Machine Learning Applied in Auto-classify of Seismic Events

In this paper, SERS （Seismic Event Recognition System） was used to classify 20 seismic events with doubtful classification in the east of Inner Mongolia cataloged by Inner Mongolia Regional Seismological Network from 2019 to 2022. Recognition rate of the model was 90.4%. Comparing with manual catalog, only 3 events were recognized as natural earthquakes, with lower probability. According to the results of amplitude-ratio and time-frequency analysis of these events, P waves of the stations close to the epicentre have evident characteristics of blast events, such as high-amplitude and high-energy. And combined with relative documents studying the geological formation and site response, the paper speculates that the classification of seismic events is mainly site-influenced.