Model Predictive Control For Continuous Crystallization Process Based On Koopman Operator

Crystallization is a crucial step in chemical industry and pharmaceutical industry.The effect of crystallization quality directly affects the downstream operation.At present,continuous crystallization is gradually used in industrial crystallization,which solves the problems of high batch crystallization cost and poor product consistency.Therefore,this thesis uses Continuous Oscillatory Baffled Crystallizer(COBC)as the research object to model,control and optimize it.The main work is as follows:First of all,according to the characteristics of continuous crystallization,the population balance model was established by combining the population balance equation and crystallization kinetics,and the model was solved by traditional algorithm.Because of the complexity of the model,it is difficult to apply the modern control theory to the continuous crystallization process,and the effect of model-free control is not good.Therefore,this paper presents a model predictive control based on Koopman operator that identifies the population balance model as Koopman linear system and applies it to the model predictive controller.The cooling curve is derived from the supersaturation change obtained by the model predictive controller.Moreover,the dual rate control scheme is used to solve the problem of measuring the specific position of the real crystallizer.Secondly,in order to describe the continuous crystallization process more accurately,the residence time distribution was incorporated into the population balance model(tanks-in-series model).The model divides the COBC crystallizer into several tanks,and the tanks are connected in series to simulate the whole crystallization process.The residence time distribution was obtained through experiments,and the number of tanks corresponding to the crystallization condition was obtained.In order to obtain the desired crystal size distribution,a nonlinear programming optimization algorithm was proposed to control the tanks-in-series model to obtain the corresponding temperature drop curve.Finally,the crystallization process of glutamic acid is simulated according to the two crystallization models and control methods.The simulation results show that the model predictive controller based on Koopman can make the crystal size distribution reach the expected value quickly,which verifies the effectiveness,rapidity and superiority of the controller.Moreover,the tanks-in-series model controlled by the optimization method can also meet the requirements of the crystal.