Research On Distributed Kalman Algorithm In Sensor Networks With Missing Data

The sensor network is deployed in a large number of micro sensor nodes distributed spatially independent in the monitoring area,which are used to monitor,sense and collect environmental or target states in real-time,such as temperature,humidity,vibration,pressure or movement.Because of the limited energy,sensing range,communication and computing capacity of nodes,distributed algorithms exhibit superior performance over centralized algorithms.Diffusion strategy has attracted extensive attention because of its simplicity,flexibility and stability among various distributed strategies.Most of previous distributed algorithms in sensor networks assume that the information sensed by sensor nodes are lossless.As the size of the deployed sensor network grows,raw data typically has significant data loss because data collection is heavily influenced by hardware and wireless conditions.In this paper,diffusion strategy of least mean square(LMS)algorithm and its derivation process are briefly introduced.However,LMS algorithm is usually used for estimating constant parameters instead of dynamic model such as target tracking which is also an important research area of sensor networks no matter military or civilian.In time-varying model,Kalman filter algorithm is one of the most popular recursion algorithm since it was proposed in 1960 s.In this paper,a diffusion strategy of distributed Kalman filter(DKF)algorithm is explored on the basis of previous,whose technical challenge is how to migrate mature central(or traditional)Kalman filtering methods to complex large dynamic systems and distribute the measurements across a large geographic area.Another advantage of Kalman filter algorithm is that estimating the states of dynamic system from an incomplete and noisy measurement.But Kalman filter algorithm cannot be directly implemented for improving estimations with missing data because of the unique data loss patterns of sensor network.According to the situation above,we provided a new imputation diffusion Kalman filter(IDKF)algorithm to track the object with missing data at random.We reconstruct the missing data by using last estimation result of Kalman filter replacement.Our simulation examples have demonstrated that the proposed algorithm can deal with the missing data problem and improve the accuracy of the sensor network.The modifications of detection and imputation can recognize the nodes who lose data when the noise is Gaussian white noise and perform better than usual strategy in the first step of DKF.Early distributed estimation algorithms for sensor networks are mostly single-task estimation problems,that is,the sensor nodes of the whole network jointly estimate a certain unknown parameter vector.In this paper,we extend the problem to multitask estimation,in which situation different nodes in the network estimate different state parameters,or nodes in different clusters are interested in different state parameters,while nodes in the same cluster still estimate the same parameter.Based on the correlation among tasks,combined with the above interpolation distributed Kalman filter algorithm,a multitask Kalman estimation algorithm is proposed in the case of missing data.The proposed algorithm can effectively adapt to multitask networks and improve the accuracy of network estimation with missing data randomly.