Balanced random search and directed search for Takagi-Sugeno type fuzzy logic controller automatic optimization

The paper presents a new cell state space based Takagi-Sugeno (TS) type fuzzy logic controller (FLC) automatic optimization algorithm. It is a new version of the Incremental Best Estimate Directed Search (IBEDS) algorithm. IBEDS starts with an initial training set that may be empty, then an FLC with randomly initialized rule output parameters is trained by a least mean square (LMS) learning algorithm in an iterative procedure. In each iteration, the trained FLC is evaluated with cell state space based global and local performance measures, and the training set is then updated based on the evaluation under best kept policy, which only keeps the best control commands for each cell center found so far. Originally, in IBEDS, all the FLCs with progressively better performance were discarded. The paper proposes to reuse and reorganize the rule bases of all the FLCs being evaluated to further expedite the search. In the new approach, the training set and the parameter set of a rule base are optimized simultaneously. Simulation results with a 4D inverted pendulum show that the new version of IBEDS is much faster than IBEDS when the initial training set is empty and the search needs to bootstrap itself.