Three-phase power transformer fault diagnosis based on support vector machines and bees algorithm

Abdusalam, Othman, Anayi, Fatih ORCID: https://orcid.org/0000-0001-8408-7673 and Packianather, Michael ORCID: https://orcid.org/0000-0002-9436-8206 2023. Three-phase power transformer fault diagnosis based on support vector machines and bees algorithm. Presented at: 3rd International Conference in Power Engineering Applications (ICPEA), Putrajaya, Malaysia, 6-7 March 2023. Proceedings of 3rd International Conference in Power Engineering Applications. IEEE, pp. 145-150. 10.1109/ICPEA56918.2023.10093147

Full text not available from this repository.

Abstract

In this paper, a new method is presented for the classification of current signals faults in three-phase transformers. In this method, Support Vector Machines are used in two different ways. The study utilized two support vector machines, SVM1 and SVM2, for detecting faults and inrush currents in 3-phase transformers, as well as differentiating between internal faults (turn-to-turn and turn-to-ground) and external faults. To evaluate the performance of the proposed model, laboratory experiments were conducted on a transformer system with both internal and external faults, and the resulting current signals were utilized to develop the model. By training machine learning classifiers to detect faults by SVM, a process for optimal feature identification has been proposed. To extract statistical characteristics from unprocessed data, discrete wavelet transform was used. An optimized subset was then created using the Bees algorithm (BA), which minimized the amount of data needed and improved the model's accuracy. 5k-fold cross-validation was used to train these models. This model has been analysed based on accuracy. The study compares SVMs to ANN-based classifiers and finds that SVMs are more reliable and provide faster results.

Item Type:	Conference or Workshop Item (Paper)
Date Type:	Publication
Status:	Published
Schools:	Engineering
Publisher:	IEEE
ISBN:	9781665475020
Last Modified:	23 May 2023 09:00
URI:	https://orca.cardiff.ac.uk/id/eprint/159728

Actions (repository staff only)

Edit Item