Automatic Seismic Phase Identification Method Based On Deep Learing

Earthquake disasters occur frequently in my country.In order to carry out earthquake relief and follow-up scientific research work in a timely manner,it is necessary to automatically identify seismic phases of seismic station monitoring data.However,traditional methods of automatic seismic phase identification have great limitations.As the amount of seismic data increases,traditional methods are difficult to process massive seismic data.At the same time,their identification results are also affected by parameters such as feature functions and thresholds,resulting in seismic phases.The accuracy of the recognition result is low,and the missed detection rate is high.With the continuous development of deep learning technology,its super learning ability is conducive to processing massive seismic data,and it also reduces manual setting of parameters.Its efficiency and accuracy are conducive to improving the accuracy of seismic phase recognition and reducing the rate of missed detection.These advantages of deep learning provide new research directions for the automatic identification of seismic phases.This paper proposes an automatic seismic phase identification method based on deep learning to achieve high-precision,low-missing detection of seismic phases.The main research work is as follows:(1)Aiming at the problems of low recognition accuracy and high missed detection rate of current phase recognition methods,the existing deep learning methods of seismic phase recognition are analyzed,and the advantages of the Bi-LSTM model in time series data processing are used to try to use Bi-The LSTM model performs automatic phase identification,solves the problem of missing phase detection in the existing phase identification methods,and improves the recognition accuracy.In order to test the effect of the Bi-LSTM model on the automatic identification of seismic phases,the Stanford University STEAD data set was used to construct the training set and the test set,and the Bi-LSTM model was trained and tested,and the phase identification results were compared with the artificial identification reference results.Contrast,and calculate three evaluation indicators of root mean square error,correct rate and missed detection rate to evaluate the effect of seismic phase recognition.Experimental results show that the Bi-LSTM model reduces the missed detection rate by 8-15% compared with the traditional method,which improves the problem of seismic phase missed detection to a certain extent.(2)Although the Bi-LSTM model reduces the missed detection rate of seismic phase identification,it still has no improvement in accuracy.In response to this problem,the U-shaped neural network and Bi-LSTM are combined to propose an automatic phase recognition method of the UBDN model.The U-shaped neural network has the advantage of high accuracy in edge detection,and the seismic data Seismic phase features are extracted;Bi-LSTM is then used to establish the time series relationship between seismic phase features to realize automatic identification of seismic phases with high precision and low missed detection rate.In order to prove the effectiveness of the UBDN model,the same data set is used to conduct comparative experiments on the UBDN model and other automatic phase identification methods,and the evaluation indicators are calculated.The experimental results show that the root mean square error of the UBDN model is about 0.26 s,and the correct rate and missed detection rate are about 87% and 13%,respectively.Compared with other models,the calculated root mean square error is reduced by about 0.2?0.3s,which is correct The rate has increased by about 5?14%,and the missed detection rate has been reduced by about 15?25%.The UBDN model has the best phase recognition effect and can realize automatic phase recognition with high precision and low missed detection rate.(3)In order to test the application effect of the UBDN model in actual earthquake events,the 2008 Wenchuan earthquake aftershock data set was used for seismic phase identification.Experiments have found that although the UBDN model meets the accuracy requirements for phase recognition on the Wenchuan earthquake aftershock data set,it can be applied in practice,but the performance of various indicators is not as good as the STEAD data set.Through the analysis of the two data sets,it is found that the signal-to-noise ratio of the STEAD test set data is significantly higher than that of the Wenchuan earthquake aftershock data set.In order to verify whether the data signal-to-noise ratio will affect the experimental results,five different signal-to-noise ratio ranges are selected The experimental results show that the larger the signal-to-noise ratio of the data set,the better the seismic phase recognition effect of the UBDN model.Although the performance of the UBDN model on the Wenchuan earthquake aftershock data set is not as good as the STEAD data set,the UBDN model still meets the accuracy requirements of seismic phase recognition under the condition of low data signal-to-noise ratio,and can meet the needs of practical applications.