Distributed Tracking via Simultaneous Perturbation Stochastic Approximation-based Consensus Algorithm

Abstract

Networked systems comprised of multiple nodes with sensing, processing, and communication capabilities are able to provide more accurate estimates of some state of a dynamic process through communication between the network nodes. This paper considers the distributed estimation or tracking problem and focuses on a class of consensus normalized algorithms. A distributed algorithm consisting of two well-studied parts, namely, Simultaneous Perturbation Stochastic Approximation (SPSA) and the consensus approach is proposed for networked systems with uncertainties. Such combination allows us to relax the assumption regarding the strong convexity of the minimized mean-risk functional, which may not be fulfilled in the distributed optimization problems. For the proposed algorithm we get a mean squared upper bound of residual between estimates and unknown states. The theoretically established properties of proposed algorithm are illustrated by simulation results.