The Power Transformer Failure Prediction with Dissolved Gas Analysis Method Using TDCG based Random Forest

Abstract

The transformer is an important component, and early detection of potential failures plays an important role in the reliable operation of the electric power system. This article describes a new approach to power transformer failure prediction based on dissolved gas analysis (DGA) by applying the TDCG method with  the Random Forest algorithm. DGA data from operational transformers is used to train and test predictive models. The random forest  method based on TDCG allows comprehensive analysis of changes in dissolved gases in transformer oil, thus enabling early detection of failure conditions. The experimental results show that  the prediction model uses a model created by applying hyperparameter tuning for optimal  parameter tuning to have high accuracy, accuracy is obtained up to 96% in detecting potential failures, the standard used for accuracy presentation uses confusion matrix as the accuracy of the prediction model. In addition, it can optimize time efficiency in analyzing failures and prevent human error when calculating gas fault  identification or potential failures.