Sensor networks are regarded as a promising technology in the field of information perception and processing owing to the ease of deployment, cost-effectiveness, flexibility, as well as reliability. The information exchange among sensors inevitably suffers from various network-induced phenomena caused by the limited resource utilization and complex application scenarios, and thus is required to be governed by suitable resource-saving communication mechanisms. It is also noteworthy that noises in system dynamics and sensor measurements are ubiquitous and in general unknown but can be bounded, rather than follow specific Gaussian distributions as assumed in Kalman-type filtering. Particular attention of this paper is paid to a survey of recent advances in distributed filtering of networked dynamic systems with non-Gaussian noises over sensor networks. First, two types of widely employed structures of distributed filters are reviewed, the corresponding analysis is systematically addressed, and some interesting results are provided. The inherent purpose of adding consensus terms into the distributed filters is profoundly disclosed. Then, some representative models characterizing various network-induced phenomena are reviewed and their corresponding analytical strategies are exhibited in detail. Furthermore, recent results on distributed filtering with non-Gaussian noises are sorted out in accordance with different network-induced phenomena and system models. Another emphasis is laid on recent developments of distributed filtering with various communication scheduling, which are summarized based on the inherent characteristics of their dynamic behavior associated with mathematical models. Finally, the state-of-the-art of distributed filtering and challenging issues, ranging from scalability, security to applications, are raised to guide possible future research.
This paper addresses the distributed resilient filtering for discrete-time large-scale systems (LSSs) with energy constraints, where their information are collected by sensor networks with a same topology structure. As a typical model of information physics systems, LSSs have an inherent merit of modeling wide area power systems, automation processes and so forth. In this paper, two kinds of channels are employed to implement the information transmission in order to extend the service time of sensor nodes powered by energy-limited batteries. Specifically, the one has the merit of high reliability by sacrificing energy cost and the other reduces the energy cost but could result in packet loss. Furthermore, a communication scheduling matrix is introduced to govern the information transmission in these two kind of channels. In this scenario, a novel distributed filter is designed by fusing the compensated neighboring estimation. Then, two matrix-valued functions are derived to obtain the bounds of the covariance matrices of one-step prediction errors and the filtering errors. In what follows, the desired gain matrices are analytically designed to minimize the provided bounds with the help of the gradient-based approach and the mathematical induction. Furthermore, the effect on filtering performance from packet loss is profoundly discussed and it is claimed that the filtering performance becomes better when the probability of packet loss decreases. Finally, a simulation example on wide area power systems is exploited to check the usefulness of the designed distributed filter.