Distributed Kalman Filtering in Presence of Unknown Outer Network Actuations

Abstract

This letter presents a fully distributed approach for tracking state vector sequences over sensor networks in presence of unknown actuations. The problem arises in large-scale systems where modeling the full dynamics becomes impractical. In this letter, the network only considers a subsection of the overall system which it can detect while accounting other inputs as unknown actuations. First a centralized technique that can consolidate all the available observation information is introduced. Then, operations of this optimal centralized solution are decomposed in a manner to allow their implementation in a distributed fashion while allowing each agent to retain an estimate of both the state vector and unknown actuations. The filter is derived in both diffusion and consensus formulations. The diffusion formulation is intended as a cost-effective solution, while the consensus formulation trades implementation complexity for accuracy.