Research On State Estimation And Fusion Algorithm For Constrained,local Strongly Coupled Systems

With the expansion of scale as well as complexity in reality,for issues regarding information science and intelligent control,engineering researchers always hope to use multi-agent cooperation and data fusion to contribute to an integrated state estimation with better performance.A type of higher intellectual "local strongly coupled" systems is introduced to describe the interaction of agent nodes for the purpose of autonomy and distributed structure,where each agent exchanges information about state estimation only with its coupled agents in a certain neighbourhood to fulfill one task or come to a decision.Eventually,in terms of state estimation(filtering)and feedback control,a unique,correct state estimation will be achieved,which is called "consensus" as well.Based on distributed,recursive state estimation,through information exchange and adaptive weighted fusion,with time going,consensus will be gradually acquired.In the modern control theory,Kalman filtering with its extended variants,is a recursive state estimation with relatively lower costs both in computation and storage.Having been proved optimal with least mean square error,Kalman filtering is applied with feasibility to problems such as integrated navigation and guidance.While in some practical instances,some a priori information,also known as "constraints" that determines quantitative relationship of state components,is frequently neglected.Constraints can be derived from physical theorems,geographical arrangements or multi-agent alignments set in advance,and will directly cause changes in the probabilistic structure of the filtering model.Therefore,estimation values from traditional Kalman filtering are not truly optimal,and we should take constraints into account.Both at home and abroad,a series of constrained filtering strategies have been studied,but a most evident limitation lies in that former researchers tended to take constraints for granted,and few original works were delved into their stochastic or imperfect characteristics.In this thesis,state estimation under constraints is applied thoroughly in local strongly coupled systems.Innovative amendments and design of algorithms are mainly concentrated on the aspects as follows:1)Description of the constrained filtering and fusion process in a local strongly coupled system is illustrated.Following the discrete protocol of consensus,we have mentioned the consensus index,and have explained the mechanism of state update and local information exchange.2)Modelling of "partial information" considering incomplete measurements as well as abnormal estimation is mathematically proposed;with methods like system error registration,covariance matching of innovation,and simplified Sage-Husa adaptive filtering,this thesis has come to the realization of fault diagnosis and handling,and a more reliable and self-consistent structure for data fusion.3)When there exist unequal relationship in a priori constraints,by referring to the active-set optimization,this thesis has firstly provided a solution to the equivalent problem with equality constraints.Then based on the exact-penalty-function method in search of better feasible solutions,typical cases with inequality constraints are resolved.And last but not least,for potentially imprecise constraints,using real-time measurements,reliability of the acknowledged constraints is concretely defined and measured,and the overall estimation will become more rationally weighted.