Control Performance Assessment And Stiction Diagnosis For Non-Gaussian System Based On Entropy

In modern process industry,the control performance of the control loop has become an important factor affecting product quality and energy consumption.The decline of control performance caused by the oscillation of the control loop will lead to excess or unqualified product quality and increase of energy consumption of the device.Accurate and timely evaluation of control performance and diagnosis are not only the hot spot of theoretical research and the urgent demand of industry,but also have become an important part of process intelligent factory construction.Aiming at the problems of incomplete signal information and poor evaluation effect of existing control performance evaluation methods in non-Gaussian systems,a variable coefficient of variation minimum entropy performance evaluation index is proposed and designed to improve the accuracy of control performance evaluation results.At the same time,the maximum information coefficient and intelligent optimization are introduced to improve the detection method for the sticking fault detection of the regulating valve.The main work of this paper is as follows:Aiming at the problem that the minimum variance performance evaluation method only considers the noise second-order signal and lacks the high-order moment information of the noise,the minimum rational entropy and coefficient of variation were introduced to evaluate the performance of CCV-MEC.The index contains the comprehensive signal information and the constraint of the index on the mean value and the degree of dispersion,which overcomes the mean deviation problem of entropy.Then,the minimum entropy performance benchmark was derived based on the structure of cascade loop.The simulation results showed that the performance index can correctly reflect the loop performance decline caused by three kinds of factors: poor controller effect,actuator viscosity and noise disturbance,and improve the accuracy and adaptability of the evaluation.In order to improve the parameter identification accuracy of non-Gaussian loop based on entropy evaluation benchmark,A Hybrid distribution estimation algorithm based on clustering is proposed.This method overcomes the problem of incomplete identification method information in non-Gaussian system,and obtains the probability distribution information of noise at the same time,improves the accuracy of model parameter identification,and improves the accuracy of performance evaluation results.The idea of K-means clustering and mixed update was introduced to improve the EDA,which improved the global search ability of the algorithm,especially in the case of multiple extreme values.Simulation results showed that the accuracy of parameter identification based on CH-EDA was higher.Aiming at the stiction detection problem of regulating valve which leads to the decline of control performance,a convex polygon valve stiction detection method based on optimization was proposed.The OP-PV ring detection and the boundary correction variable were proposed in the method.It reduced the dependence of the algorithm on special values and the limitation of the standard equation,and improved the fitting effect and the accuracy of the viscosity quantitative detection results.Aiming at the problem of sticition detection of multi-loops,a scheme of stiction location and comprehensive detection was designed based on the maximum information coefficient.The position of stiction fault loop in the system is determined by nonlinear and correlation.And then the fault loop was detected by various methods and the comprehensive results were obtained.The integrated detection scheme can locate and detect stiction faults in multiple loops.The improved algorithm is verified by TE process simulation and actual industrial data set.The proposed and improved methods have been tested and verified on industrial data,and good results have been obtained.It provided new and effective methods for the control performance assessment of non-Gaussian system and the diagnosis of valve stiction.