Construction of dynamic fuzzy if-then rules through genetic reinforcement learning for temporal problems solving

In this paper, a genetic algorithm (GA) based dynamic fuzzy network design approach is proposed. First, a dynamic fuzzy network (DyFN) constituted from a series of dynamic fuzzy if-then rules is introduced. One characteristic of DyFN is its ability to deal with temporal problems. Then, GA is adopted into the design process as a means of allowing the application of DyFN in situations where gradient information is costly to obtain or only a reinforcement signal is available. To promote the design performance, a modification to the traditional GA, the Relative-based Mutated Reproduction GA (RMRGA), is proposed. To show the efficiency of DyFN designed by GAs, including both traditional GA and RMRGA, two temporal problems, dynamic plant control and adaptive noise cancellation, are simulated. The simulated results have verified the efficiency of DyFN designed by GA