Distributed Kalman filtering.

摘要

In recent years, a compelling need has arisen to understand the effects of distributed information structures on estimation and filtering. In this thesis, distributed Kalman filtering has been on focus with various perspectives. Firstly, a bibliographical review on distributed Kalman filtering (DKF) is provided. A classification of different approaches and methods involved to DKF has been elaborated, followed by the applications of DKF are also discussed and explained separately. A comparison of different approaches is briefly carried out. Focuses on the contemporary research are also addressed with emphasis on the practical application of the techniques. An exhaustive list of publications, linked directly or indirectly to DKF in the open literature, is compiled to provide an overall picture of different developing aspects of this area.;Secondly, an approximate distributed estimation within distributed networked control formalism has been proposed. This is made possible by using Bayesian-based forward-backward (FB) system with generalized versions of Kalman filter. The analytical treatment is presented for cases with complete, incomplete or no prior information with bounds and then followed by estimation fusion for all three cases. The proposed scheme is validated on a rotational drive-based electro-hydraulic system and the ensuing results ensured the effectiveness of the scheme underpinning it.;The thesis proposes distributed expectation maximization (EM)-based reduced-order singular evolutive extended Kalman (SEEK) smoother. Optimal reduced-order smoothers complement the computation by doing re-analysis to correct the state of a dynamic system. The nature of order reduction of the SEEK smoother is fulfilling this phase, and made more precise by injecting the Kalman-like particle nature of the filter. The proposed scheme is first evaluated with its distributed full-order EM-based smoother version, followed by its reduced order version. The EM algorithm plays its role to identify and improve the estimate of process noise covariance Q in each case. The proposed scheme is then validated on a power quality system with various kinds of loads, ensuring the effectiveness and applicability of the scheme underpinning it.;An approach for distributed estimation algorithm is proposed using information matrix filter on a distributed tracking system in which N number of sensors are tracking the same target. The approach incorporates proposed engineered versions of information matrix filter derived from covariance intersection, weighted covariance and Kalman-like particle filter (KLPF) respectively. The steady performance of these filters is evaluated with different feedback strategies, moreover employing them with commonly used measurement fusion methods i.e. measurement fusion and state-vector fusion respectively to complete the picture. The proposed filters are then validated on an industrial utility boiler, ensuring the effectiveness and applicability of the scheme underpinning it.;Keywords: DKF, Bayesian approach, prior information, distributed estimation, approximate estimation, electro-hydraulic system, expectation maximization, power system quality, EM smoother, information matrix filter, covariance intersection, weighted covariance, KLPF, industrial utility boiler.