Publication: Pengesanan kerosakan bahan penebat transformer dengan menggunakan rangkaian neural buatan
| datacite.subject.fos | oecd::Engineering and technology::Electrical engineering, Electronic engineering, Information engineering | |
| dc.contributor.author | Che Osman, Suzita | |
| dc.date.accessioned | 2024-06-14T09:39:31Z | |
| dc.date.available | 2024-06-14T09:39:31Z | |
| dc.date.issued | 2006-05-01 | |
| dc.description.abstract | This paper is about a study of artificial intelligence (AI) applications for determination of transformer faults. The AI techniques include artificial neural networks (ANN), expert system, fuzzy systems and multivariate regression. The faults detection is based on dissolved gas-in-oil analysis (DGA). Several gases are formed during transformer faults. These are H2, O2, N2, CO, CO2, CH4, C2H6, C2H4 and C2H2. The causes of fault gases can be divided into three categories; corona or partial discharge, pyrolysis or thermal heating and arcing. A literature review showed that conventional fault diagnosis method, i.e. the ratio methods (Rogers, Dornenburg and IEC) and the key gas method. Various AI techniques may help to solve the problems and provide a better solution. A multilayer perceptron (MLP) is the choice among several neural network architectures that is used in this study. A three layer neural network has been used throughout this study. The data consists of 41 gas samples are divided into two sets: a set of 21 samples for training phase and the remaining 20 samples are used to test for the validity and applicability of the ANN approach. Neural network can define the transformer fault through the learning process. Matlab7 is used to design the multilayer perceptron (MLP). Three types of learning algorithm are used in this project to train the MLP network, which are resilient backpropagation, Bayesian regularization and Levenberg-Marquardt. From the result, resilient backpropagation algorithm gives the highest percentage of accuracy (91.75%) compared to Levenberg-Marquardt (89%) and Bayesian regularization (85.75%). The results proved that the MLP network has high capability to detect transformer insulation fault. | |
| dc.identifier.uri | https://erepo.usm.my/handle/123456789/19467 | |
| dc.language.iso | en | |
| dc.title | Pengesanan kerosakan bahan penebat transformer dengan menggunakan rangkaian neural buatan | |
| dc.type | Resource Types::text::report | |
| dspace.entity.type | Publication | |
| oairecerif.author.affiliation | Universiti Sains Malaysia |