An adaptive neural regulator of minimum variance

This paper presents an extension of the minimum variance control strategy (MV) to nonlinear models. The synthesis of the optimal adaptive controller of reduced complexity is presented-the regulator consists of P, PI or PID, and perhaps of nonlinearity. Parameters of this regulator (P, PI and PID gains with nonlinearity parameters) are tuned to minimize a quadratic criterion function. The process model is assumed to be known or estimated via neural network of appropriate structure. The process model has linear autoregressive part and nonlinear function of the exogenous input. Parameter adaptation is performed for several reference types and different regulator structures.