Research On Performance Evaluation Method Of Non-Gaussian Stochastic Distributed Control System

Control system performance evaluation plays a key role in industrial control systems.In order to meet the real-time requirements of modern control systems,it is necessary to evaluate the performance of the system quickly and accurately.For the control and performance evaluation of Gaussian systems,the minimum variance criterion is the core idea.Whether it is a univariate system or a multivariate system,the research on the minimum variance index is very mature.However,there are non-linearities and uncertainties in actual industrial processes.The output of the system follows a non-Gaussian distribution,and the mean and variance can not fully describe the characteristics contained in the output of the system.The disturbance of the system is often random and can be subject to arbitrary random distribution,and even if the noise of the system obeys the Gaussian distribution,it may often produce nonlinear because of the running time of the system,which makes the traditional modeling and control method based on Gaussian hypothesis can not meet the needs of the stochastic distributed control system,resulting in the deviation of the performance evaluation method with variance as the evaluation index.In stochastic systems,entropy is widely used,but rarely studied in performance evaluation of control systems.Therefore,this paper proposes several methods for evaluating the performance of non-Gaussian stochastic distributed control systems.This paper first introduces an improved distribution estimation algorithm.In the traditional distribution estimation algorithm,the preliminary estimation of the population and the cross-operation of the population are added to improve the speed of the algorithm.The improved EDA algorithm combines the two different entropy calculation methods of rational entropy and Renyi entropy to obtain the minimum entropy(baseline entropy)of the system,which improves the accuracy and speed of the method.Compared with the rational entropy,the Renyi entropy The calculation method is simpler.And in order to avoid the problem of invariance of the minimum entropy translation,the limit of the output mean is combined in the index.The two methods are compared and verified by a single-loop feedback control system.Secondly,in order to better meet the real-time requirements of modern industry and avoid the problem of translation invariance,the performance of non-Gaussian systems based on mixture correntropy is proposed under the influence of mixture correntropy filter design in non-Gaussian control systems.Combining the mixture correntropy with the improved distribution estimation algorithm,the estimated parameter values of the system and the estimated distribution of noise are obtained,and then the reference entropy is obtained.Finally,the final index of system performance evaluation is given.Finally,the performance evaluation method of some special non-Gaussian random distribution systems is introduced.When the expected output distribution of the system is known,the mixture correntropy is directly used as an index to evaluate the performance of the non-Gaussian random distribution control system whose expected output distribution is known.And proposed a superposition area index based on probability density function.The above two indexes are used in the performance evaluation of a fully decoupled multivariate random distributed control system.