Research On Identification Method Of Geomagnetic Disturbance Events Based On Deep Learning

Geomagnetic field observation data of high quality is to carry out the earthquake precursor anomaly analysis and the basis of study on the relationship between the magnetic,as the earth's magnetic field observation instrument of laid,and the expansion of the scope of human activities,interference increasing number of geomagnetic field observation instruments,the existing magnetic interference event recognition way,half man site is not only time-consuming,but also relies heavily on the expert experience.With the rapid development of Deep Learning(DL)technology,Convolutional Neural Networks(CNN)and Long Short-Term Memory Neural Network(LSTM)have obvious advantages in extracting morphological and temporal features of time series data.Self-Attention Mechanism(SA)also has a good performance in the field of feature amplification.Therefore,this paper uses a variety of deep learning technologies to build and train the geomagnetic interference event recognition model,and explores the feasibility of deep learning technology in the geomagnetic interference event recognition task.The main research work of this paper is as follows:(1)The classification of geomagnetic interference events and the morphological characteristics of the original data of several typical types of geomagnetic interference events are summarized.Based on this,the data distribution of various kinds of interference events from 2000 to 2019 is analyzed,and the geomagnetic field components which are most affected by the interference events are further analyzed.(2)Based on Convolutional Neural Network and Self-Attention Mechanism,a binary deep learning model CNN-SA for automatic recognition of interference events from geomagnetic observation data was proposed,and the sample making process was described in detail.The experimental results show that the CNN-SA model has a good recognition effect,compared with Multilayer Perceptron(MLP),Fully Convolution Network(FCN),Residual Network(Res Net),the classification accuracy is significantly improved.(3)Based on Convolutional Neural Network,Long Short-Term Memory and SelfAttention Mechanism,a six-category deep learning model CNN-LSTM-SA was constructed to distinguish five kinds of interference events from normal events.In the experimental part,the model uses data sets containing six categories for training.By comparing with the MLP,FCN and Res Net models,it is found that CNN-LSTM-SA model has a better effect on six categories of geomagnetic interference events.It shows that deep learning is a feasible method to accurately identify all kinds of interference events in geomagnetic field.