Distributed Fusion Estimation For Multi-Sensor Discrete-Time Stochastic Systems

Multi-source information fusion, also known as multi-sensor information fusion or multi-sensor data fusion, is an emerging technology. In recent years, multi-sensor in-formation fusion has received significant attention and has been applied to military and nonmilitary areas. The estimation fusion theory is an important branch of information fusion theory.There exist two important approaches to estimation fusion. The first one is the centralized fusion method, which can give the globally optimal estimate by directly combining local measurements, but it is computationally expensive. The second one, i.e., the distributed fusion method, is able to give a sub-optimal estimate by weighting the local state estimates. Compared with the first method, it can reduce computation burden and increase the rate of input data significantly. In addition, the distributed fusion method is able to facilitate fault diagnosis and isolation.In this thesis, the distributed estimation fusion problems for several classes of discrete-time linear stochastic systems are studied. The main innovations and contents of this thesis are as follows.1. The distributed fusion problem for the state estimation of discrete-time stochas-tic singular systems is addressed. In specific, the multi-sensor stochastic singular sys-tem with correlated noises is converted into a group of non-singular systems. The recur-sive predictor, filter and smoother are presented for each of these non-singular systems. Further, the local full-order filters and smoothers are given for the original singular system. The final full-order fusion filters and smoothers are obtained for the original system based on the optimal weighted fusion algorithm in the linear minimum variance sense.2. The distributed fusion problem for the state estimation of discrete-time stochas-tic singular systems with multiple time delays is considered. There are multiple time delays in both the state equations and the output equations in these systems. Similar to the above singular system, the system is converted into a group of non-singular systems with delays. Using projection theory and the innovation analysis method, the local filters and smoothers are given for the original singular system. The final distributed fusion filters and smoothers are obtained for the original system based on the optimal weighted fusion algorithm.3. The distributed optimal fusion problems for the state estimation of three classes of multi-sensor systems with packet dropouts are studied. Each class of these systems have their own model of packet dropouts. The local filters, local filtering error covari-ances and cross covariances are given. Based on the optimal weighted fusion algorithm, the distributed optimal fusion filters are obtained.4. The distributed fusion problems for the state estimation of four classes of multi-sensor systems with parametric uncertainty are investigated. First, for systems with polytopic uncertainty, the distributed fusion robust H2 and H∞filtering problems are considered. Second, the distributed optimal fusion filtering problem for systems with random parameters is studied. Third, for systems with random parameters and packet dropouts, the distributed optimal fusion filtering problem is considered. At last, the dis-tributed optimal fusion filtering problem for uncertain systems satisfied "Sum Quadratic Constraints (SQC)" is examined. The distributed fusion filters for these systems are ob-tained.5.The distributed optimal fusion problem for the state estimation of multi-sensor discrete-time stochastic systems with additional state equality constraints is consid-ered. After obtaining the filtering error cross covariance between local filters, by the distributed optimal fusion algorithm weighted by matrices, the constrained distributed optimal fusion filter is derived. A numerical example shows that the fusion filter is better than each local filter.