Accuracy Improvement of Data-Driven Algorithms in Power Transformer Assessment Using Hyperparameter Optimization on DGA Data

Abstract

Transformers are critical components in power systems. Faults within these devices can lead to substantial repair costs and prolonged service interruptions. Dissolved Gas Analysis (DGA) of transformer oil is widely used for monitoring transformer health. This research leverages data-driven algorithms, employing the Duval-Pentagon (DP) method and hyperparameter optimization, to enhance fault diagnosis accuracy in power transformers. After preprocessing the DGA dataset, it was split into training and testing sets in an 80:20 ratio. Subsequently, several data-driven algorithms, including Support Vector Machines Algorithm (SVMA), Decision Trees Algorithm (DTA), Logistic Regression Algorithm (LRA), and Naive Bayes Algorithm (NBA), were employed on the dataset. To further improve fault diagnosis accuracy, a hyperparameter optimization technique was implemented by leveraging random search. Evaluation metrics such as accuracy, F1-measure, recall, precision, and Matthews Correlation Coefficient (MCC) were used to assess impact of hyperparameter optimization. The findings demonstrate that hyperparameter optimization consistently enhances the performance of data-driven algorithms. Among the algorithms proposed in this research, DTA with hyperparameter optimization achieved the highest accuracy with an accuracy rate of 93.37% in transformer fault diagnosis. The algorithms were implemented based on Python.