Dynamic Estimation Over Distributed Sensing Network With Communication Delays

The dynamic estimation for time-varying parameters through distributed sensing network is rooted in the real-time detection and control of cyber-physical systems. It is a challenging problem while the communication delays, general time-varying directed graphs, and uncertain environment are considered. To this end, in this article, we propose the distributed online optimization algorithms for solving the distributed dynamic constrained parameter estimation problem on time-varying directed graphs in the presence of communication delays. We first develop a subgradient dynamic dual averaging estimation algorithm (subgradient DDAEA) based on subgradient information. Through analysis we show that the subgradient DDAEA achieves the tightest dynamic regret bound known to date. Further, we extend the subgradient DDAEA to the scenario of stochastic gradient or gradient-free (i.e., the gradient information is polluted by noise or is unavailable) and propose the stochastic DDAEA and gradient-free DDAEA, respectively. We demonstrate that the stochastic DDAEA and gradient-free DDAEA do not entail regret degradation compared to the subgradient DDAEA. Finally, the effectiveness of the proposed DDAEA is verified by the simulation experiments.