Deep Neural Network-Model Predictive Control (DNN-MPC) for Temperature Control in Batch Reactor

Abstract

A hybrid Deep Neural Network- Model Predictive Control (DNN-MPC) is developed regulate the temperature of batch reactor in the production of magnesium stearate. Thermal mathematical model of magnesium stearate production was employed to generate 1,000 simulation-based training and validation datasets, with a holdout ratio of 0.8. The DNN was trained using the Rectified Linear Unit (ReLU) activation function and the Adam optimizer. The trained DNN is integrated into a stable MPC with a prediction and control horizon of 60 seconds and 20 seconds respectively in MATLAB R2024a Simulink. Comparative test with standalone MPC in both servo and regulatory control system demonstrates that the proposed DNN-MPC exhibits fast settling time with zero overshoot and minimal error of 0.002794. DNN-MPC demonstrate great ability to reject disturbance where minimal error of 0.516% and small overshoot of 13% is observed. These results highlight the robustness, responsiveness, and efficiency of the DNN-MPC, establishing its potential as a highly effective temperature control strategy for batch reactor in the production of magnesium stearate.