讲座题目：Consistent Distributed Kalman Filter Over Sensor Networks
主 讲 人：方海涛 中国科学院数学与系统科学研究院
Dr. Hai-Tao Fang received his B.S. degree in probability and statistics in 1990, M.S. degree in applied mathematics in 1993 and Ph.D. degree in 1996 respectively from the Peking University, Tsinghua University and Peking University. He now is with the Laboratory of Systems and Control, Academy of Mathematics and System Sciences, Chinese Academy of Sciences as a Professor.
From 1996-1998 he was a Postdoc at the Institute of Systems Science and joined the Institute as an Assistant Professor in 1998. His current research interests include optimization，identification and control in stochastic systems and their applications in signal processes and communication.
In this talk, we investigate the network state estimation problem for a class of discrete time-varying systems. To effectively estimate the system state over the sensor network, the consistent distributed Kalman filter inspired by covariance intersection fusion (CI) is proposed, and the consistency of this filter at real time is proved. We also give a way to obtain some adaptive CI weights and prove that these weights have lower error variance bound than the constant CI weights. Based on a global observability condition and the strong connectivity of the directed network, the stability of the proposed filter is analyzed and guaranteed for time-varying systems. Then the distributed state estimation problem over a sensor network is studied with event-based communication scheme. A novel scalable event-based distributed Kalman filtering algorithm is proposed to reduce the burden of communications between sensors. Instead of event triggering condition relying on state estimation, a triggering mechanism depending on a lower bound of the true information matrix is developed. Also, the stability conclusion of the distributed filter is provided under certain conditions of collective observability and suitable connectivity. Some numerical simulations illustrate the feasibility as well as the effectiveness of the proposed algorithms.