Enhanced NMPC for Stochastic Dynamic Systems Driven by Control Error Compensation With Entropy Optimization
Ping Zhou, Xiaoyang Sun, Tianyou Chai
Abstract
In this article, an enhanced nonlinear model predictive control (En-NMPC) method driven by control error compensation with entropy optimization and unknown state estimation is proposed for high-performance control of multivariable and non-Gaussian stochastic dynamic systems. First, the extended Kalman filter (EKF) is used to estimate the unknown states online, and the estimation of posterior states is used as the input of the neural network compensator so that the unknown states, which are difficult to use in basic NMPC, can be fully applied in the compensation control input; Then, the kernel density estimation (KDE) is used to indirectly obtain the control error entropy of the non-Gaussian dynamic system. Guided by the optimization performance index, which is constructed mainly about the control error entropy, the output weight of the compensator is optimized to tune the compensation effect; Finally, compensation control and basic predictive control are integrated to achieve high-performance control of stochastic dynamic systems. The control error of the proposed method and the upper bound of the state estimation error are analyzed by inductive reasoning method to ensure that the closed-loop system has input-to-state stability (ISS) about the disturbances. Data experiments of the sewage treatment process verify the superiority and practicability of the proposed method.