Geoelectric Field Anomaly Detection And Classification Based On Machine Learning

Realizing the intelligent analysis and processing of massive seismic observation data is a must for earthquake scientists to move toward innovation-driven applications in the era of big data.Among them,as one of the important means of earthquake monitoring,geoelectric field observation,the mining of its anomalous information is particularly important.In China,a large amount of geoelectric field observation data is produced every day.It is especially important to analyze the quality of geoelectric field observation data effectively and make seismic anomaly determination.In order to identify and extract effective information from the geoelectric field signals,previous researchers have tried to process the information by spectral analysis,Fourier transform and wavelet transform,etc.Although the corresponding anomalies have been found to some extent,there are some disadvantages as well.These methods are time-consuming and inefficient,which affect our ability to investigate and detect a large amount of geoelectric field data one by one to some extent.In the 1990 s,scientists first tried to use artificial intelligence to predict earthquakes,but little progress was made at that time due to the relative backwardness of computer hardware and software performance and the lack of large amounts of data support.In today's era of big data,various high-performance technologies are gradually proposed.Meanwhile,the computing power and data storage capacity of computers have been significantly improved,bringing new opportunities for AI prediction earthquake research.Due to the great achievements of machine learning in the field of artificial intelligence and the difficulty of extracting the signal characteristics of seismic geoelectric fields,we try to introduce machine learning methods into the analysis and processing of geoelectric field observation data,and carry out research exploration work on the detection and classification of seismic anomalies in geoelectric fields using the actual observation data of stations.Based on the publicly reported research results of seismic geoelectric field,and considering the time range of the observations and the number of data samples in the training set,the geoelectric field waveform data of 20 days before the seismic event is considered to be closely related to the seismic event in the process of constructing the training samples.In this study,observations from the geoelectric field observation stations in Pingliang,Gansu,and Hotan,Xinjiang,for the past ten years were analyzed in conjunction with seismic events in the areas surrounding the stations.The geoelectric field anomaly time series before the occurrence of seismic events and the normal time series samples without obvious seismic events are trained uniformly.And support vector machines in machine learning and comparing gradient boosting tree algorithm and random forest algorithm are used to extract the waveform features of geoelectric field signals and classify the geoelectric field data according to the seismic precursor waveform information.Based on this,the training results of different test sets in different regions using different methods are also tried.Through the analysis and research,it is known that: machine learning related methods are applicable to seismic anomaly detection of geoelectric field observation data,and the models trained by using machine learning methods can achieve the purpose of anomaly detection and classification;compared with the support vector machine algorithm,the gradient boosting tree algorithm and the random forest algorithm have better classification effects on geoelectric field anomaly classification,and compared with the random forest algorithm,the gradient boosting tree algorithm shows better performance in more directions The gradient boosting tree algorithm showed better performance in more directions than the random forest algorithm.The classification accuracy reaches 69% and 65% in Pingliang,Gansu and Hotan,Xinjiang geoelectric field stations,respectively.The successful application of machine learning to geoelectric field observation data provides a new solution for processing and analyzing a large amount of observation data,effectively classifies and mines time series data for anomalies.Besides,it can improve the detection efficiency,and provide a powerful technical tool for geoelectric field data.These reasearch can support the earthquake field effectively.