Entropy optimization based linear compensation for performance enhancement of NMPC and its applications

Abstract

For non-Gaussian nonlinear dynamic systems, an enhanced nonlinear model predictive control (NMPC) method driven by error entropy optimization based linear compensation is proposed to improve the control effect of NMPC algorithm under non-Gaussian noise. The proposed enhanced NMPC method mainly includes two parts, namely the basic NMPC part and the proposed entropy optimization based linear compensation control part. In general, the cost function to measure the control effect in predictive control is only the mean and variance, but it lacks sufficient optimization effect for non-Gaussian noise. Therefore, based on the relationship between the error entropy, estimated state and control input constructed in this paper, the optimal compensation input is obtained by optimizing the weights in the form of a gradient descent algorithm under the designed cost function. Thus, the error entropy optimization is introduced into NMPC which effectively improves the control effect. The upper bound of the control error and the state estimation error caused by non-gaussian noise are analyzed by induction reasoning to ensure the bounded-input bounded-output stability of the proposed method. The data experiments of the wastewater treatment process and blast furnace Ironmaking process verify the advancement and practicability of the proposed method.