Minimum variance control of a class of nonlinear plants with neural networks

In this paper the authors introduce a technique for nonlinear control based on minimum variance control ideas, originally introduced in Astrom (1970) for the linear case. They focus their attention on a class of discrete time models depending nonlinearly on the exogenous input. A minimum variance controller, made up of neural networks and linear blocks, is designed for these models. The quality of this control scheme is strongly dependent on the possibility of devising a forward model of the whole plant and an inverse model of the nonlinearity alone: this is performed with two suitable neural networks. A simple example is provided to show the applicability and limitation of their control scheme. In addition, the overall performance is compared to that of a common linear adaptive technique.