Phased Long Short-Term Memory-Based Predictive Control of Chemical Processes With Asynchronous and Delayed Measurements

This work develops novel machine learning modeling and predictive control techniques for nonlinear chemical systems that experience asynchronous and delayed measurements that result in missingness in both offline and online data collections. Specifically, a phased long short-term memory (PLSTM) network is used to learn the process dynamics amidst the missingness in the data measurements, during the offline training process. The generalization performance of PLSTM is theoretically studied on the basis of statistical machine learning theory to better understand the capabilities of PLSTM models. The PLSTM model is subsequently employed to forecast the evolution of states for a Lyapunov-based model predictive control (LMPC). The proposed PLSTM-based LMPC is designed to account for data loss and delays in real-time implementation, and guarantees the closed-loop stability of nonlinear systems subjected to missing real-time data, provided that there is an upper bound on the number of consecutively missing real-time data. Finally, two chemical processes including an extractive dividing wall column (EDWC) and a continuous stirred tank reactor (CSTR) are used to demonstrate the effectiveness of PLSTM modeling and predictive control methods.