An Efficient Machine Learning Model for Lightning Localization via Lightning-Induced Voltages on Transmission Lines

In this paper, three machine learning (ML)-based approaches—XGBoost, Artificial Neural Network (ANN), and Random Forest algorithms—are compared for the localization of lightning through the analysis of lightninginduced voltages on power transmission lines. Across all methods, two sensors are employed to capture lightninginduced voltages on power transmission lines. Numerical simulations demonstrate that the XGBoost algorithm exhibits higher efficiency in terms of location accuracy and computational time compared to the other algorithms. Additionally, the Principal Component Analysis (PCA) algorithm is applied to reduce the dimensionality of XGBoost by 50 times without compromising accuracy, thereby accelerating calculation time and reducing computational resource usage. The R2 score obtained from the model on test data, with a Signal-to-Noise Ratio (SNR) of 30 dB, exceeded 99%, and for data with an SNR of 10 dB, it reached approximately 98%. Various configurations of transmission lines and sensor locations were tested, revealing that the accuracy of the model is dependent upon the transmission line configuration and sensor positions.