Sparse Network Identification for Consensus Discrete-Time Systems

Abstract

This paper investigates an identification problem of sparse topology among nodes for consensus networked control systems. We suppose the system is modeled by a discrete-time first-order consensus network stochastically disturbed by white gaussian process noise. Our proposed method employs a sparsity-penalized maximum likelihood approach, where edge weights are penalized by the ℓ1 norm to address the sparsity of network topology. The optimization problem is shown to be expressed as a DC (Difference of Convex functions) optimization problem. A numerical example illustrates that the proposed method is useful to locate edges and to estimate their weights.