Research On Data Based Integrated Iterative Learning Control For Batch Processes

The modern process industry is evolving from the production of basic materials in large quantities to the production of many varieties of high-quality professional products in small batches.Batch processes meet the above development needs,and they have advantages of small equipment investment and flexible production.Batch processes have been widely used in petrochemical industry,semiconductor materials,polymer materials and other manufacturing industries.They play an increasingly important role in the modern process industry.However,batch processes are characterized as multiphase repetitive operation,unstable running and mixed discrete and continuous variables,which makes traditional control theories difficult to apply.How to study the control theories and methods for batch processes based on their characteristics is an urgent problem to be solved.By using the theories of just-in-time learning(JITL),iterative learning control(ILC)and model predictive control(MPC),this thesis studies the integrated iterative learning control method for batch processes based on the essential characteristics,including batch to batch convergence and disturbance restraint within a batch.Moreover,the convergence,stability and robustness of the control system are analyzed theoretically.The main research results of this thesis are given as follows:(1)A model identification method based on JITL with hierarchical searching for batch processes is proposed.Local models are built to describe the nonlinear process by using the JITL method.It realizes the approximation of the nonlinear batch process.Firstly,the database of JITL is divided into several sub-databases based on the repeatability of batch processes.Secondly,considering the angle and distance relationship between the data,the sub-database is divided into several clusters based on the similarity between the data,and thus a three-layer searching framework is constructed.Finally,the least square method is used to identify the system model parameters.The proposed method reduces the complexity and computation of the JITL method,and improves efficiency of modeling.In order to improve the adaptive ability of the algorithm,the database is updated on-line by using the system real-time data.The proposed method can solve the problems of large computation and time consuming in the modeling of batch processes,and provides a research basis for the optimal control of nonlinear batch processes.The simulation results show that the proposed method performs well on nonlinear approximation.(2)Based on previous research results of 1,a quadratic model predictive iterative learning control method for batch processes is proposed.MPC and ILC are integrated into a quadratic objective function to realize the combination of real-time rolling optimization within a batch and control sequence iteration from batch to batch.The quadratic objective function optimization problem can be transformed into the solution of linear matrix inequality(LMI),and then the system control law with input constraint is obtained.The real-time feedback information is used to update the control input.Therefore,the external disturbance can be rejected.Simulation results demonstrate that the real-time control performance of the system is improved.(3)Based on previous research results of 1 and 2,an integrated batch-axis and timeaxis learning control method for batch processes is proposed.In the integrated learning control framework,the model-based ILC method is used to ensure the convergence of the system on the batch-axis,and the shrinking horizon MPC method is used to reduce the influence of external disturbance on the time-axis.Moreover,the tracking error information is combined with the feedback information of the current batch to realize the real-time adjustment of control variables,and then two-dimensional control for batch processes is achieved.Simulation results demonstrate that the proposed method can effectively restrain the nonrepetitive disturbance.(4)Based on previous research results of 3,considering model parameter uncertainty of the batch process,a quadratic robust predictive iterative learning control method for uncertain batch processes is proposed.Then in order to improve the process ability of disturbance rejection,a batch-axis and time-axis robust iterative learning control method for uncertain batch processes is proposed.Based on the analysis of the uncertain system model,the quadratic performance optimization index for robust iterative learning controller on the batch-axis is designed to ensure the robust convergence of the system with model perturbation,and then the quadratic optimization problem can be solved via LMI approach.In the direction of time-axis,a quadratic robust model predictive controller is designed,and it is combined with a robust iterative learning controller to improve the convergence rate of the system output.Then considering realtime feedback,a forgetting factor is introduced to make the data from the recent batch runs have more weight than that from the earlier batch runs,and the model is modified on-line based on the information of historical batch runs.It realizes the update of the model predictive controller output.Simulation results demonstrate that the proposed control method improves the robustness of the system.