Research On Fusion Filtering For Multi-sensor Stochastic Systems Via Incomplete Information

With the rapid advancement of network communication capabilities and continuous improvement in sensor technology,the interaction information among sensors can be transmitted either by using the wired or wireless communication.This development has further enhanced the information processing capabilities of multi-sensor systems that have allowed them to operate to some extent without being constrained by time and space.As a result,the multi-sensor systems have been applied in various fields such as remote monitoring of physical systems,target tracking,and industrial manufacturing.In most applications,it is common to encounter noise in the data.The filtering algorithm is used to effectively handle the issue of the noise,thereby improving the accuracy of state estimation.Therefore,in order to further obtain more accurate information,the problem of the multi-sensor fusion filtering(FF)has attracted widespread attention.At the same time,when the measurement information is transmitted in a networked environment,the incomplete information can be caused by sensor delays,missing/fading measurements,information transmission strategies,sensor resolutions,measurement outliers,etc.,which poses challenges to the FF problem for multi-sensor systems.At present,the study on multi-sensor FF problems based on incomplete information is still limited.Therefore,this thesis mainly discusses the impact of the incomplete information on multi-sensor stochastic systems(SSs).Furthermore,it proposes FF methods for multi-sensor SSs and presents boundedness criteria for filtering errors,which provides effective methods for addressing the FF problems of multi-sensor SSs.The main work of this thesis includes the following aspects.Firstly,by considering the incomplete information situations caused by random sensor delays and information transmission strategy,the resilient FF problem is first discussed for multi-sensor nonlinear time-varying SSs with random sensor delays.The random variables with specific statistical information are used to describe the random sensor delay phenomenon and the gain perturbation of filter,respectively.Posteriorly,in order to reduce the communication burden,the Round-Robin transmission strategy is introduced into the network channels,and a predictive compensation method is used to further decrease the computational complexity of the algorithm.Using relevant matrix inequalities,the recursive equations of the upper bounds on the filtering error covariance(FEC)are established for the above two classes of systems,and filter gains are designed under the minimum case of upper bounds of local FECs.Based on the relevant information of each filter,the corresponding distributed FF algorithms are proposed using matrix-weighted fusion criterion and inverse covariance intersection fusion criterion.Finally,the boundedness problems of two upper bounds of the FEC are studied and boundedness criteria are provided.Secondly,by considering incomplete information situations caused by missing measurements,fading measurements and information transmission strategy,the first part of the research focuses on the distributed FF problem for multi-sensor nonlinear time-varying SSs with missing measurements and randomly occurring uncertainty.Two series of Bernoulli random variables are utilized to depict the random occurrence of the uncertainty and missing measurement phenomenon,respectively.Next,the distributed FF issue is further investigated for a class of nonlinear delayed time-varying SSs under influences of the stochastic communication strategy and fading measurements.The random variables obeying specific statistical properties are used to describe the fading measurement phenomenon,and the stochastic communication strategy is used to reduce the transmission pressure of data in the communication process.By analyzing the above two types of systems,the recursive equations of upper bounds of corresponding local FECs are respectively established.The evaluation criterion is to minimize the local upper bounds,and suitable local filter gains are designed accordingly.The distributed FF algorithms are presented using the matrix-weighted fusion criterion and inverse covariance intersection fusion criterion.Finally,the boundedness of upper bounds of FECs is analyzed and the corresponding theoretical proofs are provided.Thirdly,by considering incomplete information situations caused by information transmission strategy and sensor resolutions,the problem of the distributed FF is discussed for a class of multi-sensor descriptor nonlinear time-varying SSs with encoding-decoding strategy and sensor resolutions.The full-order transformation method is used to convert the descriptor nonlinear stochastic system(SS)into a non-descriptor nonlinear SS.The encoding-decoding strategy is used to improve the security of data in the communication process under the case of limited bit rate.The covariance matrix of local filtering error is computed and its upper bound is iteratively obtained with the iterative form.The filter gain is designed with the optimization criterion of minimizing this upper bound.Finally,the boundedness issue of the upper bound on the FEC is discussed and the corresponding discrimination criterion is provided.Fourthly,by considering incomplete information situations caused by information transmission strategy and measurement outliers,the problem of the distributed FF is investigated for a class of multi-sensor descriptor nonlinear time-varying SSs with dynamic event-triggered strategy and measurement outliers.The full-order transformation method is used to convert the descriptor nonlinear SS into a non-descriptor nonlinear SS.A dynamic event-triggered strategy is used to decrease the number of information transmissions during communication.Consider the occurrence of measurement outliers and construct an anti-outlier filter with a saturated innovation structure to reduce the impact of outliers on estimation accuracy.By using the correlation matrix inequalities,the local FEC and its upper bound are obtained,and the filter gain is designed in the light of the minimum value of the local upper bound as the performance index.Then,the results regarding the FF are obtained by means of the inverse covariance intersection fusion criterion.Finally,a sufficient condition is given to ensure that the local upper bound on the FEC is bounded uniformly.Fifthly,by considering incomplete information situations caused by information transmission strategy and external disturbance,the FF filtering problem is investigated for a class of multi-sensor descriptor nonlinear time-varying SSs with weighted try-once-discard strategy and unknown dynamic bias.The full-order transformation method is used to convert the descriptor nonlinear SS into a non-descriptor nonlinear SS.A weighted try-once-discard strategy is introduced to reduce the number of network congestion occurring in the network,and the local filter with an augmented structure of state-unknown bias is constructed.Furthermore,the recursive expressions of the local FEC and its upper bound are derived based on the filter,and the expression of the local filter gain is obtained in terms of the goal of minimizing the trace of this upper bound.Then,the distributed FF algorithm is proposed using the covariance intersection fusion criterion with the help of the related information of the local filter.Finally,a sufficient condition with respect to the boundedness of the obtained upper bound on the FEC is given.