Speaker: 甘叠

Title: Distributed Extended Least Squares Algorithm Over Directed Graphs

Language: Chinese

Time & Venue: 2024.05.14 13:40-14:20 思源楼415

Abstract: We present a fully distributed estimation method to estimate an unknown parameter matrix for autoregressive-moving average with exogenous input models and over directed sensor networks, based on the extended least squares (ELS) algorithm and diffusion strategies. Unlike the centralized method, each sensor in the network is only allowed to communicate with its neighbors in the proposed diffusion-based ELS algorithm in order to make communication over the network possible and to increase the robustness, scalability, and privacy of the system. Without resorting to independence, stationarity, or persistent excitation (PE) conditions for the regressors and the Gaussian property for the system noises, the convergence of the proposed distributed ELS algorithm is established in this paper. The cooperative decaying excitation condition for the convergence results also shows that even if any individual sensor cannot estimate the unknown parameter matrix accurately by using the traditional non-cooperative ELS algorithm, the whole network can still fulfill the estimation task by using the distributed ELS algorithm, which is called the cooperative property of the distributed estimation algorithms. Furthermore, some simulation results are provided in the paper to demonstrate the efficiency and cooperative property of the proposed distributed ELS algorithms.