Real-Time P-Wave Onset Automatic Recognition Using XGBoost Model

Accurate identification of P-wave onsets is crucial for improving the reliability of seismic emergency response in major engineering projects. However, conventional P-wave onset detection methods often suffer from weak anti-interference capability and delayed triggers. This study proposes an XGBoost-based model for real-time P-wave onset detection, utilizing four parameters that describe period and energy features. The model was trained using 9978 vertical seismic records from 197 large earthquakes (1996-2017) from Japan's K-NET network and tested on 4262 records from 278 events (K-NET and KiK-net). Comparison with the widely used EDP-Picker method showed that 90.2 % of samples had a picking error within <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\pm 0.5$</tex> s using the XGBoost model, compared to 86.7 % with EDP-Picker. The XGBoost model also exhibited stronger anti-interference capability, with fewer false positives and delayed triggers.