An Acquiring Method of Fuzzy Reasoning Rules by Genetic Algorithm with Variable Gene Length
Tatsuya Masuda, Akio Ito, Koji Sato · IEEJ Transactions on Electronics Information and Systems · 1995
Some methods for acquiring fuzzy reasoning rules by standard genetic algorithm have been proposed.These methods use a genetic code which contains lotuses corresponding to all consequent selections of fuzzy reasoning rules, so that the length of gene becomes very longer as the number of reasoning rules increases.As the result, it is difficult to identify exactly all reasoning rules and to understand their contents.In this paper, we propose a new method for acquisition of fuzzy reasoning rules to solve this problem.The genetic algorithm we use for this method has a dynamic adjustable function of the length of gene, so that we can obtain only necessary rules to express the characteristics of a controlled object.The rule acquisition process used by this method closely resembles that by which humans create fuzzy reasoning rules by trial and error.We demonstrate the effectiveness of this method by applying it to an obstacle avoidance problem of mobile robot.