Performance Assessment of Customized LSTM based Deep Learning Model for Predictive Maintenance of Transformer

Abstract

To improve predictive maintenance of transformers with small DGA datasets, customized LSTM network named C-LSTM is devised to circumvent the boundaries of the standard-LSTM network, which had an increased rate of classification error than conventional machine learning techniques. The study compares the performance of traditional machine learning algorithms with the customized LSTM model using various metrics such as validation accuracy, test accuracy, precision, recall, and F1-score. Additionally, the comparison includes the evaluation of seven fault detecting diagnostic techniques, including discharges of low/high energy, thermal/electrical faults, partial discharge, and low/medium/high thermal faults. The results indicate that the customized LSTM model outperforms traditional machine learning methods with a validation accuracy of 100% and a test accuracy of 98.57%.