Performance Assessment Algorithm And Application Of Non-gaussian Batch Control Systems

Due to its high production efficiency and flexible control methods,batch process are widely used in the fields of papermaking,medicine,metallurgy,food and beverages,chemical reactors,etc.Affected by noise and other factors,the batch control system cannot always meet the expected performance requirements in actual operation.The control performance of batch process affects the product quality,operating costs,etc.Therefore,it needs to be accurately evaluated in order to make decisions as soon as possible,thereby improving product quality and system operating efficiency.Therefore,the control performance assessment(CPA)of batch process has became a research hotsport recently.Under the assumption that the process variables obey the Gaussian distribution,the traditional CPA benchmark based on the minimum variance can accurately evaluate the performance of batch process control.However,process disturbance and measurement noise do not necessarily obey the Gaussian distribution.In this case,the variance index is not enough to describe its random characteristics and cannot effectively evaluate the control performance of non-Gaussian batch processes.In recent years,performance assessment index based on minimum entropy have been proposed for non-Gaussian continuous processes.However,this method cannot be directly used for the CPA of batch processes.Therefore,statistical information is used to study the CPA problem for non-Gaussian batch control systems,which has very important theoretical and practical significance,The main research contents of this thesis are as follows:(1)For the time-invariant non-Gaussian batch process,the minimum rational entropy(MRE)is employed to design the inner and outer controllers according to the batch error,and a new benchmark of minimum rational entropy performance assessment for the non-Gaussian batch control system is established.For batch processes with unknown models in actual industrial processes,a minimum rational entropy-based estimation of distribution algorithm(MRE-EDA)is proposed to identify the mathematical model of the batch process,and then according to the identified model the optimal controller is designed,and a new benchmark for performance assessment of the MRE is established,so as to achieve the purpose of performance assessment of the batch process.(2)The batch error of the time-invariant non-Gaussian batch process consists of two parts: deterministic error and stochastic error.The inner controller only affects the stochastic error,while the outer controller affects both the deterministic error and the stochastic error.This thesis analyzes the relationship between the deterministic error and the stochastic error and the convergence of the batch process.According to the deterministic error and stochastic error,the inner and outer controllers are designed based on the minimum rational entropy.And the inner and outer MRE-CPA are established.In addition,for batch processes with unknown models in actual industrial processes,an improved MRE-EDA method is proposed to identify batch process model parameters and estimate the PDF of non-Gaussian disturbances.Then design the best controller and establish the performance assessment benchmark,wich realize the performance assessment of the batch control system.(3)Since the batch process in actual industry is time-varying in nature,it is more difficult to design controller for a time-varying non-Gaussian batch process.In this thesis,analyzing the batch tracking error for the time-varying non-Gaussian batch process and design the mixture correntropy(MC)inner and outer optimal controllers,and establish a new benchmark for the performance assessment of the MC batch control system.In addition,for batch processes with unknown models,the traditional distribution estimation algorithm is improved,and the mixture correntropy estimation of distribution algorithm(MC-EDA)is proposed to identify time-varying non-Gaussian batch process model and estimate the non-Gaussian disturbance PDF.Furthermore,the optimal controller is designed for evaluating the batch process control performance of the time-varying non-Gaussian batch process.