Research On The Performance Index Of A Moving Horizon Estimation Method

State estimation is a method of estimating system state based on measurement data,which is widely used in many industrial application fields such as automatic control,fault diagnosis and soft sensor.Moving horizon estimation is a filtering algorithm which uses the rolling window data to estimate the state.It can explicitly deal with state estimation problems of constrained systems.In the process of application,the selection of window size and the calculation of arrival cost are two important aspects that affect the estimation effect of the algorithm.To improve the solution efficiency and estimation accuracy,this paper focuses on studying the reasonable selection of window size and the approximate calculation of arrival cost.The typical numerical example and actual case are analyzed and verified.On the basis of reading a large number of literatures,this paper has carried out some research work on moving horizon estimation algorithm,which is as follows:1.Aiming at the optimization problem of moving horizon estimation algorithm in linear systems,the corresponding method for solution is proposed by transforming the full information estimation problem and the approximate estimation problem into the quadratic programming problem described in matrix form.Based on the least square principle,a recursive method of moving horizon state estimation is presented.By rolling forward the batch output data,blocking the matrix with system parameters and the matrix with output information respectively,the estimated state value is given in the form of recursion like the Kalman filter.Finally,the method is applied to the linear model of white noise and colored noise respectively to verify the validity and universality of the method.2.In order to solve the problem that window size will affect the estimation accuracy and computational efficiency of moving horizon estimation algorithm,a selecting method of window size is proposed to balance the two indexes.On the basis of the excellent characteristics of genetic algorithm and simulated annealing mechanism,the fitness function of precision index and efficiency index of different dimensions is designed by using the principle of normalization,and the optimal window size under the current weight ratio is obtained.The linear unconstrained discrete model and the continuous stirred tank reactor model with constraints are taken as examples to verify that the parameter optimization method can effectively calculate the optimal window size under the preset weight ratio for the application of moving horizon estimation.3.To solve the problem that the application of moving horizon estimation algorithm in nonlinear system,a method of calculating arrival cost by obtaining prior covariance matrix through the unscented Kalman filter is presented.Firstly,the principle of simplex sampling is used to replace the unscented transformation of symmetric distribution,and the number of sigma points in the original method is reduced.Secondly,an adaptive calculation strategy is proposed for the scale factor in unscented transform to improve the calculation accuracy of arrival cost.The simulation results of two nonlinear cases show that this method has more accurate estimation effect and is more suitable for higher order nonlinear systems than the extended Kalman filter.