Earthquakes are known as the worst evil of natural disasters,and they have always received widespread attention from domestic and foreign experts and scholars.In particular,the occurrence of huge earthquakes and the resulting fires,mudslides,landslides,etc.,have brought serious losses to the safety of human lives and properties,and destroyed their homes.In the process of gestating an earthquake,the electromagnetic field,thermal field,and gravity field of the earth will have some changes.Because the ionospheric electron content TEC can reflect the active state of the ionosphere to a certain extent,it has become a study on the precursor ionospheric anomalies of earthquakes.Important data.However,how to extract earthquake-related anomalies from TEC data is the key to the study of earthquake precursor electromagnetic anomalies.This paper takes earthquakes of magnitude 7 or higher in the middle and low latitudes of the world as the research object,and uses the SVR support vector machine regression algorithm and the LSTM long and short-term memory model to extract and statistically analyze the total electron content(TEC)of the ionosphere.(1)Developed a method for constructing TEC non-seismic background field based on SVR model.Using SVR support vector machine method,construct a TEC non-seismic dynamic background field considering the influence of solar activity and geomagnetic activity,so as to more accurately extract earthquake-related anomalies.This paper compares the SVR non-seismic dynamic background field method with the traditional sliding time window method.The results show that the method in this paper does not extract anomalies during the non-seismic time,while the sliding time window method extracts anomalies,which is important for subsequent ionospheric anomalies.Detection will have an important impact.At the same time,this paper is divided into 6 study areas based on the type of geological structure(intra-plate inter-plate),based on the SVR non-seismic background field method,the pre-earthquake ionospheric TEC anomaly extraction and statistical analysis of 180 earthquakes of magnitude 7 or above in low and middle latitudes around the world.(2)Developed a method of constructing TEC non-seismic background field based on LSTM model.Using deep learning LSTM neural network,TEC,solar index,geomagnetic index and other data are used as input parameters for model training.In this paper,the LSTM neural network and the RNN recurrent neural network are compared in the non-seismic period.The results show that the fitting effect of the LSTM neural network is significantly better than the RNN recurrent neural network and no abnormalities are extracted.At the same time,this paper uses the TEC non-seismic background field method based on the LSTM model to extract ionospheric TEC anomalies 15 days before three earthquakes of magnitude 7 or above.The results show that different degrees of anomalies were extracted within 6 days before the three earthquakes,and When the anomaly appeared,it was not accompanied by strong solar and geomagnetic activities.Therefore,the use of deep learning LSTM neural network to extract ionospheric TEC anomalies before earthquakes is an effective attempt.It can eliminate non-seismic interference to a certain extent,thereby improving the accuracy of TEC anomaly detection before earthquakes. |