Fast Training Algorithm for Genetic Fuzzy Controllers and Application to an Inverted Pendulum with Free Cart

EasyChair Preprint 5041

Abstract

The classical control theory cannot be applied in those systems whose complexity is too high to be analytically modeled. In these cases, math-ematical methods with more degrees of freedom are used because they provide better adaptation. One method widely used in control problems is the Fuzzy In-ferencing Systems. However, the process of calibration of the parameters re-quired may involve a high computational cost. Among them, Genetic Algo-rithms have demonstrated great convergence towards ideal solutions. As the dimensions of the control problem (input features) increase, the optimization process requires much more time. Therefore, the present work proposes a grad-ual search and parameter update criteria for Genetic Fuzzy Controllers because it improves several orders of magnitude in the processing time. The algorithm developed has been applied to the control problem of the Inverted Pendulum with Free Cart. The results obtained demonstrate an effective parameter calibra-tion in seconds, while the traditional method of tuning for the same problem takes more than 2 hours. Currently, many of the mechanical systems of the dif-ferent industries undergo sudden changes in their properties during use, there-fore an instant effective recalibration of the controllers is necessary. This meth-od allows fast adaptation and also guarantees the same performance in the con-trol process.

Keyphrases: Chained Genetic Algorithms, Fast Tuning, Fuzzy Inferencing System, Gradual Update, Symmetric Mechanical Systems

@booklet{EasyChair:5041,
  author    = {Javier Viaña and Kelly Cohen},
  title     = {Fast Training Algorithm for Genetic Fuzzy Controllers and Application to an Inverted Pendulum with Free Cart},
  howpublished = {EasyChair Preprint 5041},
  year      = {EasyChair, 2021}}