Precise and Computationally Efficient Nonlinear Predictive Control Based on Neural Wiener Models

摘要

This paper describes a nonlinear Model Predictive Control (MPC) algorithm based on a neural Wiener model. The model is linearised on-line along the predicted trajectory. Thanks to linearisation, the algorithm is computationally efficient since the control policy is calculated on-line from a series of quadratic programming problems. For a nonlinear system for which the linear MPC approach is inefficient and the MPC algorithm with approximate linearisation is inaccurate, it is demonstrated that the described algorithm gives control quality practically the same as the MPC approach with on-line nonlinear optimisation.