Data-driven Consensus Filtering In Distributed Sensor Networks

Based on the data-driven theory and the methods of recursive least squares and distributed consensus filtering,some practical problems in sensor networks are solved in this paper.Identification and estimation technology of nonlinear systems in sensor networks with switching topologies are proposed.The main research contents and innovations of this paper are summarized as follows:1.Considering a general discrete-time nonlinear system with process noise,using the nonlinear model is equivalent to the data-driven state-space model utilizing Full Form Dynamic Linearization.A novel data-driven adaptive optimization recursive method is proposed to solve the identification problem of fast time-varying parameters.By using the optimization index and Lyapunov convergence,an optimal and convergence estimation method for fast time-varying parameters is obtained including parameter update rate and error covariance update rate.The method makes full use of historical data to improve the tracking ability of time-varying parameters.The damping factor and forgetting factor are used to improve the utilization rate of new data,so as to obtain better estimation effect.Finally,the Lyapunov convergence analysis of the method is given.2.Aiming at consensus optimal state estimation for discrete-time nonlinear systems,a data-driven distributed information-weighted consensus filtering is proposed to achieve consensus optimal state estimation for nonlinear systems in sensor networks with switching topologies.According to the topologies of a sensor network,a novel collectively distributed direct and indirect measurement model is designed.The direct and indirect learning gains,error covariance update rate and state estimation update rate in global form are derived and obtained.This method effectively solves the problems such as limited observation of sensor nodes,insufficient utilization of network topology information and communication noise in sensor network.Then,using Lyapunov stability theory,it is proved that the method can strictly guarantee the local optimal state estimation by distributed sensors.3.The method of data driven identification and sensor network state estimation proposed in this paper are further analyzed and verified by MATLB simulation platform.Simulation examples and comparative experiments are used to verify that the proposed recursive method can effectively solve the identification problem of fast time-varying parameters.The addition of historical data is also verified,which has an effect on improving the performance of the method.Furthermore,the numerical simulation of complex sensor networks with switching topologies is given,and the effectiveness of the proposed filtering method is verified by comparing with other existing distributed filtering for nonlinear systems.Finally,the simulation research and comparison experiment of high-speed train speed estimation are given to illustrate the superiority of the method.