Some Studies on Distributed Filtering for Networked Systems

Abstract

Recently, networked systems have received increasing research attention due to their numerous applications in process monitoring, intelligent transportation, power grids, and other fields. Compared with point-to-point communication in traditional systems, different system components in networked systems perform data transmission through shared communication networks. Networked systems have the advantages of low cost, high flexibility, and convenient installation. For networked systems, distributed filtering/estimation that merely utilizes local and neighbor information to complete the estimation tasks is a crucial issue.

Communication among system components plays a key role since only partial information is available in distributed filtering for networked systems. However, data transmission is subject to restrictions due to limited network and device resources. Several schemes, namely quantized communication methods, event-triggered communication protocols, and communication mechanisms using energy harvesting techniques, can be used to handle resource issues. Furthermore, some undesirable phenomena (e.g., faults and non-Gaussian noises) often occur in practical engineering systems. Thus, the distributed filtering designs under the imperfect conditions are worth discussing. In summary, it is highly desirable to develop a distributed filtering framework for networked systems to solve the resource limitation problem. The criteria for guaranteeing the filtering performance will also be derived.

The following topics will be presented in detail in this thesis: (a) an information-based distributed filtering algorithm for networked systems under dynamic quantization; (b) two edge event-triggered distributed filtering schemes for networked systems; (c) distributed estimation for energy-harvesting-constrained networked systems; (d) distributed maximum correntropy filtering under energy harvesting constraints over networked systems.

The main contributions of this thesis are listed as follows:

A distributed filtering algorithm for networked systems with dynamic quantization is proposed. The system state is dynamically quantized before being transmitted by the communication channels. Such a quantization setting allows the system state to be processed adaptively, thereby solving the bottleneck problem in static quantization. A practical online adjustment rule of the quantizer's parameter is presented to ensure that the encoder and decoder can access the changing parameter. An information-based fusion strategy is employed to improve the filtering performance. The stochastic stability analysis of the proposed algorithm is provided.

The distributed filtering problem for networked systems under edge event-triggered communication is investigated. Two dynamic edge-based event-triggered schemes are designed, i.e., the sender-receiver and sender event-triggered schemes. Both are applied to the distributed filtering framework to reflect a general fully distributed filter design for the dynamic edge event-triggered mechanism. In the sender-receiver scheme, the receiver node is given the privilege to receive or reject the incoming state information, which reduces redundant transmissions from the sender node. A sufficient criterion pledges the exponential boundedness of the estimation errors.

Two distributed estimators are developed for energy-harvesting-constrained networked systems. One is for simultaneously estimating the state and the fault over sensor networks limited to energy harvesting. The other is for the state estimation in time-varying complex networks under energy harvesting constraints. Different types of harvesting models are characterized for various energy resources. Transmission probabilities can be obtained recursively to deal with intermittent communication caused by energy harvesting. For each estimator, the estimation error covariance has an upper bound, which is minimized by designing the energy-based filter parameters.

The distributed maximum correntropy filtering issue is studied over networked systems subject to energy harvesting. By exploiting the properties of correntropy, high-order statistical features of the error signal are captured to improve the estimation performance in non-Gaussian noises. Under the coupled influence of nonlinear/non-Gaussian systems and energy harvesting constraints, a new distributed filter is devised. The convergence of the fixed-point iteration is guaranteed to derive the optimal gain.

Date of Award	3 Aug 2023
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Wing Cheong Daniel HO (Supervisor) & Xiaosheng ZHUANG (Co-supervisor)