Model Independent Dynamic Predictive Controller Design Using Differential Extreme Learning Machine for Composition Control in Binary Distillation Column

Abstract

This paper presents a novel design framework termed differential Extreme Learning Machine (DELM) for addressing nonlinear process dynamics in time series modelling. DELM is constructed via a single‐layer feed‐forward ELM network featuring a skip net topology. This innovative network is engineered to accurately assess nonlinear time series patterns utilizing an nth order Legendre polynomial activation and imposing constraints at the output layer. The DELM persistently monitors trends in streaming process data and adjusts dynamic model predictive control (DMPC) settings inside the feedback loop. The Adaptive Distributed Model Predictive Control (ADMPC) is engineered to provide optimal control responses that meet both local and global stability requirements. The efficacy of DELM‐driven DMPC is evaluated for reference tracking and disturbance rejection goals and compared with RELM‐based DMPC and model‐based adaptive MPC (AMPC). The DELM‐DMPC surpasses alternative methods by providing superior generalization, stability, and computational efficiency. Average performance accuracy of 95% is attained across the operational range, exhibiting superior computing speed relative to its controller counterparts.