Enhancing Control Performance through ESN-Based Model Compensation in MPC for Dynamic Systems

Deriving precise system dynamic models through traditional numerical methods is often a challenging endeavor. The performance of Model Predictive Control (MPC) is heavily contingent on the accuracy of the system dynamic model. Consequently, this study employs Echo State Networks (ESN) to acquire knowledge of the unmodeled dynamic characteristics inherent in the system. This information is then integrated with the nominal model, functioning as a form of model compensation. The present paper introduces a control framework that combines ESN with MPC. By perpetually assimilating the disparities between the nominal and real models, control performance experiences augmentation. In a demonstrative example, a second-order dynamic system is subjected to simulation. The outcomes conclusively evince that ESN-based MPC adeptly assimilates unmodeled dynamic attributes, thereby elevating the system’s control proficiency.