ADMM-Based Distributed Recursive Identification of Wiener Nonlinear Systems Using WSNs

Abstract

The distributed estimation over wireless sensor networks (WSNs), as opposed to least-squares and fusion-center based estimations, is proficient to work with real-time applications. In this paper, a block-structured Wiener model is identified in a distributed fashion by minimizing the least-squares cost function on prediction error. As the block-structured Wiener model can approximate a large class of nonlinear systems with a small number of characteristics parameters hence makes it more suitable to work with. The global minimization task is reformed into several constrained subtasks in a manner that each node in WSN can obtain the parameters of interest locally. Each node in the network has the ability to combine its local estimates with the single-hop neighbors’ estimates to obtain the global parameters of interest. The optimization of the reformulated cost is accomplished using a powerful distributed method called alternating direction method of multipliers. Simulations are carried on an infinite-order nonlinear system under the impact of observation noise. The obtained results are juxtaposed to the results of non-cooperative algorithm to show the effectiveness and superiority of the proposed algorithm.