一种改进的分布式动态分簇粒子滤波算法

Improved distributed dynamic clustering particle filter algorithm

针对Kalman和UKF算法不适合WSN非高斯环境中目标追踪,而粒子滤波适合却能量消耗大的问题,提出了一种改进的分布式动态分簇粒子滤波(DDCPF)算法,动态建立分布式跟踪簇。簇成员节点结合各自最新的最优估计值和方差产生粒子,根据类似Kalman算法将粒子滤波的预测值与测量值结合得到的值作为预测值,提高跟踪精度。最后搭建目标跟踪的WSN仿真环境,仿真结果表明,此算法具有较高的跟踪精度,且显著降低了系统的能耗和跟踪时间。

Kalman filter and UKF are the most typical filter algorithms in target tracking domain in WSN. But their filtering performance will descend or even diverge when non-Gaussian distribution occurs. The particle filter completed by Gordon can resolve this problem. In order to improve the tracking performance and to reduce the energy cost,this paper proposed an improved distributed dynamic clustering particle filter（ DDCPF） algorithm. It organized dynamic clusters,and constructed the decentralized tracking structure. It used the optimal estimate positions of the cluster member nodes to incorporate the newest observation into the proposal distribution of the DDCPF. Meanwhile using a method similar to Kalman filtering,it integrated predicting value obtained by particle filter and measured values as the real predictive values to improve tracking accuracy. It established a performance evaluation system. The results indicate that DDCPF has similar good performance in the tracking accuracy,can reduce the communication cost and tracking response time significantly.