Linear System Modeling And State Estimation Of Missing Data

Missing and unobservable data are problems caused by multiple sampling frequencies,limited equipment,and bad environment in the process of system data acquisition and processing,resulting in deviations in data processing and analysis results,and providing wrong mathematical models for designing system control algorithms.EM algorithm can improve the quality of data processing with missing data,and is one of the effective methods for system modeling and parameter identification.The EM based Kalman filter(EM-KF)applies RTS smoothing algorithm for state estimation,and the estimation results are more accurate.This algorithm is widely used in areas such as blind source separation,but the existing application of the EM-KF algorithm with the stochastic state space model is very narrow.The actual industrial system often uses a more generalized state space model,which is more suitable for input signal control.The EM-KF algorithm can not only handle the system with missing data,but also have a good application to the generalized state-space model with input signal.The control input is considered in the stochastic state-space model as the generalized state-space model.The EM-KF algorithm is modified for different data missing pattern,and the relevant RTS smoothing algorithm is employed to fit the new model.In this paper,the proposed algorithm is verified and applied to the industrial dryer system.The algorithm is compared with the subspace algorithm and the result is slightly better.Furthermore,the effectiveness of the proposed algorithm under random data loss with different proportions of data is verified.