Integrative fault diagnostic analytics in transformer windings: Leveraging logistic regression, discrete wavelet transform, and neural networks.

Abstract

Protection of transformers is crucial in the power industry due to their susceptibility to various electrical and mechanical faults over time. Traditional methods like Frequency Response Analysis (FRA) have limitations in accurately diagnosing these faults. This paper highlights the potential of combining advanced signal processing techniques with machine learning algorithms by presenting an innovative hybrid model for accurately detecting transformer winding faults, utilizing Logistic Regression, Artificial Neural Networks (ANN) and Discrete Wavelet Transform (DWT). The primary novelty of this approach lies in the use of Logistic Regression to evaluate the impact of each wavelet decomposition, which aids in selecting the most effective wavelet bases, reducing data volume, and decreasing computational complexity. By integrating these methods, the proposed model significantly enhances fault detection accuracy and system performance. The effectiveness of the algorithm is validated through a practical case study, demonstrating a 97 % success rate in detecting transformer faults and reducing misclassification to 2.9 %.