Heat Management in Batch Reactors Using Flatness-Based Neural Predictive Control

Effective heat management in batch reactors (BR) ensures accurate temperature control, enabling optimal reaction conditions, product quality, and process safety. Challenges in real-time temperature tracking involve balancing computational time with accuracy. Efficient Algorithms, as well as a simplified problem formulation, are essential in reducing computational complexity. This study exploits polymerization systems' differential flatness property (DFP) to simplify the temperature trajectory tracking problem. Expressing states and inputs as a function of flat outputs and their derivatives allows a lower-dimensional flat space, reducing computational complexity. A flatness-based model predictive control (FMPC) is proposed for efficient trajectory generation, enabling the handling of constraints while effectively producing feasible flat trajectories. This control architecture couples feedback from (MPC) with flatness feedforward linearization. It offers the computational advantage of requiring the solution of a convex quadratic programming (QP) instead of a nonlinear program. Meanwhile, a neural network-based model-free control (NN-MFC) has been designed to enhance the feedback loop of the FMPC and obtain improved trajectory tracking performance. In this paper, a nonlinear optimal tracking control problem forAa nonlinear batch reactor is formulated and investigated. Moreover, a nonlinear model predictive control (NMPC) has been compared to the proposed control algorithm in crucial aspects such as tracking efficiency and computational complexity. The result of the proposed scheme has the advantages of excellent tracking and significantly reduced complexity compared to NMPC.