Energy-efficient LSTM-based predictive control for nonlinear systems

This paper presents an energy-efficient control approach based on long short-term memory (LSTM) neural networks for nonlinear thermal systems. The proposed strategy uses a trained LSTM network to predict future system outputs and optimize control actions over a prediction horizon. The control scheme minimizes both the tracking error and energy usage. To improve control performance, the optimal control input generated by the LSTM-model predictive control (MPC) scheme is further refined using a discrete proportional–integral (PI) compensator. The approach is evaluated against classical baseline controllers under identical operating conditions. Simulation results and an experimental validation in a transesterification reactor show that the proposed LSTM-based control strategy significantly improves reference tracking while reducing energy consumption, demonstrating its effectiveness for controlling nonlinear systems.