Hybridizing Infeasibility Driven and Constrained- Domination Principle with MOEA/D for Constrained Multiobjective Evolutionary Optimization - Technologies and Applications of Artificial Intelligence - page 254

3

Experimental Study

In order to evaluate the performance of the proposed algorithm, we tested its perfor-
mance on several widely used benchmark problems. The parameters' settings are
given in Table 1. Features of the selected test problems are listed in Table 2. The aim
of the experiment is to test if the ID mechanism added to MOEA/D-CDP or NSGAII-
CDP is effective and can help improve the performance of the algorithms. We there-
fore compared the performance of MOEA/D-CDP-ID with MOEA/D-CDP, and
NSGAII-CDP-ID with NSGAII-CDP.

Table 1. Parameter Settings

1 Population size (N) 100

2 Maximal number of generations 500

3 Neighborhood size(T) 20

4 Crossover rate(CR) 0.5

5 Mutation rate(F) 0.5

6 Probability of selecting

mating parents from 0.9
neighborhood

7 Number of runs: 30

ʱ

8 Proportion of infeasible solutions( ): 0.2

9 FEmax: 50000

In this study, hypervolume metric [9] is used to compute the performance of the
proposed algorithm MOEA/D-CDP-ID, as well as the abovementioned algorithms
MOEA/D-CDP, NSGAII-CDP-ID, and NSGAII-CDP. The larger the hypervolume
mean value, the better quality of the obtained non-dominated set. The experimental
results in table 3 show that the hypervolume mean value is larger when the ID me-
chanism is added to MOEA/D-CDP and NSGAII-CDP for the test functions OSY,
wherein the biggest hypervolume mean value is achieved in MOEA/D-CDP-ID. In
the test function CONSTR, the hypervolume mean value in MOEA/D-CDP-ID is
larger than MOEA/D-CDP. In addition, the hypervolume mean values are almost
equal in NSGAII-CDP-ID and NSGAII-CDP. In this test function, the biggest hyper-
volume mean value is also achieved in MOEA/D-CDP-ID. In the test function SRN,
when the ID mechanism is added to MOEA/D-CDP and NSGAII-CDP, the hypervo-
lume mean values also become larger, even though in this case the biggest hypervo-
lume mean value is achieved in NSGAII-CDP-ID. Based on the above results, it can
be safely concluded that the ID mechanism is effective to help the constrained
multiobjective optimization algorithms to achieve better performance. In addition,
embedding the ID and CDP mechanisms in MOEA/D can lead to effective new
algorithm for handling constrained multiobjective optimization problems.