Speaker: 甘叠

Title: Distributed extended SG algorithm for stochastic systems with colored noise

Language: Chinese

Time & Venue: 2024.05.14 13:00-13:40 思源楼415

Abstract: In this talk, we study the distributed identification problem of dynamic systems described by stochastic regression models with colored noise. To deal with challenges brought by the colored noise, we provide estimates for the unknown noise over a previous time period, and incorporate them with observed signals to construct extended regression vectors. Based on this, we develop a distributed extended stochastic gradient (SG) algorithm to estimate unknown parameter matrices by integrating the diffusion strategy of extended regression vectors with the consensus strategy of neighbors’ estimates. We introduce cooperative non-persistent excitation conditions to reflect the temporal and spatial union information condition, under which the almost sure convergence of the proposed distributed algorithm is established. Our results are obtained without assuming the independence, stationarity or persistent excitation (PE) conditions for the regressors and the Gaussian property for the system noises. Finally, numerical results are provided to verify the effectiveness of our proposed algorithm.