Controle Preditivo Não-Linear Baseado em Modelo Obtido via Aprendizado de Máquina

Abstract

Model-based predictive control (MPC) is a control design methodology based on the existence of a prediction model to solve a constraint optimal control problem with a finite receding horizon. However, in the absence of mathematical models to describe the system’s dynamical behaviour, it is necessary to use identification techniques to enable the design of predictive controllers. In this context, this work addresses the predictive control problem using an identification technique for nonlinear dynamical systems based on a non-parametric machine learning method known as Lazily Adaptive Constant Kinky Inference (LACKI). Therefore, with the prediction process being inferred from a data set describing the system input to output relations, a nonlinear predictive control strategy is presented and corroborated through a numerical example.