Multi-stage Nonlinear Model Predictive Control with Online Scenario Update for Semi-batch Polymerization Processes

摘要

In this paper, the problem of multi-stage nonlinear model predictive control with scenario update is investigated for semi-batch polymerization processes. The objective is to propose novel online scenario update schemes such that the more reasonable scenario tree can be generated. Firstly, based on the Orthogonal Configuration of Finite Elements (OCFE) method of direct radau configuration, the dynamic optimization problems are converted to Nonlinear Programing (NLP) problems such that the speed and accuracy of real-time optimization problem solving are effectively improved. Then, the scenario deviation is calculated based on model prediction information of each scenario and process measurement information. After that, calculate the bayesian probability weight of corresponding scenario is obtained. The online scenario reduction scheme uses the weight information update scenarios gradually reduce the scope of scenario tree representation. The online scenario weight update scheme uses the weight information as the basis for weight assignment of each scenario in the optimization problem. They use different methods to make the scenario tree modeling approach the real realization of uncertainty, and reduce the conservativeness compared with the traditional MSNMPC fixed scenario tree method. Through multiple batches numerical simulations of a semi-batch polymerization process, the advantages and effectiveness of the two proposed schemes are verified.