甘叠,中关村实验室

Parameter estimation or filtering is one of the important issues in diverse fields including statistical learning, signal processing, system identification and adaptive control. With the development of sensor networks, we can collect more and more data, which provides favorable conditions for identification or estimation of unknown parameters of systems. Compared with the centralized algorithms, the distributed ones have been widely applied due to its advantages of scalability, robustness to the node or link failures as well as reducing communication load and calculation pressure. As far as we know, most of the theoretical results on the distributed adaptive estimation and filtering algorithm are established by requiring the independency in time or space, stationarity or ergodicity conditions of the system signals. However, it is hard or even impossible to satisfy the aforementioned assumptions, because the data measured is often generated by feedback systems. This talk will discuss the distributed adaptive estimation and filtering algorithms based on sensor networks to identify or track the unknown parameters for several kinds of classical discrete time stochastic systems. The convergence and stability analysis of the corresponding distributed algorithm are provided without relying on the assumptions of the independency, stationarity and ergodicity of the regression signals. Moreover, we can reveal the cooperative effect of multiple sensors from a theoretical point of view.