Comparison between Principal Component Analysis and Wavelet Transform ‘Filtering Methods for Lightning Stroke Classification on Transmission Lines

• The analyses and study of different signal processing techniques (SPT) applied to TL lightning stroke classification are presented.
• The most important characteristic in a lightning stroke classification problem is related to the extraction of features.
• Daubechies mother wavelet family has been the most applied function in analyzing EPS transients.
• The eigenvectors are used as mother functions.
• PCA could be used to overcome the shortcomings of Wavelet Transform in mother function selection.

This paper presents an assessment between Principal Component Analysis (PCA) and Wavelet Transform (WT) signal processing techniques applied for Transmission Lines (TLs) lightning stroke classification. In this work, the atmospherics discharges signals are analyzed in two steps. The first step objective is patterns extraction, which is developed through Principal Component Analysis and the Wavelet Transform. The second step objective is pattern classification, which is developed using three different techniques: Artificial Neural Network (ANN), k-Nearest Neighbors (k-NN) and Support Vector Machine (SVM).This work presents as assessment of lightning stroke classification, providing useful information, especially in extraction and selection of mother functions and the use of PCA. Both methodologies are assessed under different lightning stroke conditions. Features as extraction, speed, orthogonal functions and others are comparatively assess.Resu lts show that by using PCA, optimal mother functions can be extracted, presenting a new alternative for relaying protection.