Performance Monitoring Methods Of Feedback Control System Based On Information Entropy

During actual industrial manufacturing processes,feedback control system performance may get worse due to a variety of problems such as poor controller design,actuator failure,sensor degradation,and mismatch of model objects.Therefore,performance monitoring needs to be conducted.This secures that any potential performance degradation in the loop can be fed back to the relevant staff in a timely and effective manner for further actions.The main practical performance monitoring methods currently used in the industry are minimum variance and its improved index.This method is limited as it requires a priori knowledge of the model and is not very sensitive to the non-Gaussian and non-linear cases.Over the last few years,information entropy measures have found some applications in industrial processes due to their excellent ability to quantify stochasticity.This paper develops a study for this situation,with the following main research parts:(1)The basic theory of feedback control system performance monitoring is presented in detail.Considering the possible non-stationary factors in the actual implementation,a minimum variance indicator obtaining methodology based on the time series analysis technique is developed and validated.The simulation results show that this model correctly calculates the minimum variance index under both stationary and non-stationary conditions.It shows that the method is feasible.(2)A detailed description of the minimum information entropy-based performance monitoring index for control systems is presented.On this basis,an improved approach to obtaining a minimum entropy index based on a time series analysis model is proposed.This method adopts the entropy of the feedback invariant as a benchmark and employs a regression model to calculate the residual for estimation of the feedback invariant while fitting continuous information entropy values using kernel density estimation.An exponential function mapping is used to construct a minimum entropy index.Simulation results show that this method is effective for performance monitoring in both Gaussian and non-Gaussian cases.However,as continual entropy is translation invariant,the index still requires an estimate of process noise,making this method less useful in industrial processes.(3)A detailed description of the performance monitoring index for control systems based on improved relative entropy is presented.This method requires only process data and removes the need for a priori knowledge of the model,which makes it highly flexible and practical.The original relative entropy measurement is first improved to a symmetric,bounded performance monitoring index.The calculation of the index is then completed by Monte Carlo simulation and kernel density estimation.Simulation results of the consistency test,non-Gaussian noise test,and control valve non-linearity test all indicate that the index has a high sensitivity to non-Gaussian and non-linear processes.Compared with the minimum variance index,it can correctly reflect the actual performance degradation of the systems.(4)The method studied in this paper is validated in the actual thermal process data of a power plant.Through the process of obtaining real-time monitoring signal output,dealing with the actual information,and calculating the indexes by sliding window method finally giving the real-time performance monitoring results.The results show that the proposed method in this paper has some feasibility in the actual system.