Particle Filter Based Methodology Of Robust State And Parameter Estimation For Nonlinear System

The state and parameter estimation of nonlinear system plays an important role in the process control and fault detection.In recent years,in the aspects of state and parameter estimation of nonlinear system,particle filter methods get extensive attention.Its basic thought is using a Monte Carlo method to create a large number of random samples to approximate state posterior distribution thereby to realize the state estimation.This method can be applied to arbitrary state space model.However,there are still some problems of particle filter based methods in the aspects of state and parameter estimation.This paper researches the problems existing in general state and parameter estimation.1.The development of particle filter,the application field and the existing problems are introduced.Based on the state space model of nonlinear system,the sequential importance sampling method is introduced in the unified framework of the recursive Bayesian filtering theory.The main problems existing in this method,namely particle degradation phenomenon,are pointed out.And then,the re-sampling method based on the particle degeneration phenomenon is introduced.The general particle filter method is formulated.Three kinds of particle filter methods are presented according to different application environment.2.The paper introduces the general state estimation method based on the particle filter,and points out the general problems existing in the state estimation method,namely,the initial value uncertainty leads to poor effect of general state estimation.Based on this,it proposes a multiple population particle filter method,which decides whether to enter the multi-group particle filter method or not according to the observation error leaded by the initial value uncertainty of status.The method shortcoming is time-consuming,than a robust state estimation method is proposed,which decides whether to adjust the initial value or not according to the observation error leaded by the initial value uncertainty of status.Robust state estimation method not only solve the problem that the initial value uncertainty leads to poor effect of general state estimation,and solving the shortcoming of the multi-group particle filter.3.This paper introduces the general state and parameter estimation based on particle filter,and points out the problems of general parameter estimation,the problem is that the estimate parameters cannot track model once iteration model parameters change.Based on this,it proposes a robust parameter estimation method.The method decides whether to adjust the variance of parameters or not based on the observation error generated by model parameter changes.The volatility of this parameter estimation method is larger than general parameter estimation method,but it can respond quickly and track accurately when the model parameters change.