Data-Driven Approximated Optimal Control for Chemical Processes with State and Input Constraints

Abstract

Input and state constraints widely exist in chemical processes. The optimal control of chemical processes under the coexistence of inequality constraints on input and state is challenging, especially when the process model is only partially known. The objective of this paper is to design an applicable optimal control for chemical processes with known model structure and unknown model parameters. To eliminate the barriers caused by the hybrid constraints and unknown model parameters, the inequality state constraints are first transformed into equality state constraints by using the slack function method. Then, adaptive dynamic programming (ADP) with nonquadratic performance integrand is adopted to handle the augmented system with input constraints. The proposed approach requires only partial knowledge of the system, i.e., the model structure. The value information of the model parameters is not required. The feasibility and performance of the proposed approach are tested using two nonlinear cases including a continuous stirred-tank reactor (CSTR) example.