| 英文摘要 |
A transformer is a device used for electricity-related purposes, one of which is to distribute electricity from a supplier, in this case the State Electricity Company (PLN), to customers or the community. Considering the transformer’s essential role, it is crucial to conduct research to minimize damage to this device, which can be caused by a variety of factors, where gas and electrical conditions of the transformer, among other things, can be used as indicators of damage. Therefore, this study focused on creating models using Artificial Neural Network (ANN) algorithms to assess transformer conditions based on data obtained from transformers. In this study, correlation analysis was used to determine six features that served as leading indicators in evaluating the condition of the transformer. The six features were dibenzyl disulfide (DBDS), interfacial voltage, hydrogen, methane, ethylene, and water content. In modelling and testing, 80% of the data was distributed for the training dataset and 20% was for the testing dataset, with a total of 470 data points. This study also applied the Synthetic Minority Oversampling Technique-Rechecked, Reused, and Edited (SMOTE-R2E) method, which is an improvement of the SMOTE method. SMOTE-R2E is proposed in this study to overcome the limitations of unbalanced transformer data. In this study, model training was carried out using three approaches, i.e., model training using a training dataset obtained without the SMOTE method, model training using a training dataset obtained with the SMOTE method, and model training using a training dataset obtained with the SMOTE-R2E method. Each training approach was performed 100 times. Based on model testing on the testing dataset, the best model was the model obtained by applying the SMOTE-R2E method, with average accuracy and F1-score obtained from 100 iterations of 83.04% and 81.96%, respectively. |
本站僅提供期刊文獻檢索。
