Research On Sampling Nonlinear Filter

Sampling nonlinear filter technology is widely used in practical systems.With the improvement of sensor accuracy,the filter algorithm which repeatedly uses the measurement with the predicted value plays an increasingly important role in real life.In this paper,the deterministic sampling filters and random sampling filters are studied.The main research results are as follows:1?Deterministic sampling iterated filter.Since the coupling relationship between estimation and measurement noise is not taken into account in the iterative process of the traditional iterated filter algorithm,it has the problem of low estimation accuracy.In order to obtain more accurate filter estimation,a deterministic sampling iterated filter(DSIF)algorithm is proposed in the nonlinear system.In the iteration process,the coupling relationship between estimation and measurement noise is fully considered.Means of measurement noise,covariance of measurement noise and covariance of estimation and measurement noise are updated as the estimators to be estimated.Then the complexity of DSIF algorithm and traditional nonlinear iterated filter algorithm is analyzed.Finally,The results of various simulation scenarios show that the proposed algorithm can improve the filter estimation accuracy.2?Measurement conversion nonlinear filter.In nonlinear systems,the nonlinear filter(NLF)algorithm,which only uses measurements to update the predicted values,does not make full use of the measurements,resulting in a slightly worse filter estimation effect.In order to improve the accuracy of filter estimation,a measurement conversion nonlinear filter(MCNLF)algorithm is proposed.The proposed algorithm converts the measurements several times,then updates the filter estimates with the transformed measurements in turn,and fully considers the coupling relationship between the estimation and the measurement noise in the updating process.The simulation results show that,compared with NLF algorithm,the proposed MCNLF algorithm can obtain more accurate filter estimation,and it has good scalability.3?Multi-proposal distribution model sampling particle filter.Particle degradation and impoverishment exist in particle filter(PF)algorithm,which results in inaccurate filter estimation.In order to obtain better filter estimation effect,starting with the important density function,a multi-proposal distribution model sampling particle filter(MPDMSP)is proposed.The proposed algorithm obtains the distribution of proposals in the form of weighted proposal distributions,then obtains the weight of each proposal distributions by calculating the likelihood of each particle point,and finally obtains the estimation and its covariance in the form of weighted sum of particles.The simulation results show that the proposed algorithm can achieve better filter estimation effect under the lower operation time.