Design, development, and testing of a comprehensive and automated classification system for single and multiple PD source identification based on the relationship between the variation of PRPD patterns and the sources of PD is proposed. The proposed system consists of feature extraction methods and classifier algorithms that are implemented for recognition of partial discharge patterns. For single PD source identification, twelve high performance, applicable feature extraction techniques on PRPD patterns are employed to extract features. In order to present a comprehensive classification system, 10 well-known algorithms for the classification of PD sources have then been used. To evaluate the performance of the classification system, three laboratory test setups are designed and built to simulate various types of PD activities. The first test setup is designed to model common sources of PD in air, oil, and SF6. Using this setup, the application of automated classification system on different sources of PD in different HV insulation media is investigated. The second and third test setups are designed to test the classification system on identification of different sources of PD in oil-immersed insulation and power transformer cellulose insulation under both electrical and thermal stresses, respectively. In many practical situations, the interest lies in the identification of multiple, simultaneously activated PD sources in insulation. Multi-source PDs sometimes results in partially overlapped patterns, which makes them hard to be identified by single source identification techniques. To further enhance the proposed classification system, a novel algorithm to identify Multi-source PDs is developed and appended to the system. To evaluate the performance of this algorithm, a number of multi-source PD models have been designed. The overall results show that the classification system is well able to identify the single and multi-source of partial discharges. More importantly, this identification system is able to assign a ``degree of membership" to each PRPD pattern, besides assigning a class label to it. This enables probabilistic interpretation of a new PRPD pattern that is being classified and results in safer decision making based on the risk associated with different sources of PD. The results of this research is beneficial for the design of a solid basis for an automated, continuous 24/7 monitoring of equipment, which facilitates PD source identification in early stages and safe operation of HV apparatus. / October 2016
| Identifer | oai:union.ndltd.org:MANITOBA/oai:mspace.lib.umanitoba.ca:1993/31655 |
| Date | 08 September 2016 |
| Creators | Janani, Hamed |
| Contributors | Kordi, Behzad (Electrical and Computer Engineering), Swatek, David (Electrical and Computer Engineering) Jafari Jozani, Mohammad (Statistic) Jayaram, Shesha (University of Waterloo) |
| Source Sets | University of Manitoba Canada |
| Detected Language | English |
Page generated in 0.0019 seconds