Fuzzy Control of Exploration and Exploitation Trade-Off with On-Line Convergence Rate Estimation in Evolutionary Algorithms

Abstract

In this paper, the fuzzy control of exploration and exploitation trade-off with on-line convergence rate estimation in evolutionary algorithms is presented. We introduce to the proposed algorithm three fuzzy systems (Mamdani type-1). These fuzzy systems are responsible for controlling the parameters of an evolutionary algorithm, such as selection pressure, crossover probability and mutation probability. While creating the fuzzy rules in the proposed fuzzy systems, we assumed that, at the start, the algorithm should possess maximal exploration property (low selection pressure, high mutation probability, and high crossover probability), while at the end, the algorithm should possesses maximal exploitation property (high selection pressure, low mutation probability, and low crossover probability). Also, in the paper we propose a method for estimating the algorithm convergence rate value. The proposed approach is verified using test functions chosen from literature. The results obtained using the proposed method are compared with the results obtained using evolutionary algorithms with a different selection operator, and standard values for crossover and mutation probability.