Transformer Fault Diagnosis Utilizing Feature Extraction and Ensemble Learning Model

Abstract

This paper proposes a novel method for diagnosing faults in oil-immersed transformers, leveraging feature extraction and an ensemble learning algorithm to enhance diagnostic accuracy. Initially, Dissolved Gas Analysis (DGA) data from transformers undergo a cleaning process to ensure data quality and reliability. Subsequently, an interactive ratio method is employed to augment features and project DGA data into a high-dimensional space. To refine the feature set, a combined Filter and Wrapper algorithm is utilized, effectively eliminating irrelevant and redundant features. The final step involves optimizing the Light Gradient Boosting Machine (LightGBM) model using IAOS algorithm for transformer fault classification; this model is an ensemble learning model. Experimental results demonstrate that the proposed feature extraction method enhances LightGBM model’s accuracy to 86.84%, representing a 6.58% improvement over the baseline model. Furthermore, optimization with IAOS algorithm increases the diagnostic accuracy of LightGBM model to 93.42%, an additional gain of 6.58%.