Adaptive Switching Controller Design Based on MPC and Virtual Unmodeled Dynamics Compensation

This paper proposes an adaptive switching controller (ASC) to address nonlinearity, strong coupling, and unmodeled dynamics in complex industrial processes. The ASC integrates model predictive control (MPC) with a virtual unmodeled dynamics compensator. Firstly, a linear discrete‐time multivariable system serves as the controlled plant, a linear controller is designed based on MPC and decoupling compensator. Then, to compensate for the dynamic tracking error between the linear controller and the actual system, a nonlinear controller with a virtual unmodeled dynamic compensator is proposed, combining an adaptive neural fuzzy inference system (ANFIS) with stochastic configuration networks (SCN). A multi‐information stochastic gradient algorithm estimates the plant parameters online, while an adaptive switching mechanism selects the optimal controller in real time. Finally, the accuracy of the virtual unmodeled dynamics compensator and the convergence of ASC are rigorously validated. In the cement raw meal decomposition process, ASC outperforms mainstream controllers, reducing the tracking root mean square error (RMSE) by over 12.8% and the output variance by approximately 63.6%. This demonstrates its robust potential for complex industrial applications.