Online Distributed Identification for Nonlinear Networks With Homogeneous Clusters

The identification of heterogeneous nonlinear networks consisting of homogeneous clusters is investigated, which is challenging due to the high computational complexity and partial state observations. To improve the computational efficiency, a finite-time horizon particle-based online Expectation-Maximization (EM) algorithm is proposed that enables distributed identification of unknown parameters across all agents even under complex agent couplings. To overcome the limitations caused by partial state observations, a neighbor-centered adapt-then-combine (ATC) strategy is developed for homogeneous clusters. This adaptive mechanism dynamically diffuses parameter estimates among neighboring agents, improving both accuracy and scalability for identifying large-scale networks. Theoretical analysis establishes the convergence of the proposed algorithm. Simulation examples validate the effectiveness and reliability of the proposed method, demonstrating its capability for a wide range of applications related to large-scale networks.