MEAPCA: a multi-population evolutionary algorithm based on PCA for multi-objective optimization. The simulated binary crossover (SBX) and differential evolution operators (DE) are two most representative evolutionary operators. However, due to their different search pattens, they are found to face difficulty on multi-objective optimization problems (MOPs) with rotated Pareto optimal set (PS). The regularity model based multi-objective estimated distribution algorithm, namely, RM-MEDA that adopts a segmented PCA model to estimate the PS shows good performance on such problems. However, determining the offering number of segments (clusters) of the PCA model is difficult. This study therefore proposes a multi-population multi-objective evolutionary algorithm based on PCA (MEAPCA) in which the optimization process is divided into two phases. The first phase employs a multi-population method to quickly find a few well-converged solutions. In the second phase, new offspring are generated under the guidance of the PCA model. That is, the PCA model utilizes information of those well-converged solutions so as to ensure the generation of good offspring. The DTLZ1-5 with modified PS are used as test problems. The MEAPCA is then compared against RM-MEDA as well as two representative MOEAs, i.e., NSGA-II and MOEA/D, and is found to perform well on MOPs with complex Pareto optimal front.
Keywords for this software
References in zbMATH (referenced in 1 article , 1 standard article )
Showing result 1 of 1.
- Dong, Nan-jiang; Wang, Rui: MEAPCA: a multi-population evolutionary algorithm based on PCA for multi-objective optimization (2020)