EMD–Conv1D–LSTM for P-Wave Picking in High-Speed Rail Earthquake Early Warning

Abstract High-speed railways (HSR) are characterized by intense ambient noise, complex vibration interference, and an extremely short early-warning time window, which jointly impose stringent requirements on seismic phase-picking algorithms. To meet the engineering demand of high precision with low false-alarm and missed-detection rates in high-speed railway earthquake early warning (HSR-EEW), this study proposes an automatic P-wave arrival-time picking method based on an empirical mode decomposition–one-dimensional convolution–long short-term memory network (ECL). The overall performance is evaluated in terms of accuracy, error-distribution characteristics, and parameter sensitivity. The results indicate that the proposed ECL algorithm achieves a P-wave picking accuracy of 98.6% with a mean absolute error (MAE) of 0.097 s, outperforming empirical mode decomposition–long short-term memory (EL), empirical mode decomposition–one-dimensional convolution (EC), short-term average/long-term average (STA/LTA), and Akaike information criterion (AIC) algorithms. The accuracy of traditional algorithms is mainly controlled by amplitude features, whereas AI-based algorithms exhibit better robustness to conventional source parameters. Combining the first-station accuracy analysis, the ECL algorithm achieves high picking accuracy with low error dispersion and strong reliability, providing reliable technical support for automatic P-wave picking in HSR-EEW systems. Combined with the first-station accuracy analysis and field-measured validation, the ECL method demonstrates high P-wave picking accuracy, low error dispersion, and strong robustness to operational interference, providing reliable technical support for automatic P-wave picking in high-speed railway earthquake early warning (HSR-EEW) systems.