Study On Short-term And Impending Earthquake Forecasting In Sichuan-yunnan Region Based On Convolution Neural Network

Earthquake is a kind of natural phenomenon that can bring disaster to human society.The destructive earthquake will bring economic loss and casualties to the country and region.After a long-term exploration and research by experts and scholars in the earthquake industry in the world,significant progress has been made in earthquake prediction,especially in the medium and long term forecasting.However,the short-term and imminent earthquake prediction is still in the exploring stage,and its prediction level is far from the social demand.Therefore,since the 1980 s,China began to build a seismic network,and now has accumulated a large amount of high-precision precursory observation data such as deformation,geoelectricity,geomagnetism,gravity,fluid and so on.Based on the research and analysis of precursory observation data,considering the highly nonlinear characteristics between the preparation and occurrence of earthquakes and the subjectivity of artificial selection of seismicity parameters,and inspired by convolution neural network technology,this paper attempts to apply CNN algorithm to the field of earthquake prediction.We propose a short-term and imminent earthquake prediction model based on convolution neural network,and compare it with those models based on BP neural network and support vector machine algorithms.The main research work and results are as follows:1)proposed a method of seismic prediction region division based on clustering algorithmLocation prediction is an important part of earthquake prediction.In this paper,the area of 26.50°N~30.00°N,100.00°E ~105.50°E in Sichuan-Yunnan area is selected.Even if the time and magnitude of the future earthquake can be predicted accurately in such a large range,its practical value is greatly reduced.Therefore,it is necessary to divide the research area into small blocks.In theory,the attributes of the data in the block are close after clustering algorithm classification,and the attributes of different blocks are quite different,so we choose to use clustering algorithm to divide the historical seismic location in the region into six sub blocks.2)proposed a method of earthquake prediction based on the data fusion of multi term precursory observationAt present,most of the data used in earthquake prediction research are seismicity parameters,some of them use seismic catalogue data,and the prediction research using precursory observation data also uses a certain measurement item.The combination of multiple items of precursory data is an innovation of this paper.Before the earthquake,it may or may not have obvious influence on the data of a certain measurement item.Only using one item to predict will lose the abnormal information that other items may contain.Based on this,the multi-channel data composed of geomagnetic data,deformation data and water level data are used to predict earthquake magnitude and area,and the accuracy is increased to 95% through data enhancement and parameter optimization.3)explored a convolution neural network seismic prediction method based on precursory observation dataThanks to the automatic feature extraction ability of convolutional neural network,reasonable network structure design and super parameter optimization,its accuracy in the field of two-dimensional data such as image recognition has exceeded that of human beings.In the precursory observation data which belongs to one-dimensional data in essence,this paper designs a convolution neural network prediction method based on precursory observation data in principle,and optimizes the super parameters through quantitative analysis.Finally,this method can achieve high accuracy and good regional prediction ability at the same time.4)constructed a model that can predict both the region of occurrence and magnitude of earthquakeThere are a lot of researches on using machine learning method to predict the magnitude of a region in the future,but there are not many researches on predicting the region and magnitude of an earthquake at the same time.In this paper,a model is designed based on the characteristics of convolution neural network.The model input is multi-channel data composed of multiple observation data.After a group of feature extraction layers composed of convolution layer,BN layer,pooling layer and two layers of full connection layer,the model output is a label for predicting the earthquake occurrence area and magnitude range in the next day,which has a certain value in the field of earthquake prediction.