Fuzzy Adaptation of Parameters in a Multi-swarm Particle Swarm Optimization (PSO) Algorithm Applied to the Optimization of a Fuzzy Controller

Many control problems require running numerous simulations to find a suitable configuration for the controller’s parameters. Population-based distributed algorithms can be used to speed up this procedure. One way to do this is to use multiple independent populations, each running an independent search algorithm in parallel. It is difficult to find the ideal configuration for these algorithms, such as the number of swarms, defining how particles are exchanged between swarms, and especially the parameters that affect exploration and exploitation in the search. We suggest a version of Particle Swarm Optimization (PSO) that includes a Fuzzy Inference System (FIS) to change the algorithm parameters dynamically. The adjustment considers two variables: population diversity and the number of iterations performed on the population. As an output of the FIS, we obtain the adapted parameters, representing the new values for the social and cognitive coefficients to be used in the next iteration. We aim to evaluate if this strategy helps minimize the evaluation time and minimize the root-mean-square error (RMSE). As a case study, the distributed PSO algorithm is applied to optimize the membership functions of a fuzzy controller for tracking the trajectory of an autonomous mobile robot. When compared to other configuration strategies, experimental results achieve a similar RMSE.