple demand response program. Energies 2019;12:3980. 2014;39(12):8913–23.
[9] Tavakoli M, Shokridehaki F, Akorede MF, Marzband M, Vechiu I, Pouresmaeil E. [27] Fahrioglu M, Alvarado FL. Using utility information to calibrate customer demand
CVaR-based energy management scheme for optimal resilience and operational cost management behaviour models. IEEE Trans Power Syst 2001;16: 317–22. Zifa L,
in commercial building microgrids. Int J Electr Power Energy Syst 2018;100:1–9. Jianhua Z. Optimal planning of substation of locating and sizing based on GIS and
[10] Mohammad RE, Nima A. Adaptive robust optimization framework for day-ahead adaptive mutation PSO algorithm. In: IEEE Int Conf Power Syst Tech; 2006. p. 1–5.
microgrid scheduling. Electric Power Energy Syst 2019;107:213–23. [28] Zakariazadeh A, Jadid S, Siano P. Economic-environmental energy and reserve
[11] Xiang Y, Liu J, Liu Y. Robust energy management of microgrid with uncertain re- scheduling of smart distribution systems: a multiobjective mathematical program-
newable generation and load. IEEE Trans Smart Grid 2016;7:1034–43. ming approach. Energy Convers Manage 2014;78:151–64.
[12] Zhang C, Xu Y, Li Z, Dong ZY. Robustly coordinated operation of A multi-energy [29] Kim HJ, Kim MK. Multi-objective based optimal energy management of grid-con-
microgrid with flexible electric and thermal loads. IEEE Trans Smart Grid 2018;1–1. nected microgrid considering advanced demand response. Energies
[13] Shi Z, Liang H, Huang S, Dinavahi V. Distributionally robust chance constrained 2019;12(21):4142.
energy management for islanded microgrids. IEEE Trans Smart Grid [30] Mohammadreza M, Alireza Z, Shahram J, Pierluigi S. Integrated scheduling of re-
2019;10:2234–44. newable generation and demand response programs in a microgrid. Energy Convers
[14] Li Y, Yang Z, Li G, Zhao D, Tian W. Optimal scheduling of an isolated microgrid and Manage 2014;86:1118–27.
with battery storage considering load and renewable generation uncertainties. IEEE [31] Mohiti M, Monsef H, Anvari-moghaddam A, Guerrero J, Lesani H. A decentralized
Trans Ind Electron 2019;66:1565–75. robust model for optimal operation of distribution companies with private micro-
[15] Bai L, Li F, Cui H, Jiang T, Sun H, Zhu J. Interval optimization based operating grids. Int J Electr Power Energy Syst 2019;106:105–23.
strategy for gas-electricity integrated energy systems considering demand response [32] Rezaei N, Ahmadi A, Khazali AH, Guerrero JM. Energy and frequency hierarchical
and wind uncertainty. Appl Energy 2016;167:270–9. management system using information gap decision theory for islanded microgrids.
[16] Li Y, Wang P, Gooi HB, Ye J, Wu L. Multi-objective optimal dispatch of microgrid IEEE Trans Ind Electron 2018;65:7921–32.
under uncertainties via interval optimization. IEEE Trans Smart Grid [33] Fioriti D, Giglioli R, Poli D, Lutzemberger G, Micangeli A, Del Citto R, et al.
2019;10:2046–58. (chaotic group search). Stochastic sizing of isolated rural mini-grids, including effects of fuel procurement
[17] Zakariazadeh A, Jadid S, Siano P. Smart microgrid energy and reserve scheduling and operational strategies. Electr Power Syst Res 2018;160:419–28.
with demand response using stochastic optimization. Int J Electr Power Energy Syst
16