A Data Driven Model Predictive Controller for a Polybutylene Succinate (PBS) Synthesis Reactor

Polybutylene succinate (PBS) is a biodegradable plastic that has received attention due to its strength and versatility in a variety of applications. In this research, a neural network model-based predictive control strategy (NNMPC) and multiple neural network model based predictive control (Multi-NNMPC) using Python are developed for temperature control of the polybutylene succinate polymerization process during the esterification and polycondensation steps. The polymerization process is highly nonlinear. Therefore, a conventional model predictive controller requires a long time to perform the optimization at each time step. A neural network model can learn the process dynamics and efficiently predict the optimal value of the manipulated variable 5 to 20 times faster using the SLSQP optimization method from the SciPy library. NNMPC and Multi-NNMPC were compared to split range PID and first-principles model based MPC control using the IAE performance criteria for several scenarios.