Efficient MOEAs with an Adaptive Sampling Technique in Searching Robust Optimal Solutions


Searching robust solutions in multi-objective evolutionary algorithms (MOEAs) usually optimize effective objective functions instead of original objective functions. It needs sampling in a delta-neighborhood of an individual. Fitness of every sample must be evaluated, so large computation of MOEAs becomes an urgent problem. To solve the flaw of low efficiency resulted from fixed sample size, this paper combined NSGA-II with Random Sampling(RS) and Latin Hypercube Sampling (LHS) to search robust optimal solutions, and proposed an adaptive sampling technique which changes the sample size adaptively in the sampling process. It reduces unnecessary samples and fitness evaluations. Accordingly, two sampling methods named Adaptive RS (ARS) and Adaptive LHS (ALHS) were designed. The experimental results demonstrate that ARS and ALHS can reduce CPU time and fitness evaluations significantly compared with RS and LHS in the condition of not degrading the performance of MOEAs in searching robust optimal solutions. In a word, the adaptive sampling technique can improve the efficiency of MOEA obviously.