| With the large-scale deployment of modern digital seismometers and the increasing frequency of global geological activities,the output of seismic signal data is unprecedented.The rapid processing and recognition of seismic signal is the first problem to be solved in the face of massive seismic signal data.Seismic P-waves(longitudinal waves)are the first waveforms that reach the surface after an earthquake.The pickup of seismic P-waves when they arrive is a key step in seismic signal processing.Its high-precision pickup is very important in seismological problems such as epicenter location and focal mechanism research.significance.Aiming at the problems of low intelligence,low processing efficiency,and poor picking accuracy of the existing arrival picking technology,this paper carried out the seismic P-wave arrival picking modeling based on deep learning technology,and verified the scientificity and effectiveness of the model construction through experiments.sex.The specific work of this paper is as follows:(1)Aiming at the problems of insufficient efficiency,generalization ability and poor time series analysis ability of the existing seismic phase pickup methods,an improved network for seismic P-wave arrival pickup is proposed and constructed based on the integration of inception structure and GRU.Compared with the traditional method,the model does not need to manually set the threshold,and only needs to input the three-component data of the seismic waveform to intelligently identify the arrival of the seismic P-wave.The model has a lightweight structure and excellent feature extraction ability,which can effectively identify data with low signal-to-noise ratio and has strong robustness.The experimental results show that,within the allowable error of 0.1s,0.3s,and 0.5s,the P-wave pickup rate reaches 74.45%,96.79%,and 98.68%,respectively,and the pickup error when the P-wave arrives is 0.031 s.The performance is better than AR-Traditional methods such as AIC+STA/LTA and mainstream deep learning methods such as GRU.(2)Aiming at the problems of poor pickup accuracy and insufficient robustness of the existing seismic waveform arrival pickup methods,this paper designs a new seismic P-wave arrival automatic pickup model based on deep learning technology.The model is based on the UNet and realizes the P-wave arrival pickup by point prediction of seismic signals.In this model,firstly,the encoder is used to analyze and extract different scale features of the seismic signal data input to the network,and the potential features of the data are mined.Secondly,combined with the residual structure,the data features are further explored on the premise of ensuring network activity,strengthening the model’s ability to perceive the time information.Then the spatiotemporal attention mechanism is added to the decoder to increase the time feature weight,which further improves the network picking accuracy and helps the network converge faster.Finally,the encoding and decoding feature fusion is effectively controlled to reduce the shallow feature fusion and prevent the noise information from polluting the arrival feature sequence.Through comparative experiments,it can be seen that under different allowable errors(0.1s,0.2s,0.3s),the pickup hit rate of the constructed model reaches 75.04%,94.6%,and 97.37%,respectively,and the mean square error and mean absolute error are0.036 and 0.092 s respectively.Compared with the existing traditional arrival picking method and the existing deep learning automatic arrival picking method,the model has higher P-wave arrival picking accuracy. |
