A Machine Learning Based to the Enhancement of Electrical Fault Detection and Classification

Abstract

The electricity transport system relies on transmission lines, which remain vulnerable to faults that can cut off operations and cost the system massive monetary losses. Transmission lines are very crucial over long distances in delivering electricity, but they also have reliability problems. Images of defects that might disrupt the power supply and put people in risk accompany reliability.  Therefore, this study uses machine learning on the Electrical Fault Detection and Classification dataset from Kaggle, which comprises voltage and current observations from an 11 kV transmission system. Two models, K-Nearest Neighbors (KNN) and Long Short-Term Memory (LSTM), were constructed following data preparation, which included label encoding and Minmax normalization. The models' performance was then evaluated using metrics such as confusion matrices, accuracy (Acc), precision (Prec), recall (Rec), and F1-score. KNN offered the highest Acc of 99.72, high Prec of 99.99, high Rec of 99.55, and a high F1-score of 99.77, and hence it beats LSTM and the earlier Random Forest and Decision Tree tricks. In addition, the LSTM model demonstrated a high performance also since the training and validation loss were both stable and convergent, which indicates that learning was effective. The above findings indicate that machine learning, including multi-feature fusion, is significant to enhance the accuracy of electrical fault detection; thus, a very precise and reliable solution can be offered to minimize the loss of time and improve the safety of power transmission networks.