Optimization Scheme of Fuzzy Control System based on Reinforcement Learning Compensation

Fuzzy control technology has been extensively studied and widely used in reality due to its ability to transform human control experience, described in vague terms, into automatic control law. However, when dealing with complex systems that require valuable control experience, fuzzy control technology may not necessarily guarantee optimal control performance. Therefore, developing an advanced optimization scheme of fuzzy control systems is essential. In this paper, a model-free optimization scheme is developed for the fuzzy control system based on reinforcement learning algorithm, which is named Reinforcement Learning Assisted Fuzzy Control (RLAFC). The main feature of this scheme is that it can automatically obtain an effective compensation policy for the fuzzy controller by employing a data-driven reinforcement learning algorithm. Compared to the standalone reinforcement learning algorithm, the proposed method can achieve a synergistic combination of experiential and data-driven learning approaches, which brings about better control performance as well as improved safety assurance during the learning process. Finally, the simulation results of a car-following control task demonstrate the effectiveness of the proposed method. We believe that RLAFC is an effective solution for the control problems of complex systems by the combination of human control knowledge and artificial intelligence technology.