Research On Data Rectification Technology Based On Historical Data

A large number of instruments are used for measurement,monitoring and control of the production indicators in modern industrial system.Accurate and reliable measurement data is the basis for production decision making,process control and plant management.However,in the actual production process,the measurement error and the existence of gross errors make the measurement data different from the actual situation.The data rectification technology is proposed to solve this problem.By using the redundancy between the measured data and a certain algorithm,it can get a group of new data which would meet the basic material balance,energy balance or chemical laws,the rectificated data are closer to the real value.This paper focus on how to use the historical data to extract the information and make data rectification.The main contents of this paper are as follows:(1)Research on principal component analysis(PCA):This paper analyzes the application of PCA in data rectification.In order to overcome the shortcomings of this algorithm in inability of gross error detection,the IMT method is introduced here,and PCA-MT,IPCA-MT iterative method algorithm are proposed.These algorithms can identify gross errors in the historical data,and obtain more accurate system models and reconciliation results.(2)Research on generalized T-distribution data rectification method:This paper analyzes how to use the generalized T-distribution to obtain the error distribution model,and how to use this model for robust data rectification.Compared with the least squares method and the Huber rectification method,it shows the advantage of the GT rectification technique when the error distribution model is ambiguous.(3)Research on swarm optimization algorithm:In order to identify the parameters of the generalized T-distribution model through the historical data,the feasibility of finding optimal solution by group optimization algorithm is analyzed.The advantages and disadvantages of the Particle Swarm Optimization(PSO)algorithm and the firefly algorithm(FA)in parameter identification are compared.(4)Research on Kalman filter method:The shortcomings of traditional Kalman filter are analyzed,and the EM-KF method is introduced and discussed here in its ability of gross error detection.The EM-KF method is improved by Sage-Husa adaptive Kalman filter method,and the EM-AKF method is proposed,which can realize error variance adaptive estimation and gross error detection.