Distributed joint estimation and identification for sensor networks with unknown inputs

Abstract

In this paper we consider the problem of distributed, joint, state estimation and identification for a class of stochastic systems with unknown inputs (UI). A distributed Expectation-Maximization (EM) algorithm is presented to estimate the local state at each sensor node by using the local observations in the E-step, and three different consensus schemes are proposed to diffuse the local state estimate to each sensor's neighbours and to derive the global state estimate at each node. In the M-step, each sensor identifies the local unknown inputs by using the global state estimate. A numerical example of target tracking in distributed sensor network is given to verify the three different distributed EM algorithms compared with the centralized EM based measurement-level and track-level fusion.