Data-Driven Techniques for Divide and Conquer Adaptive Control

Abstract

This paper presents an approach for modeling and controlling discrete-time non-linear dynamical system. The controller consists of a multiple step ahead direct adaptive controller. At each time step a forward simulation of the system composed by the controller and the plant model is performed. This dynamic information is then used to adapt the parameters of the controller. In order to obtain good results it is necessary to have a good model of the process to control. Takagi-Sugeno fuzzy systems and lazy learning, are two approaches which can be successfully used to model the controller plant. This paper focuses on case when a model of the plant is not given a priori and has to be learned starting from a limited amount of data and it is necessary to add some adaptation capabilities to perform on-line learning. Simulation examples of the control of the manifold pressure of a car engine using adaptive and non-adaptive versions of Takagi-Sugeno and Lazy models are given.