Improvement Of Correlation-based Just-in-time Learning Method And Its Research In Chemical Process Model-free Control System

The chemical process control system is important in the modern chemical industry,but it is hard to design the process control system based on the first-principle model as the non-linearity,time delay,uncertainty and other issues of the chemical process.With the development of computer technology,the chemical process data can be easily measured and stored online.Therefore,the data-based process control design has attracted more and more attention by researchers and institutions.Among them,the Correlation-based JITL(Co-JITL)method is applied to a model-free adaptive PID control system for nonlinear chemical processes(JITL-PID)because of its simple model structure,easy update,and strong adaptability.In this thesis,the Co-JITL method is improved and used in the design of the JITL-PID control system.The main research work of this thesis is as follows:(1)Aiming at the problem of the historical database in the Co-JITL method,the sliding window method is introduced to improve the structure and update mechanism of the original historical database.Therefore,an improved Co-JITL method based on sliding window is proposed,and used in the model-free adaptive PID control system.Simulation experiments with two nonlinear chemical process cases are carried out,and the simulation results show that the control system designed based on the improved method proposed in this thesis not only has less online calculation time,but also has higher control accuracy and better robustness and stability.(2)When the Co-JITL method is used to design controller,it takes a lot of time to adjust the threshold and other parameters for different operating conditions,which makes the control system lack of convenience in use and may fall into over-fitting.In order to solve this problem,the upper limit of statistical analysis is applied to the Co-JITL method,thus two methods of adaptive threshold J and adaptive threshold Φ are proposed and are used in the model-free adaptive PID control system.Simulation results show that the control system based on the two improved Co-JITL methods proposed in this thesis will greatly reduce the offline optimization time when facing new operating conditions.Among them,the proposed adaptive threshold Φ control system has the best controller performance and robustness and stability.(3)In the modeling process of the Co-JITL method,Principal Component Analysis(PCA)will be used to deal with the correlation between data.However,the PCA algorithm is difficult to effectively extract the nonlinear relationship between the data.Thus,algorithms such as Kernel Principal Component Analysis(KPCA)and Principal Polynomial Analysis(PPA)have been developed.Therefore,the KPCA and PPA algorithms are introduced into the original Co-JITL method to replace the PCA algorithm,respectively.Therefore,the KPCA-based Co-JITL method and the PPA-based Co-JITL method are proposed and used for the model-free adaptive PID control system design.Finally,the simulation results show that the two new control systems proposed in this thesis have good control performance.