Neuro-Optimal Tracking Control for Continuous Stirred Tank Reactor With Input Constraints

This paper proposes a novel data-based optimal control algorithm for continuous stirred tank reactor (CSTR) system based on adaptive dynamic programming (ADP). To overcome the challenge of establishing an accurate mathematical model for the CSTR system, neural networks are employed to reconstruct the dynamics of the CSTR system using the production data of the system. A new nonquadratic form performance index function is provided, where the control input is constrained in order not to exceed the bound of the actuator. Then, the operational optimal control problem of CSTR is formulated. Furthermore, an iterative ADP (IADP) algorithm is developed to obtain the optimal tracking controller for the CSTR system with control constraints. In particular, the convergence analysis of the IADP algorithm is developed. The proposed IADP algorithm is implemented via the dual heuristic dynamic programming structure. Finally, the proposed approach is applied to the real CSTR system to verify the effectiveness and performance.