Cumulative Absolute Velocity (CAV) parameter estimation in earthquake emergency response based on a support vector machine

Abstract

Rapid and accurate estimation of emergency response parameters during earthquakes is important in earthquake early warning (EEW) systems. Because earthquake rupture is not instantaneous, to accurately, safely, and reliably determine parameters and thresholds for emergency response, the cumulative absolute velocity (CAV) is used as the target parameter, and 7 P-wave characteristic parameters of strong ground motion records occurring 3 s after P-wave arrival at K-NET and KiK-net stations in Japan are used as inputs to construct a machine learning (ML) CAV prediction model based on the support vector machine (SVM) algorithm. The results show that compared with a single-parameter prediction algorithm, the proposed ML model can significantly reduce the error standard deviation and effectively address the phenomena of small value overestimation and large value underestimation. A confusion matrix analysis demonstrates that the 6-parameter model P_a&P_v&P_d&CAV&I_a&IV2 shows the best performance in improving the prediction accuracy and provides a threshold selection strategy for threshold-based EEW emergency response.