Study of Hybrid Artificial Bee Colony Algorithm with Reinforcement Learning to Fuzzy Controller Design

  • 田 智輝

Student thesis: Master's Thesis

Abstract

The present study utilizes a hybrid artificial bee colony (ABC) algorithm with reinforcement learning Q-learning to design a fuzzy controller The optimization method demonstrates a poor balance between exploration and exploitation In order to accelerate the rate of convergence and avoids convergence to a local optimum Therefore we use the concepts of reward and penalty from reinforcement learning methods to train a hybrid ABC algorithm Our study works on the principle of reward and penalty to assist the ABC algorithm to avoid premature convergence and escape the local optimum The proposed hybrid ABC algorithm with Q-learning is successfully applied to solving 23 benchmark problems of global numerical optimization which have unimodal and multimodal functions The simulation results show that the proposed hybrid ABC algorithm with Q-learning offers more opportunities to find the optimal solutions Simultaneously the ABC algorithm with Q-learning is compared with the original ABC algorithm The results show that the hybrid ABC algorithm with Q-learning has a superior performance in convergence rate Finally this study aims to tune the parameters of the fuzzy logic controller for nonlinear systems The simulation results demonstrate the effectiveness and feasibility of the proposed method
Date of Award2018 Aug 10
Original languageEnglish
SupervisorTzuu-Hseng S. Li (Supervisor)

Cite this

'