Efficient P-wave detection in real time for Earthquake Early Warning System based on Artificial Intelligence

Abstract:

Real-time earthquake detection is the highest priority for Earthquake Early Warning System (EEWS) effectiveness, in order to benefit society and mitigation efforts. Unfortunately, there is no anticipated earthquake prediction, however, seismic observations in real time, at strategic points close to the seismic source, open the possibility of estimating parameters that allow generating an early alarm before the earthquake waves shake the population. In this work, Artificial Intelligence (AI) algorithms for optimal detection seismic primary compressional waves (P-wave) are investigate in order to improve the response of EEWS. The Stanford Earthquake Dataset (STEAD) was used in order to train robust Machine Learning models that allows us to choose the suitable algorithm to detect an earthquake and accurately pick the P-phase time. The Support Vector Machine (SVM) provided the best results in terms of computational time response and the success rate. This procedure has been tested on earthquakes data (magnitude upper 5, period 2018 - 2019) recorded by Instituto Geofisico del Peru (IGP), obtaining a success of 100% in detection and a 0.1 second of Root Mean Square Error (RMSE) for the picked time of the seismic P-wave, compared with the IGP catalog. Finally, the performance of this algorithm compared with traditional Short Time Average through Long Time Average trigger (STA/LTA) algorithm is briefly discussed.