Research On Distributed State Estimation Methods For A Type Of Dynamic Targets

State estimation is the prerequisite for high-level information extraction and control,playing a pivotal role in the fields of process monitoring,tracking,unmanned aerial vehicles,etc.Therefore,it attracts much attention of scholars across various research areas.With the development of networking,communication,and computing technologies,the traditional single information detection and processing device,as well as centralized state estimation methods,are no longer applicable for some practical applications.Thus,the distributed state estimation based on the multi-agent cooperation,which has greater robustness,flexibility,and scalability,has been the subject of a surge of activities in the field of state estimation.In this paper,we consider the distributed state estimation for a class of dynamic targets,of which the evolution of states can be modeled as linear or nonlinear systems.Most of the existing works on this problem focus on the scenario with a single dynamic target,in which the proposed distributed state estimation methods cannot deal with the estimation for multiple coupled target states.Furthermore,a few studies on this problem consider the case where the multi-agent system suffers malicious attacks in the distributed state estimation process.However,the resilient distributed state estimation methods introduced in these works have limited applications due to the strict requirements on the state evolution models of the targets.As a result,to address these gaps,this paper follows the principle of studying from single to multiple targets and from simple to complex evolution models of target states,studying the distributed state estimation methods for a class of dynamic targets.First,we investigate the resilient distributed state estimation methods under deception attacks for the dynamic targets with linear and nonlinear state evolution models in the single-target scenario.Then,we focus on the two-target scenario,researching the distributed estimation methods for the coupled nonlinear-evolution states of two dynamic targets.Finally,we further extend our research in the two-target case to the multi-target scenario.The main works and contributions are as follows:1)Consider the distributed state estimation for the general linear time-invariant systems under deception attacks.To deal with the negative impacts on the estimation performance of the time-invariant deception attacks and the stability issue of the dynamic estimation error system resulting from the arbitrary system matrix of the target system,we propose a novel adaptive dynamic-target regulative gain estimation algorithm which is capable of detecting the attacks.In this algorithm,the adaptive gain and the regulative gain matrix are designed cleverly,which can restrict the impacts of the time-varying and arbitrary measurements on the estimate and guarantee the stability of the dynamic error system,respectively.Through defining the average estimation error and consensus estimation error,we analyze the estimation performance of the algorithm,and verify that all the agents(have been attacked or not)in the multi-agent system can achieve the state estimation task under the malicious deception attacks.2)Consider the distributed state estimation for the moving robots with nonlinearevolution states under deception attacks.To deal with the negative impacts on the estimation performance of the time-invariant deception attacks,the difficulties of the estimation algorithm design caused by the nonlinear evolution model of the target state,and the challenges in the estimation accuracy resulting from the measurement collection which is not continuous,we propose a novel invariant extended Kalman filter-based resilient distributed state estimation algorithm based on the exponential mapping between the Lie group and corresponding Lie algebra.This algorithm avoids the introduction of the estimation biases induced by the linearization of the target’s nonlinear state evolution model.Moreover,it can monitor the occurrence of the attacks and the collection of the measurements,restricting the impact of the anomalous measurements by the adaptive gain whenever an attack is detected.Via the convergence analysis of the estimation error,it is illustrated that all the agents in the multi-agent system can recover the target robot’s state.3)Consider the distributed state estimation for multiple dynamic targets,among which the information interactions exist.The states of these targets are coupled and evolve according to nonlinear models.First,for the case with two targets,to deal with the difficulties of the information collection and fusion performed by the multi-agent system induced by the coupling of the targets’ states and the challenges in the estimation algorithm design,we divide all the agents into groups such that agents in different groups observe various targets,and propose consensus-based distributed two-target state estimation algorithms using the measurements collected by the agents.These algorithms fuse the information from different groups of agents by the covariance intersection fusion rule separately,which guarantees the consistency of the algorithm.We further extend to the multitarget(more than two targets)case,proposing consensus-based distributed multitarget state estimation algorithms.The algorithms repeat the consensus iteration,fusing the local estimates of more agents and improving the estimation accuracy.