Centralized predictive controller design based on an uncertain neural network for multivariable non-linear systems

Abstract

In this paper, we propose a feedforward neural networks-based robust predictive controller for a class of multi-input–multi-output non-linear systems. Using the structured uncertainties of the output layer’s weights of the neural networks model, the non-linear model of the real system is determined at each operating point. The control law is formulated as a minimax problem, which is solved online. The non-convex optimization is developed by minimizing the worst case of the objective cost function, taking into account the uncertainties of the non-linear model and the input control signal constraints. The efficiency of the proposed neural predictive controller is illustrated, in simulation, with a multivariable system example.