Extended Model Predictive Control Based on Multi-Structure RBF Networks: Design and Application to Clutch Control

Abstract

For many control systems in engineering practice, the nonlinear, slow time varying and multistage dynamics make it difficult to realize precise control. Therefore, the extended model predictive control (EMPC) based on multi-structure radial basis function neural networks (MSRBFNN) is proposed in this paper. To begin with, the standard model predictive model (MPC) is established from system analysis or identification to formulate the problem with known disturbances. Secondly, the original MPC is extended with radial basis function neural networks (RBFNN) to deal with the nonlinearity and slow time varying dynamics of the system. Further, considering that a single RBFNN is too complicated to cope with the multistage dynamics of the system for real time application, the MSRBFNN, composed of a series of simpler RBFNNs, is designed to replace the original RBFNN. The running system can then switch between different RBFNNs at different working conditions. Finally, a successive iteration method is introduced to derive the controller with nonlinear compensations and the stability issue is also discussed. Application to the dry clutch control system shows that the proposed method has a much better performance than the standard MPC. Moreover, it can also be extended to other similar nonlinear control systems.

Citation

Recommended citation: C. Huang, L. Li and X. Wang, “Extended Model Predictive Control Based on Multi-Structure RBF Networks: Design and Application to Clutch Control,” in 2018 the 5th IFAC Conference on Engine and Powertrain Control, Simulation and Modeling (E-CoSM), vol. 51, no. 31, pp. 653-658, Sep. 2018. (Paperurl)