Hybrid modified evolutionary particle swarm optimisation-time varying acceleration coefficient-artificial neural network for power transformer fault diagnosis. (August 2016)

• Record Type: Journal Article
• Authors: Illias, Hazlee Azil; Chai, Xin Rui; Abu Bakar, Ab Halim
• Is Part Of: Measurement. Volume 90(2016)
• Page Start: 94
• Page End: 102
• Publication Date: 2016-08
• Subjects: Modified particle swarm optimisation -- Artificial neural network -- Power transformer -- Artificial intelligence; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2016.04.052
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: Existing DGA methods sometimes yield incorrect diagnosis results. The accuracy of the existing methods is believed to have rooms for improvement. Hybrid modified EPSO-TVAC-ANN was successfully proposed. The performance of the ANN was optimised through the proposed MEPSO-TVAC. The proposed method obtained the highest accuracy than the previous methods. Abstract: In power transformer fault diagnosis, dissolved gas analysis (DGA) has been widely used to identify the type of the fault. The common methods of DGA are IEC 60599 method, Doenenberg's ratio method and Roger's ratio method. The accuracy of the DGA diagnosis will determine the cost, duration and workload of the maintenance since it can influence the error in the maintenance. Although DGA methods have been used widely, sometimes they still yield incorrect diagnosis results. Thus, many works on transformer fault diagnosis have been proposed previously, which include artificial intelligence methods, to improve the accuracy of transformer fault diagnosis. However, the accuracy of the previously reported works is believed to have rooms for improvement. Therefore, in this work, hybrid modified evolutionary particle swarm optimisation-time varying acceleration coefficient (MEPSO-TVAC)-artificial neural network (ANN) was proposed for transformer fault diagnosis based on dissolved gas data. This is due to these two methods have never been proposed for transformer fault diagnosis in the past. The performance of the ANN was optimised through the proposed MEPSO-TVAC. The superiority of the proposed method was demonstrated through comparison with the existing DGA methods, unoptimised ANN and previously reported methods in literatures. The comparison shows that the proposed hybrid MEPSO-TVAC-ANN obtained the highest accuracy among all methods, which can then be used for power transformer fault diagnosis.