Use this URL to cite or link to this record in EThOS:
Title: Automatic pattern recognition and knowledge discovery for partial discharge on-line condition monitoring of cable systems
Author: Peng, Xiaosheng
ISNI:       0000 0004 2718 1999
Awarding Body: Glasgow Caledonian University
Current Institution: Glasgow Caledonian University
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
Partial Discharge (PD) based on-line monitoring has increasingly been adopted for Condition Based Maintenance (CBM) of underground power cable utilities systems. However, two significant challenges still greatly affect the diagnostics result. The first significant challenge is the lack of effective method for automatic PRP recognition of on-line monitoring data: The second significant challenge is previous knowledge rules for insulation diagnostics and PD identification are not effective when applied to on-line monitoring system. The objectives of the current research are to develop an effective automatic phase resolved PD pattern recognition algorithm and an effective knowledge discovery algorithm for on-line PD cable systems monitoring. In order to enable automatic PD pattern recognition, PD signal processing and feature extraction are carried out. An effective data denoising algorithm-Adaptive Second Generation Wavelet Transform based denoising technology (ASGWT}-is developed. A series of PD feature extraction techniques, raw PD signal analysis, phase-resolved PD pattern, are applied and compared. Based on data denoising and feature extraction, a data base management system is developed, which is used for managing raw data, PD feature, cable information and insulation diagnosis results for further automatic PD pattern recognition and knowledge discovery. For automatic PD pattern recognition, a K-Means method based PD pattern recognition algorithm is developed. This algorithm contains coordinate transformation, data overlapping, data discretization, centroid identification, and pattern judgement. On-site data testing and on-line data monitoring is tested by the algorithm, proving it effective for automatic PD pattern identification for different types of discharge. Furthermore, when PD from three-phase data is processed as the input signal, the algorithm successfully identifies this kind of discharge. The recognition rate is 80% when the algorithm is tested by 85 sets of on-line monitoring data, if signal overlapping is not adapted. The recognition rate is 100% if signal overlapping is adapted. For knowledge discovery, Rough Set (RS) based knowledge acquisition algorithm is developed. Signal discretization, attribution reduction, and decision table generation are discussed and proved to form an effective knowledge discovery flow chart of Rough Set theory. One simulation example containing training and application process is presented. The results shows that 59 certain rules can be extracted from 400 sets of data, and a rate of 81.5% of another 600 sets of data can be identified by the rules obtained. Three experimental studies, including on-site PD tests in substations and PD testing in HV lab, are carried out to validate the proposed approaches. The most effective PD features describing PD pulses and PD sources and corresponding criteria are extracted by RS theory.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available