Data-Driven Distributed Robust Consensus Kalman Filtering for Sensor Networks

Abstract

This paper mainly studies the distributed robust filtering problem for wireless sensor networks without knowing the variance of the driving noise and measuring noise. A distributed filtering algorithm based on leader-follower consensus strategy is developed. Firstly, from limited sampling data of noises, the empirical distribution of the noises which play a critical role in filtering is given. Secondly, by formulating the problem as a zero-sum game problem and using dynamic programming approach, an optimal distributed robust filtering policy with Wasserstein penalty for the leader is obtained. Then, by using an adaptively weighting average consensus protocol, the filtering algorithm for followers is given. The effectiveness of the proposed filter is demonstrated through simulations.