Deep Learning And Its Application In Geoelectric Field Anomaly Detection

The observation of ground electric field at a fixed station is of great significance to the monitoring and research of seismic events,because geoelectric field anomalies may be appearing in every stage of earthquake preparation and occurrence.In China,geoelectric field observation started from the mid-1980 s with reference to the VAN(Varotsos,Alexopoulous and Nomicos)method of Greece.After the "Ninth Five-Year Plan","Tenth Five-Year Plan" and the background field project,geoelectric field observation has begun to take shape in China.At present,there are 123 geoelectric field stations in Chinese mainland,which produce a large amount of data every day.How to effectively evaluate the quality and abnormality of the data is very important.At present,Deep learning has become one of the most popular machine learning techniques and has achieved remarkable achievements in the field of time series anomaly detection.Based on the introduction of Long short-term memory network,dilated causal convolution network,singular spectrum analysis and logarithmic probability density function method,simulating experiments are carried out to detect the outliers set in the simulated data,and the availability and effectiveness of the method are verified.And it is determined that the dilated causal convolution network method will be used to carry out the investigation in this paper.The method was applied to the observational data of Nanjing,Haian,Gaoyou geoelectric field stations in Jiangsu Province and Pingliang,Jiayuguan and Shandan geoelectric field stations in Gansu Province.Through the processes of dilated causal convolution network model establishment,training and actual prediction,some hourly mean values of the observed data at 6 stations are analyzed and studied in this paper.The results show that the logarithmic probability density(LPD)value of the prediction error of the model on the observation data of the Nanjing observatory decreased significantly within one and a half months before the Ms7.2 earthquake in the east China sea.In the three months before Wenchuan Ms8.0,two Wenchuan aftershocks,Yushu Ms7.1,Lushan Ms7.0 and Min-Zhang Ms6.6 earthquakes,the LPD value of the prediction error of the model on the observation data of Pingliang observatory was significantly reduced.In the three months before Wenchuan Ms8.0,Haixi Ms6.3 earthquakes,the LPD value of the prediction error of the model to the observation data of Jiayuguan observatory decreased significantly.In the two months before the earthquakes of Menyuan Ms6.4 and Ms5.2 in Qilian county,the LPD value of the prediction error of the model on the observation data of Shandan observatory was significantly reduced.In the month before Nepal Ms8.1,Alxa Right Banner Ms3.9,and Jiuzhaigou Ms6.4,the LPD value of the prediction error of the model on the observation data of Shandan observatory decreased.Combined with the spatial electromagnetic environment and the change of microcracks before the earthquake,the possible causes of the anomaly of the geoelectric field before the earthquake are tentatively explained theoretically.The successfully application of deep learning technique in geoelectric field data will greatly help to improve the utilization of observation data and the efficiency of anomaly detection,and provide technical support for the observation of data to better serve seismic investigations.