Optimization of Fuzzy Controllers Using Distributed Bioinspired Methods with Random Parameters

Abstract

The optimization of fuzzy controllers is a task that requires a high demand for computational resources since it is necessary to carry out a large number of simulations to evaluate the operation of the controller. This is time-consuming work, and because of this, distributed and asynchronous bio-inspired algorithms have been proposed recently to make the execution achieve acceptable results in terms of scaling and optimization improvement. These algorithms use multiple populations and processors to search algorithms in parallel on each population. A problem with these algorithms is finding the ideal configuration that we will use to execute the algorithms, such as the number of populations, the number of processors needed, and particularly the parameters that affect the exploration and exploitation in the search. In this work, we compare homogeneous configurations for all populations against heterogeneous configurations in which we randomly define the parameters. We want to evaluate if this strategy helps us to minimize the evaluation time and minimize the RMSE. We will use genetic and particle swarm optimization algorithms in this experiment. As a case study, we applied the algorithms to optimize the membership functions of a fuzzy controller for the trajectory tracking of a mobile autonomous robot. Results show that even when we use random parameters in our algorithms, we continue to obtain an RMSE similar to the previous results. We also observe a reduction in execution time.