Tuning discrete PI controllers by fuzzy fitness based genetic algorithms

Abstract

Different methods and schemes have been proposed in literature for tuning continuous and discrete PI (ProportionalIntegral) controllers. This paper proposes a scheme in which, this controller structure is explored in a different way, by looking its behavior as a lag compensator and tuning it by genetic algorithms. A difference with conventional approaches is the manner to evaluate every individual generated by the evolutionary algorithm. That evaluation is achieved by a set of measurements which becomes the input of a fuzzy inference system that models the expert's knowledge. This scheme is simulated and tested over two nonlinear dynamical systems. Results show that a widely variety of discrete PI controllers can be obtained for one dynamical system, based on the same tuning criterion and having high performance levels in comparison with conventional methods.