Identification of earthquakes and microtremors based on a combined model of generative adversarial network and random forest

Abstract

Seismic event identification is one of the basic tasks of earthquake monitoring. Especially with the construction of seismic stations, the collection of massive seismic data, and the development of earthquake early warning (EEW), how to distinguish between earthquakes and noise from continuous waveforms become increasingly vital. To accurately identify seismic events, a combined model based on a generative adversarial network (GAN) and a Random Forest (RF) is provided to distinguish between earthquakes and microtremors. We first train a GAN with 52537 strong ground motion records, and use the modified discriminator to extract waveform features, and then use a RF to discriminate seismic events in the testing set which has 5378 data of earthquakes and microtremors, thereby transforming seismic events identification into a classification problem. The results show that, the classification accuracy of the combined model for classifying seismic events and microtremor can reach more than 99%. It illustrates that the proposed model can accurately identify seismic events and has application prospects in earthquake monitoring.