Reducing Time of Evolutionary Computation using Fitness Estimation

Abstract

Evolutionary Computation (EC) techniques have shown their outstanding performance in the engineering fields for solving combinatorial optimization problems. Although the CPU is getting faster, ECs still take much time because they need a lot of iteration. Some acceleration methods were proposed that modified genetic operators such as mutation, recombination and/or their control parameters. Although these modifi-cations show improvement to some extent, ECs still need much time.
In many engineering applications of ECs, fitness evaluation takes most of their time. This paper presents a new approach on the acceleration of ECs by reducing time for fitness evaluation. Reducing time for fitness evaluation will result in acceleration of ECs in time domain. Only one chromosome in the population is actually evaluated in the proposed method. Fitness value for the rest of population are estimated using the fitness database updated by the real evaluation. The time for fitness estimation of whole population can be negligible comparing with the actual evaluation.
The simulation results for designing fuzzy logic controller using Genetic Algorithm (GA) show the effective-ness to accelerate the evolution using the proposed method. Simulation results for De Jong's test functions show that the proposed method is applicable to other problems.