2148 Wang R.

/ Energy Reports 6 (2020) 2146–2152

3.2. Multi-objective configuration optimization method for a hybrid

energy system based on the fuzzy artificial bee colony algorithm

If the solution to be optimized in the artificial bee colony

algorithm is denoted as θ , then θ = [NWT , NPV , NBatt ]. For the
multi-objective configuration optimization of the hybrid energy
system in this paper, θ(l) dominates θ(j), which is denoted as
θ(l) ≺ θ(j). They must satisfy both Equations (14) and (15).
∀i ∈ {1, 2}: Fi (l) ≤ Fi (j) (14)

at least ∃k ∈ {1, 2}: Fk (l) < Fk (j) (15)

In Eqs. (14) and (15), F1 (l) and F2 (l) represent CT in Eq. (5) and
Fig. 3. Optimization mechanism of the ABC algorithm.
LPSP in Eq. (11), respectively. The 2 objective functions, CT and
LPSP, are contradictory, so we use fuzzy logic rules to quantify the
relationship between them; the fuzzy logic rules are described by
where Pt , Ren = Pt , PV + Pt , WT + Pt , Batt ; Pt , Batt is the output power of Eqs. (16)–(17).
the energy storage module. The smaller the loss of power supply ⎧
probability (LPSP) is, the better the system load power shortage ⎪
⎪ 0 Fk (l) ≤ Fkmin
performance. Fkmin
⎪
Fk (l) −
⎨
The configuration optimization of the diesel-based hybrid en- ϕlk = max
Fkmax ≥ Fk (l) ≥ Fkmin (16)
ergy system in this paper is used to determine the best values
⎪
⎪ F k − Fkmin
⎪
Fkmax ≤ Fk (l)
⎩
of NPV , NWT , and NBatt and to simultaneously minimize the 2 1
objective functions, CT and LPSP. ∑Nobj
=1 ϕl
k
φl = ∑ k∑ (17)
3. Multi-objective configuration optimization method for a NF N
k=1 ϕl
obj k
l=1
hybrid energy system based on the fuzzy-artificial bee colony
algorithm Here, k = 1, 2, . . . , Nobj , where Nobj is the number of objective
functions.
3.1. Artificial bee colony optimization algorithm The process of using the artificial bee colony algorithm to
optimize the configuration of the diesel hybrid energy system
The artificial bee colony (ABC) is a group intelligent optimiza- is as follows: the minimum function, Eq. (17), is used as the
tion algorithm developed by Karaboga in 2005 to simulate the objective function, and then the combination of the ABC and
process of a bee searching for nectar (Karaboga, 2005; Wang, multi-objective dominance solution is used to optimize the global
2016). According to the labor division in the process of collecting minimum of the function in Eq. (17), and the best optimal so-
nectar, the bee colony is divided into employed bee (EB), onlooker lution for the quantitative relationship between CT and LPSP is
bee (OB), and scout bee (SB) segments. The ABC algorithm is obtained. In addition, the array of parameters corresponding to
a cluster intelligent algorithm that simulates the mechanism of the optimal solution is shown in Eq. (18).
the labor division of a bee colony to find nectar to solve the
multidimensional optimization problem. The ABC optimization θ (l, d) = θmin (d) + r [θmax (d) − θmin (d)] (18)
mechanism is shown in Fig. 3. If the parameter to be optimized
is regarded as the nectar source, then 1 optimization iteration for where θmax (d) and θmin (d) are the maximum and minimum values
the parameter to be solved needs to complete 3 stages: in the of the optimized solution d-dimension, respectively, which are
EB stage, it is looking for the next new solution in the proximity limited by Eqs. (19)–(22).
of the solution obtained from the last optimization process (the max
0 ≤ NWT ≤ NWT (19)
iterative equation is Eq. (12)); in the OB stage, a new solution is
max
generated based on the solution generated by the EB stage, and 0 ≤ NPV ≤ NPV (20)
the optimization is performed (the iterative equation is Eq. (13)); 0 ≤ NBatt ≤ max
NBatt (21)
and in the SB stage, the solution that is not updated during several
min max
consecutive iterations is abandoned and a new solution is gener- EBatt ≤ Et , Batt ≤ EBatt (22)
ated. Compared with group optimization algorithms such as the
For ease of analysis, the implementation steps are summarized
differential evolution in Wang et al. (2015b), genetic algorithm in
Wang et al. (2016), and particle swarm algorithm in Wang et al. in Fig. 4.
(2015a), the ABC algorithm provides a better optimization mech-
anism and convergence performance, which is why we chose the 4. Simulation analysis
ABC algorithm.
To verify the effectiveness and rationality of the multi-
θEB (l, d) = θ (l, d) + rld [θ (l, d) − θ (lr , d)] (12) objective configuration optimization method for the diesel-based
where l = 1, 2, . . . , NF . NF is the size of the bee colony; d = hybrid energy system in Section 3.B, different methods were
1, 2, . . . , D.D is the dimension of the optimization solution. used to optimize the configurations of the 3 modes of the hy-
V (l, d) and θ (l, d) are the original and new optimization solutions. brid energy system (PV/energy storage/diesel, wind/energy stor-
rld is a random number between 0 and 1; lr is a random integer age/diesel, and PV/wind/energy storage/diesel) in an edge region
between 1 and NF , and lr ̸ = l. located in Idaho, USA. The choice of Idaho in the United States was
based on data availability. The load power consumption in the
θOB (l, ds ) = θ (l, ds ) + rld [θ (ls , ds ) − θ (lr , ds )] (13)
region from January 2014 to December 2014 is shown in Fig. 5.
where ds is a random integer between 1 and D, ls is a random The wind speed and solar radiation collected by a weather station
integer between 1 and NF , and ls ̸ = lr ̸ = l. are shown in Figs. 6 and 7, respectively. The data come from