Method for calculating earthquake relative travel time difference based on deep learning and DTW

Abstract

The invention discloses a method for calculating an earthquake relative travel time difference based on deep learning and DTW, and belongs to the technical field of earthquake signal processing, and the method comprises the steps: obtaining earthquake data which needs to be subjected to relative travel time difference calculation, and carrying out the preprocessing of the earthquake data through the seismic software, such as Obspy; meanwhile, an input file required by the PhaseNet is generated; inputting the processed data and the corresponding file into a PhaseNet to obtain a first arrival pickup result of all seismic data and outputting the first arrival pickup result; and on the premise that a seismic event pair of which the relative travel time difference needs to be calculated is known, performing similarity matching on the seismic events in the event pair by using a DTW technology in combination with a first arrival pickup result, and taking the first arrival time difference when the similarity of the two events reaches the maximum as the relative travel time difference and outputting the relative travel time difference. The method for calculating the relative travel time difference of the earthquake based on deep learning and DTW provided by the invention solves the problem that the calculation of the relative travel time difference by a WCC technology has a large error when the magnitude difference is significant, and improves the accuracy and efficiency of the calculation of the relative travel time difference.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic signal processing, and in particular to a method for calculating seismic relative travel time difference based on deep learning and DTW. Background Art

[0002] Earthquake location is a prerequisite for seismological research such as source mechanism analysis, seismic tomography, and fault activity analysis. It is also crucial for earthquake early warning, disaster reduction, and engineering design. In relative earthquake location, the calculation accuracy of the relative travel time difference between earthquake events (i.e., the time difference between the first arrival times of two earthquake events) has a direct impact on the accuracy of earthquake location. At present, waveform cross-correlation (WCC) is widely regarded as the most accurate method for calculating relative travel time differences. Studies have shown that using this method to calculate the relative travel time difference between earthquake events can significantly improve the accuracy of earthquake location results, achieving an improvement of one order of magnitude or more, and its accuracy even exceeds the level that can be achieved by manual calculation.

[0003] However, with the promotion of the application of this technology, its limitations have gradually become apparent. In practical applications, if the magnitude difference of earthquake events is large, the use of WCC to calculate the relative travel time difference will introduce a large error due to the mismatch of the first pulse width of the two earthquake waveforms. This error will cause the source location results to shift by several hundred meters. In this case, researchers usually manually delete earthquake events with significantly larger magnitudes in the data set (such as main shocks), and only use WCC to calculate the relative travel time difference for the remaining earthquakes and relocate them to improve the reliability of earthquake location results. Although this method can effectively reduce the errors introduced by magnitude differences, it may also weaken the comprehensiveness and accuracy of the positioning results, which is ultimately not conducive to subsequent detailed interpretation and comprehensive analysis. In addition, waveform cross-correlation technology involves a lot of calculations, and seismic data often have a huge amount of data, which also makes it a huge challenge to use waveform cross-correlation technology to efficiently calculate relative travel time differences. Summary of the invention

[0004] The purpose of the present invention is to provide a method for calculating the relative travel time difference of earthquakes based on deep learning and DTW, aiming to solve the problem that when there is a significant difference in earthquake magnitude, the use of WCC technology to calculate the relative travel time difference of two earthquake events will introduce large errors.

[0005] To achieve the above object, the present invention provides a method for calculating earthquake relative travel time difference based on deep learning and DTW, comprising the following steps: S1. Data preparation: obtain the earthquake data for which relative travel time difference needs to be calculated, preprocess the data using seismological tools, and generate the station list and data name list files required to run PhaseNet; S2, first arrival picking, input the station list and data name list file obtained in S1 and the processed earthquake data into PhaseNet, perform first arrival picking on all earthquake data, output the first arrival picking results of each earthquake event and save them; S3, relative travel time difference calculation, according to the actual application requirements, obtain the information of the earthquake event pair for which the relative travel time difference needs to be calculated, use the DTW technology combined with the first arrival picking results in S2 to perform similarity matching on the earthquake events in the event pair, and use the first arrival time difference when the similarity of the two events reaches the maximum as its relative travel time difference and output it.

[0006] Preferably, the preprocessing in S1 includes filtering and denoising the data, removing linear trends, and converting the data into a SAC format that meets the PhaseNet processing requirements.

[0007] Preferably, S2 uses PhaseNet to quickly process the seismic data obtained in S1, accurately and efficiently pick up its first arrival and output it.

[0008] Preferably, S3 is specifically: Any earthquake in the event pair is used as a template sequence, and a 0.2s waveform is intercepted starting from its first arrival time obtained in S2. The waveform is then matched with the waveform of the other earthquake in the range of -0.06s to 0.3s before and after its first arrival using the DTW technology for similarity matching.

[0009] Therefore, the present invention adopts the above-mentioned method for calculating the relative travel time difference of earthquakes based on deep learning and DTW, uses PhaseNet for first arrival picking, and selects a small range of waveforms near the first arrival of the target earthquake for relative travel time difference calculation, which greatly reduces the amount of data for similarity calculation and improves the calculation efficiency. At the same time, the use of DTW technology to calculate the relative travel time difference effectively solves the problem of large errors in WCC calculation when the magnitude difference is significant, thereby improving the calculation accuracy of the relative travel time difference.

[0010] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 is a flow chart of a method according to an embodiment of the present invention; Figure 2Schematic diagram of the specific application of an embodiment of the present invention, wherein (a) represents the alignment result of similarity matching of two seismic waveforms with significant magnitude difference using WCC technology, and (b) represents the alignment result of similarity matching of two seismic waveforms with significant magnitude difference using DTW technology. DETAILED DESCRIPTION

[0012] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0013] See also Figure 1-Figure 2 , a method for calculating earthquake relative travel time difference based on deep learning and DTW, including the following steps: S1. Data preparation: obtain the earthquake data for which relative travel time difference needs to be calculated, and use seismological tools such as Obspy to perform preprocessing on the data such as filtering, denoising, and de-linear trending, convert the data format into the SAC format that meets the PhaseNet processing requirements, and generate the station list and data name list files required to run PhaseNet.

[0014] PhaseNet is a seismic signal processing model based on deep learning, which is mainly used to automatically identify and pick up the first arrival time of seismic waves. It uses a convolutional neural network (CNN) architecture and can process large amounts of seismic data quickly and accurately. It is currently widely used in various aspects of seismological research.

[0015] S2, first arrival picking, input the station list and data name list file obtained in S1 and the processed earthquake data into PhaseNet, perform first arrival picking on all earthquake data, output the first arrival picking results of each earthquake event and save them.

[0016] S3, relative travel time difference calculation, according to the actual application requirements, obtain the information of earthquake event pairs for which relative travel time difference needs to be calculated (for example, in earthquake relocation, use HypoDD to obtain event pairs), use DTW technology combined with the first arrival picking results in S2 to perform similarity matching on the earthquake events in the event pairs, and use the first arrival time difference when the similarity of the two events reaches the maximum as its relative travel time difference and output it.

[0017] Specifically: Any earthquake in the event pair is used as a template sequence, and a 0.2s waveform is intercepted starting from its first arrival time obtained in S2. The waveform is then matched with the waveform of the other earthquake in the range of -0.06s to 0.3s before and after its first arrival using the DTW technology for similarity matching.

[0018] Dynamic Time Warping (DTW) is a nonlinear algorithm used to calculate the similarity of two time series. It allows the time series to be elastically scaled when the similarity matches, so as to achieve the most realistic and accurate alignment between the two series. The algorithm has high flexibility in calculation and can perform similarity analysis on time series of different lengths. Therefore, it is welcomed by many fields such as speech recognition, financial data analysis, and biological signal processing.

[0019] Figure 2 (a) in the figure shows the alignment result of similarity matching of two earthquake waveforms with significant magnitude differences using WCC technology. The gray curve represents the template earthquake waveform with a larger magnitude, and the black curve represents the matched waveform with a smaller magnitude. The gray dashed line and the black solid line represent the WCC picked first arrival and the true first arrival, respectively. The time interval between them is the relative travel time difference of the two earthquake waveforms.

[0020] (b) shows the alignment result of similarity matching of two earthquake waveforms with significant magnitude differences using DTW technology. The gray curve and the black curve have the same meaning as above. The gray dotted line and the black solid line represent the first arrival picked up by the method of the present invention and the true first arrival, respectively. The time interval between them is the relative travel time difference of the two earthquake waveforms.

[0021] Therefore, the present invention adopts the above-mentioned method for calculating the relative travel time difference of earthquakes based on deep learning and DTW, uses PhaseNet for first arrival picking, and selects a small range of waveforms near the first arrival of the target earthquake for relative travel time difference calculation, which greatly reduces the amount of data for similarity calculation and improves the calculation efficiency. At the same time, the use of DTW technology to calculate the relative travel time difference effectively solves the problem of large errors in WCC calculation when the magnitude difference is significant, thereby improving the calculation accuracy of the relative travel time difference.

[0022] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for calculating relative travel time difference of earthquakes based on deep learning and DTW, characterized in that: The following steps are involved: S1. Data preparation: obtain the earthquake data for which relative travel time difference needs to be calculated, preprocess the data using seismological tools, and generate the station list and data name list files required to run PhaseNet; S2, first arrival picking, input the station list and data name list file obtained in S1 and the processed earthquake data into PhaseNet, perform first arrival picking on all earthquake data, output the first arrival picking results of each earthquake event and save them; S3, relative travel time difference calculation, according to the actual application requirements, obtain the information of the earthquake event pair for which the relative travel time difference needs to be calculated, use the DTW technology combined with the first arrival picking results in S2 to perform similarity matching on the earthquake events in the event pair, and use the first arrival time difference when the similarity of the two events reaches the maximum as its relative travel time difference and output it.

2. The method for calculating relative travel time difference of earthquakes based on deep learning and DTW according to claim 1, characterized in that: The preprocessing in S1 includes filtering and denoising the data, removing linear trends, and converting the data into the SAC format that meets the processing requirements of PhaseNet.

3. The method for calculating relative travel time difference of earthquakes based on deep learning and DTW according to claim 2, characterized in that: S2 uses PhaseNet to quickly process the seismic data obtained in S1 to achieve its first arrival picking.

4. The method for calculating relative travel time difference of earthquakes based on deep learning and DTW according to claim 3, characterized in that: S3 is specifically: Any earthquake in the event pair is used as a template sequence, and a 0.2s waveform is intercepted starting from its first arrival time obtained in S2. The waveform is then matched with the waveform of the other earthquake in the range of -0.06s to 0.3s before and after its first arrival using the DTW technology for similarity matching.

Images