Design of a Fuzzy Logic Controller With Evolutionary Q-Learning
Min-Soeng Kim, Ju-Jang Lee · Intelligent Automation & Soft Computing · 2006
Abstract In this paper, an Evolutionary Q-Learning(EQL) algorithm is proposed, which is based on the modified Q-learning and evolutionazy algorithm. The objective of the proposed EQL algorithm is to fmd a fuzzy logic controller (FLC) when only a binary reinforcement signal is available from un unknown target environrnent. The proposed EQL algorithm utilizes and evolves a group of FLCs simultaneously to obtain more feasible solution set. By defining Q-values as functional values of states and FLCs, whole FLCs in the group experience Q-learning process together during the same generation. The Q-learning process assists the proposed EQL algorithm in fmding better FLCs with good quality consequent parts. At the end of each generation, the best FLC is constructed by the unique elite construction algorithm. In usual case where evolutionazy process, which is basically parallel process, is used with reinforcement learning, multiple instances of tazget systems are necessary to make the algorithm applied on-line. O...