Research On High Performance Filtering Algorithm For Vehicle Integrated Navigation System

With the rapid developments of modern science and technology, the requirements ofpeople for real-time and rapidity of the vehicle navigation calculation method are higher andhigher. The kalman filtering is a commonly used computational method in vehicle integratednavigation. But the application of kalman filtering requires that the kinematic system must belinear mathematical model. When the system is nonlinear model, if still using the kalmanfiltering method to carry on navigation calculation, the navigation solution will cause largeerror and even divergence. Therefore, in order to improve the calculation accuracy of vehicleintegrated navigation, research of the higher precision and nonlinear filtering method appliedto vehicle integrated navigation system is a very important and urgent task in the trafficinformation engineering and control field.Based on the study of existing navigation filtering algorithms, this paper presents a set ofhigh performance filtering algorithms. These new algorithms include nonlinear modelpredictive unscented particle filtering, fading memory square root unscented particle filtering,fuzzy robust adaptive unscented particle filtering, state dependent coefficient robust adaptivefiltering, nonlinear robust adaptive state dependent riccati equation filtering and sage randomweight adaptive filtering for kinematic model error. The proposed algorithms are applied tothe vehicle integrated navigation system and compared with the existing filtering algorithms,the results show that these algorithms have small amount of calculation, high precision. Theirfiltering performances improve significantly than the existing filtering algorithms.The main research work and innovative contributions are as follows.(1) This paper presents a new nonlinear model predictive unscented particle filteringmethod. The method considers model error of real-time estimation to correct the nonlinearand non-Gaussian system model, and then calculates by unscented particle filtering.Simulation experiment results show that the proposed method can improve navigationcalculation accuracy than model predictive filtering and unscented particle filtering.(2) Fading memory square root unscented particle filtering algorithm absorbs theadvantages of the fading memory filtering and square root filtering. This method uses thefading factor to adjust the information of the current measurement for estimation and reducethe influence of historical information, and then, the method replaces the covariance matrixwith the square root of the covariance matrix of iterative calculation to ensure symmetry andpositive semi-definite of covariance matrix. Simulation results and their analysis demonstratethat the proposed algorithm can improve significantly the filtering performance and accuracy.(3) Based on fuzzy control theory this paper presents a fuzzy robust adaptive unscentedparticle filtering method which absorbs the advantages of the unscented particle filtering,adapative filtering and robust estimation. The method considers the influence of the grosserrors in the observation vectors on filtering, and structures equivalent weight function basedon fuzzy theory, and then uses the useful information reasonably by the equivalent weightfunction and adaptive factor to control the influence of gross errors on navigation results. Theresults show that the proposed method can improve the accuracy and real time.(4) This paper presents a new robust adaptive filtering method based on state dependentcoefficient. The method adopts the state dependent coefficient to convert nonlinear systeminto state dependent system for readucing the errors caused by the linearization process. Usingequivalent weight matrix and adaptive factor to assign information reasonably, it can controlinfluence of abnormal dynamic model and abnormal observation on navigation system. Simulation results demonstrate that the performance of the proposed method improvessignificantly compared with the EKF and the UKF.(5) This paper presents a new nonlinear robust adaptive filtering method based on thestate dependent riccati equation technique. The method adopts the state dependent coefficientform to convert nonlinear system model into linear system for decreasing the errors caused bythe linearization process. It also demonstrates the stability on specified conditions.Experimental results show that the proposed filtering method can not only effectively resistdisturbances from nonlinear system state noise and observation noise, but it also can achievehigher accuracy than the EKF and SDRE filtering.(6) According to research of existing literatures, many methods use the arithmetic meanto estimate the covariance matrices of innovation vectors and observation residual vectors.The estimate of the observation noise vector covariance matrix containing state predictederrors. If the state predicted errors are large, then the predicted residuals will be large and thereliability of the estimations that are calculated by the covariance matrices of innovationvectors and observation residual vectors will decrease. This paper uses random weightingestimation method to estimate the covariance matrices of observation noise and state noise,with the aim of controlling the influence of the observation anomaly and the dynamic modelnoise anomaly. Experimental results demonstrate that the proposed method can improve thepositioning accuracy for dynamic navigation and reduce the amount of calculation.(7) This paper presents a sage random weighting adaptive filtering method to estimatekinematic model error. The proposed algorithm adopts the window smoothing method of sagefiltering to obtain covariance matrices of observational residual vector and predictive residualvector. By using the random weighting factor to adjust observation residual and predictedresidual to control observational residuals and predicted residuals for resisting thedisturbances of navigation accuracy. Experimental results demonstrate that the proposedfiltering algorithm can resist disturbances of state noise and observation noise.In this paper, the research results have a certain contribution to the vehicle integratednavigation filtering calculation, multi-source information fusion technology, error estimation,computer simulation and other fields. The research results not only can be applied to filteringcalculation in the fields of military and civilian on vehicle integrated navigation, but alsoprovide successful experiences and reference for other integrated navigation filteringcalculation in the fields of aviation and aerospace by extension.