Topology Inference for Consensus-based Cooperation under Time-invariant Latent Input

Topology inference is a critical problem in multiagent systems. Knowledge of the topology helps to better understand the system behavior, e.g., predicting states of a multivehicle system. Different from most prior works, this paper is dedicated to inferring the directed network topology from the time-series observations of the system, which is driven by the unmeasurable latent input. Independent of the system state, the latent input is designed by the system user or injected by external excitations. To decouple its influence from the observations, an input filtering-based topology inference algorithm (IF-TIA) is proposed. Specifically, we demonstrate that the influence of the time-invariant input on the system state evolution is separable, and transform the goal into solving an optimization problem based on the separability. The novel algorithm significantly improves the inference performance by reformulating the empirical risk, where the decreasing weight setting is adopted for the cost terms in the objective function. Compared with the benchmark methods, extensive simulations illustrate the outperformance of the IF-TIA method in terms of both the topology structure and the edge weight.