Research On Aftershock Prediction Based On Deep Learning

It is found that strong mainshock always triggers a series of aftershocks.Great aftershocks cause secondary damage of structures and collapse of buildings which lead to the serious casualties and huge economic loss.Therefore,making accurate aftershock prediction is of great importance.It can help direct the reconstruction and minimize the serious impact above.In this article,we collect the information of large numbers of history earthquakes as well as their aftershock catalogues.Based on the database and the relationship between mainshocks and aftershocks,we put forward three deep learning networks which can predict the location and magnitude of aftershocks.The main research contents are:(1)According to the correlation between the mainshock and the aftershock in time and space,historical earthquake events in 1968 to 2012 as well as their own aftershock sequences are selected from the SRCMOD database and the ISC database.The information of these earthquakes are collected and caculated,which is the data basis of the latter research.(2)The prediction effect of static Coulomb rupture stress and dynamic Coulomb rupture stress on the location of aftershocks are evaluated.An 8-layer deep learning network based on the combined effect of dynamic and static stress triggering is designed to predict the location of aftershocks.The accuracy of the prediction model is quantitatively evaluated and compared with other methods.(3)The correlation among the maximum magnitude of aftershocks and the magnitude,source mechanism,fault characteristics of the mainshocks is analysed.According to the result,the input of the network was selected.A 3-layer neural network is proposed to predict the maximum magnitude of aftershocks.On basis of this network,more information are included in the input parameters.a 5-layer neural network that can predict the magnitude of aftershocks at a certain location is proposed.The accuracy of the prediction results is evaluated.