Research On The Method Of Predicting Medium And Strong Earthquakes Based On Deep Learning

China is one of the countries with the worst earthquake disasters in the world,and earthquakes usually cause serious casualties and property damage.Earthquake prediction is a common scientific problem worldwide,but researchers have not stopped exploring.China started to build a seismic station network in the 1980 s and set up the China Earthquake Science Experimental Field in Sichuan and Yunnan in2018,and after years of efforts by seismologists,a large amount of high-quality multidisciplinary earthquake precursor observation data,such as geomagnetic field,electromagnetic field,gravity,and fluid,have been accumulated by now.In this paper,we apply deep learning algorithms to study and analyze the geomagnetic field precursor observation data to explore the complex nonlinear relationship between earthquake breeding and earthquake precursors.Using the seismic science experimental site in China,which is prone to earthquakes,as the study area,we construct a deep learning algorithm-based earthquake prediction model for moderate to strong earthquakes based on long-time geomagnetic field precursor observation data,and conduct a comparative experimental study by combining deep learning algorithms such as convolutional neural network and Transformer.The experimental results show the feasibility of convolutional neural networks and Transformer in earthquake prediction work,which provides a new research direction for earthquake prediction work.The main research work and results of this paper are as follows.(1)A classification summary of the recent earthquake prediction work.In this paper,earthquake prediction methods are classified into three major categories,which are earthquake prediction methods based on precursor anomaly statistics,earthquake prediction methods based on traditional machine learning,and earthquake prediction methods based on deep learning.This paper summarizes and analyzes the shortcomings and limitations of these earthquake prediction methods in detail,and looks forward to the future research directions of earthquake prediction.(2)At present,most of the research work on earthquake prediction uses seismic activity parameters or seismic waveform data combined with earthquake catalogs,but this paper uses geomagnetic field precursor observation data and innovatively combines traditional seismic-magnetic anomaly feature extraction methods and deep learning methods to build earthquake prediction models with good results.(3)In this paper,E-CNN and E-Transformer,a medium-intensity earthquake prediction model,are constructed using convolutional neural network and Transformer,respectively.After the models are trained,the advantages of the models in earthquake prediction tasks are verified by comparison experiments with MLSTM-FCN,MALSTM-FCN,LSTM-FCN,ALSTM-FCN,LEMPCNN and2016 BP models.