Fuzzy strength-based XCS: An application on multi-step environment problems

Abstract

This paper proposes an algorithm to perform an online fuzzy system construction was called FsXCS (Fuzzy Strength-Based XCS). FsXCS is the extended XCS system combining XCS with fuzzy logic theory to tackle the multi-step continuous input-output problems. Indeed, XCS is a great attention discrete-valued system considering from the rule generalization system. However, it becomes more difficult when extending to continuous-valued systems, and even more difficult when addressed in multi-step continuous input-output problems. In order to develop FsXCS, we propose new computation details on some XCS components while these changes do not affect the original XCS's learning procedures. The performance of FsXCS was tested on simulation classical continuous problems; n-Environment, and automatically back driving a truck. The experimental results ware compared to the tabular Q-Learing with high discretization.