Practical Nonlinear Model Predictive Control Algorithm for Long Short-Term Memory Networks

Abstract

This paper presents a practical nonlinear model predictive control algorithm using Long Short-Term Memory (LSTM) artificial neural networks as process model. The controller is based on the Practical Nonlinear Predictive Controller (PNMPC) algorithm, a general framework which can be used for the implementation of nonlinear model predictive controllers using nonlinear models of almost any class. LSTM has gained attention during the last years for its ability to build high fidelity models in data-driven problems, such as system identification. In the proposed control algorithm, the LSTM model is used to obtain on-line a nonlinear prediction of the system free response and a linearized version of the neural model is used to obtain a local approximation of the system step response, which is used to build the dynamic matrix of the system at each sampling instant. The proposed controller was tested in a benchmark neutralization reactor control problem. The process was identified using the LSTM approach and the identification and closed-loop results are compared with linear and Hammerstein nonlinear models. The LSTM approach outperformed the other alternatives in both identification and control problems. The results obtained in this study show that the proposed control algorithm is an option to use a state-of-the-art nonlinear MPC in a computationally affordable way.