Research On State Estimation And Fusion For Uncertain Parameters Coupling Systems

The state estimation and fusion for dynamical systems in manoeuvering target tracking,networked system control,multi-sensor decentralized localization and tracking under complex environment,are challenged by multiple uncertain parameters coupling characteristics,such as the multi-mode,nonlinearity,noise correlation,additional and multiplicative noises coexistence,networking and distributed architectures.On the basis of modeling the multi-mode uncertainty obeying a Markov chain,the corresponding systems are always transformed into uncertain parameters coupling systems along with Markov jump processes.Thus,accompanied with Markov jump processes,it has the theoretical significance and practical applications to develop a series of state estimation and fusion methods for uncertain parameters coupling systems.In this thesis,focused on the dynamical state estimation and distributed fusion along with Markov jump processes,the main contributions are as follows:1.For the Markovian jump linear systems with stochastic coefficient matrices,the corresponding linear minimum mean square error(LMMSE)estimator is derived and the stability of the derived LMMSE estimator is discussed.Based on the derived LMMSE estimator,a general filter framework about state estimation and data association in clutters is established for tracking manoeuvering targets,through reconstructing a novel measurement equation with stochastic coefficient matrices.The proposed method obtains a more accurate estimate than those of the corresponding interacting multiple model probabilistic data association filter and interacting multiple model joint probabilistic data association filter for tracking a single maneuvering target under Gaussian and Glint noises and two maneuvering targets,respectively.2.For the Markovian jump systems with colored measurement noises obeying an autoregressive process,a new measurement equation with adjacent two Markov jumping parameters is reconstructed through exploring the left zero divisor,perturbed by mutually independent,Gaussian white noises.Based on the definition of a novel hypothesis set constituted of all possible values of two adjacent Markov jumping parameters,the posterior density about state estimation is derived recursively.By pruning the Gaussian components with too small weights appropriately,the adaptive Gaussian mixture recursion about approximating the posterior density is realized on the basis of measuring the Kullback Leibler divergence.In the maneuvering target tracking accompanied by range gate pull-off,the proposed method achieves a better estimated accuracy than those of the related interacting multiple model method and Gaussian sum filter for Markov jump nonlinear systems,when considering the colored measurement noises.3.For the Markovian jump linear systems with random parameter matrices and crosscorrelated noises,the recursive LMMSE estimator is proposed under a centralized framework,and the corresponding square-root array implementation is presented.Furthermore,the distributed fusion estimation with square-root array implementation is derived,incorporated with consensus strategy.Meanwhile,the convergent condition and the convergence point are also discussed with respect to the consensus strategy.In the maneuvering target tracking in sensor networks,the proposed method obtains a more accurate estimate than those of the LMMSE estimators for related Markovian jump systems,with multiplicative noises and cross-correlated noises coexisting simultaneously.4.For the networked nonlinear systems including the sensor network,transmission network and processing network,by considering multi-step random measurement delays obeying a Markov chain,a novel Gaussian recursive filter with estimating the posterior probability of delay online is presented in each local processing unit.Then,the distributed Gaussian-consensus filter is proposed by using the consensus strategy among processing neighbors to share information in processing network.The proposed method gains a better estimate accuracy that those of modified unscented Kalman filters with randomly delayed measurements in a single processing unit and distributed processing network under one-step and two-step randomly delayed measurements conditions.5.For the nonlinear systems with randomly delayed fading measurements and multiplicative noises in sensor networks,on the basis of Gaussian mixtures approximating the posterior densities appropriately,a novel Gaussian-mixture recursive filter is given in a single processing unit.Moreover,a distributed Gaussian-mixture information filter is proposed via consensus strategy based on the statistical linear regression applied to nonlinear measurement equations.By considering the coexistence of measurement random delay,channel fading and multiplicative noises,the proposed method obtains a more accurate estimate than those of Gaussian approxi-mated filters with randomly delayed measurements in both a single processing unit and distributed processing network.