Template matching earthquake detection method based on artificial intelligence

Abstract

The invention discloses a template matching seismic detection method based on artificial intelligence. The method comprises the following steps: extracting effective seismic phase data by using a deep learning detection model; taking the effective seismic waveform as a machine detection seismic waveform template; intercepting an artificial seismic monitoring waveform as an artificial detection seismic waveform template; co-building a template seismic waveform library; obtaining a template waveform of each seismic phase; and intercepting a to-be-detected continuous waveform with a suspected seismic event from the continuous waveform data, and carrying out matching calculation on the to-be-detected continuous waveform and a template waveform of each seismic phase of each seismic waveform template in a template seismic waveform library. According to the template matching seismic detection method, the to-be-detected continuous waveform and the template waveform of each seismic phase of each seismic waveform template in the template seismic waveform library are matched and calculated, so that the to-be-detected continuous waveform can be quickly and accurately identified and judged, and whether a seismic event exists in the to-be-detected continuous waveform or not can be quickly judged.

Description

Technical Field

[0001] The invention relates to an earthquake detection method, in particular to an artificial intelligence-based template matching earthquake detection method. Background Art

[0002] Accurate earthquake source parameters are an important prerequisite for studying seismic active structures and hidden faults. There are two main types of traditional earthquake detection technologies. One is based on seismic phase picking. Commonly used methods include manual picking, long and short window methods, and machine learning methods. This type of method integrates time picking, seismic correlation, and earthquake positioning. It can identify and detect a large number of earthquakes, but the accuracy of seismic phase picking of this type of method is reduced for waveforms with low signal-to-noise ratio or when there are many other seismic phases.

[0003] The other type is waveform-based earthquake detection technology, commonly used are MFT matched filtering method, M&L matching positioning method and image-based GPU accelerated detection technology. This type of method mainly uses waveform cross-correlation technology, uses the waveform of a specified time window to scan continuous waveforms, and determines whether there is an earthquake to be detected based on the size of the cross-correlation coefficient. This type of method has a smaller lower limit for earthquake detection, and is especially suitable for repeated earthquake detection, but has a higher generalization ability for template matching, and the overall calculation is relatively time-consuming, so it is more suitable for waveform detection with a low signal-to-noise ratio.

[0004] To sum up, although the two existing types of technologies have been widely used in the field of earthquake detection and positioning, neither the phase-based earthquake detection technology nor the waveform-based earthquake detection technology can achieve the requirements of both fast and accurate earthquake detection, and cannot solve the problem of imbalance between the phase picking accuracy and the earthquake picking speed in the existing technology. Summary of the invention

[0005] The purpose of the invention is to provide a template matching earthquake detection method based on artificial intelligence to solve the problem of imbalance between earthquake detection accuracy and detection speed in the prior art.

[0006] Technical solution: The template matching earthquake detection method based on artificial intelligence described in the present invention comprises the following steps: Step 1: Obtain the original continuous waveform data of historical earthquake monitoring at each seismic station, and use the trained deep learning detection model to extract the effective seismic phase data from the original continuous waveform data; Step 2, correlating the extracted effective seismic phase data to determine the effective seismic waveforms therein, and using the effective seismic waveforms as machine detection seismic waveform templates; Step 3, obtaining the identified artificial earthquake catalog, and intercepting artificial earthquake monitoring waveforms according to the earthquake events of each seismic station, and then using the intercepted artificial earthquake monitoring waveforms as templates for each artificial detection earthquake waveform; Step 4, merging the machine-detected seismic waveform template and the manually-detected seismic waveform template to jointly build a template seismic waveform library; Step 5, intercepting each seismic waveform template in the template seismic waveform library to obtain the template waveform of each seismic phase; Step 6, read the continuous waveform data to be detected, extract the continuous waveform to be detected that is suspected to contain an earthquake event from the continuous waveform data, and match the continuous waveform to be detected with the template waveform of each seismic phase of each seismic waveform template in the template seismic waveform library. If the calculation result shows that the waveform matching is successful, it is determined that an earthquake event exists in the continuous waveform data to be detected. If the calculation result shows that the waveform matching fails, it is determined that no earthquake event exists in the continuous waveform data to be detected.

[0007] Furthermore, in step 1, the specific steps of extracting effective seismic phase data from the original continuous waveform data using the trained deep learning detection model are as follows: Step 1.1, setting the time window length and step length for segmented interception of the original continuous waveform data, and then sequentially intercepting the original continuous waveform data of each seismic station according to the time window length and step length to obtain each segmented waveform of each seismic station; Step 1.2, input each segmented waveform into the trained deep learning detection model in sequence, and use the deep learning detection model to extract and calculate the probability values ​​of the P phase and the S phase in each segmented waveform; Step 1.3, analyzing and judging the probability values ​​of the P phase and the S phase in each segmented waveform: if the probability value of the P phase and the probability value of the S phase of a segmented waveform are both greater than the segment preset threshold, the current segmented waveform is judged to be valid phase data, otherwise it is judged that there is no valid phase data in the current segmented waveform; Step 1.4, repeat step 1.3 until all segmented waveforms of each seismic station are analyzed and judged, and all valid seismic phase data of each seismic station are output respectively, and each valid seismic phase data of each seismic station is arranged in chronological order.

[0008] Furthermore, in step 2, the specific steps of correlating the extracted effective seismic phase data to determine the effective seismic waveforms therein are: Step 2.1, obtaining the coordinates of the seismic station points corresponding to each effective seismic phase data and the acquisition time of each effective seismic phase data, and then setting the source limiting circle for limiting the possible range of the source, the extraction time interval for defining the data extraction time range, and the effective waveform interception time length; Step 2.2, according to the acquisition time sequence, the effective seismic phase data of each seismic station are allocated to the extraction time intervals of the corresponding time in the sequential arrangement, and then the effective seismic phase data of each seismic station in each extraction time interval are combined into a seismic phase data group; Step 2.3, taking out a seismic phase data group in sequence according to the arrangement time sequence, and decomposing the P seismic phase wave and the S seismic phase wave in each valid seismic phase data in the current seismic phase data group; Step 2.4, determine whether there is any valid seismic phase data that has not been taken out in the seismic phase data group. If there is any valid seismic phase data that has not been taken out, take out a valid seismic phase data and go to step 2.5, otherwise go to step 2.6; Step 2.5, determine the correlation between the P phase wave and the S phase wave in the extracted valid phase data: if the S phase wave arrival time in the valid phase data of a certain seismic station is greater than the P phase wave arrival time, then determine that the current valid phase data is possible event data, and then return to step 2.4; otherwise, determine that the current valid phase data is invalid event data, delete the current valid phase data, and then return to step 2.4; Step 2.6, determine whether the number of possible event data in the current seismic phase data group is greater than the event number threshold. If the number is less than the event number threshold, the current seismic phase data group is determined to be an invalid data group, and then proceed to step 2.10, otherwise proceed to step 2.7; Step 2.7, calculate the distance D between each seismic station and the earthquake source according to the arrival time of the S phase wave, the arrival time of the P phase wave, the velocity of the S phase wave and the velocity of the P phase wave in each possible event data, and then draw each earthquake source location circle with the coordinates of the earthquake point of each seismic station as the center and the corresponding distance D as the radius. After drawing the earthquake source location circle corresponding to each possible event data, mark the intersection of each earthquake source location circle as the positioning reference point; Step 2.8, select each positioning reference point as the center of the source limiting circle in turn, and then determine the number of positioning reference points in each source limiting circle respectively, and then compare the determined number of points to find the maximum number of points. If the maximum number of points is greater than the point number threshold, the current seismic phase data group is determined to be a valid data group, and then proceed to step 2.9; otherwise, the current seismic phase data group is determined to be an invalid data group, and then proceed to step 2.10; Step 2.9, find the source limiting circle corresponding to the maximum number of points, then determine each positioning reference point within the source limiting circle, then find the corresponding source positioning circle based on each positioning reference point, then determine each seismic station based on each source positioning circle, and finally extract the effective seismic waveform from the effective seismic phase data of each seismic station based on the effective waveform interception time length, and the effective seismic waveform includes the complete waveform of the S seismic phase wave and the P seismic phase wave; Step 2.10, determine whether there are any unretrieved seismic phase data groups, if there are any unretrieved seismic phase data groups, return to step 2.3, otherwise output the effective seismic waveforms of each seismic station.

[0009] Furthermore, in step 3, the specific steps of intercepting artificial earthquake monitoring waveforms according to earthquake events at each seismic station are: Step 3.1, determine the earthquake events of each seismic station according to the identified artificial earthquake catalog, and obtain the artificial monitoring data of each earthquake event; Step 3.2, using TAUP ray tracing technology to calculate the arrival time of the S phase wave and the P phase wave of each earthquake event at each seismic station; Step 3.3, set the starting point of the effective waveform interception interval to the 7 seconds before the arrival of the P seismic phase wave, and then set the end point of the effective waveform interception interval to the 15 seconds after the arrival of the P seismic phase wave, and then extract the corresponding artificial earthquake monitoring waveform from the artificial monitoring data of each seismic station according to the set effective waveform interception interval.

[0010] Furthermore, in step 4, the specific steps of merging the machine-detected seismic waveform template and the manually-detected seismic waveform template to jointly build a template seismic waveform library are: Step 4.1, collecting each machine-detected seismic waveform template and each manually-detected seismic waveform template, and obtaining the seismic stations corresponding to the machine-detected seismic waveform template and the seismic stations corresponding to the manually-detected seismic waveform template; Step 4.2, using the names of the seismic stations as the first-level directories of the template seismic waveform library, and placing the machine-detected seismic waveform templates and the manual-detected seismic waveform templates into the corresponding first-level directories; Step 4.3, extract the P seismic phase wave in the machine-detected seismic waveform template or the manually-detected seismic waveform template under each first-level directory, classify the obtained seismic waveform templates according to the amplitude of the P seismic phase wave, and use the classified levels as the second-level directory, and then put each machine-detected seismic waveform template and each manually-detected seismic waveform template into the corresponding second-level directory respectively. If there are multiple seismic waveform templates under a certain level, delete the redundant seismic waveform templates so that only one seismic waveform template is retained in each level, and finally a template seismic waveform library is formed.

[0011] Furthermore, in step 4.3, the specific steps of classifying the obtained seismic waveform templates according to the amplitude of the P seismic phase wave are as follows: First, decompose each machine-detected seismic waveform template and each manually-detected seismic waveform template collected in step 4.1 to obtain the P seismic phase wave in each seismic waveform template; Then, the maximum amplitude value in each P seismic phase wave is screened out and defined as Tmax, and the minimum amplitude value in each P seismic phase wave is screened out and defined as Tmin; Then, 0.8Tmin is set as the minimum level value, 1.6Tmax is set as the maximum level value, and 8 level values ​​are set equally between 0.8Tmin and 1.6Tmax. The amplitude of the P phase wave is divided into 10 level intervals using the minimum level value, the maximum level value, and the 8 level values ​​in between. Finally, the amplitude of the P seismic phase wave is determined according to which level of the 10 level intervals the amplitude of the P seismic phase wave belongs to, which is the classification level of the corresponding seismic waveform template.

[0012] Furthermore, in step 5, the specific steps of intercepting each seismic waveform template in the template seismic waveform library to obtain the template waveform of each seismic phase are: Step 5.1, obtaining a seismic waveform template in a secondary directory in a template seismic waveform library, and separating a P seismic phase wave and an S seismic phase wave from the obtained seismic waveform template; Step 5.2, respectively intercept the separated P seismic phase wave and S seismic phase wave: the waveform in the interval from 1s before to 5s after the P seismic phase wave is used as the template waveform of the P seismic phase wave, and the waveform in the interval from 1s before to 5s after the S seismic phase wave is used as the template waveform of the S seismic phase wave; Step 5.3, using the obtained P phase wave template waveform and S phase wave template waveform to replace the seismic waveform template in the corresponding secondary directory, and then judging whether there are still seismic waveform templates in the template seismic waveform library that have not been intercepted, if there are still seismic waveform templates that have not been intercepted, then return to step 5.1, otherwise go to step 5.4; Step 5.4, update the template seismic waveform library: name the P seismic phase wave template waveform in each secondary directory as P wave, and name the S seismic phase wave template waveform in each secondary directory as S wave.

[0013] Furthermore, in step 6, the specific steps of extracting the continuous waveform to be detected that is suspected to contain an earthquake event from the continuous waveform data are: Step 6.1.1, extracting each data segment of the waveform in the continuous waveform data, and then performing waveform analysis on each of the extracted data segments to obtain a suspected waveform of each data segment; Step 6.1.2, using the deep learning detection model to identify the separated suspected waveforms, and determine whether each suspected waveform is a P seismic phase waveform or an S seismic phase waveform. If the P seismic phase waveform and the S seismic phase waveform are identified, proceed to step 6.1.3; otherwise, it is determined that there is no earthquake event in the continuous waveform data, and then proceed to step 6.1.6; Step 6.1.3, statistically judge the identified P seismic phase waveform and S seismic phase waveform. If there is one P seismic phase waveform and one S seismic phase waveform, proceed to step 6.1.4. If there are multiple P seismic phase waveforms and S seismic phase waveforms, proceed to step 6.1.5. Step 6.1.4, waveform matching of the S seismic phase wave and the P seismic phase wave. If the S seismic phase wave and the P seismic phase wave belong to the same earthquake source and the arrival time of the S seismic phase wave is greater than the arrival time of the P seismic phase wave, it is determined that there is a suspected earthquake event in the continuous waveform data, and then the S seismic phase wave and the P seismic phase wave are spliced ​​into a continuous waveform to be tested in time sequence according to the arrival time of the S seismic phase wave and the arrival time of the P seismic phase wave, and then the process proceeds to step 6.1.6. Otherwise, it is determined that there is no earthquake event in the continuous waveform data, and then the process proceeds to step 6.1.6. Step 6.1.5, waveform matching is performed on each P seismic phase waveform and each S seismic phase waveform, and the matched P seismic phase waveform and S seismic phase waveform are taken as a seismic matching group. If the number of seismic matching groups obtained by matching is ≥ 1, it is determined that there is a suspected earthquake event in the continuous waveform data, and then the S seismic phase waves and P seismic phase waves in each seismic matching group are spliced ​​into each continuous waveform to be tested according to the time sequence of the arrival time of the S seismic phase wave and the arrival time of the P seismic phase wave, and then the process proceeds to step 6.1.6. Otherwise, it is determined that there is no earthquake event in the continuous waveform data, and then the process proceeds to step 6.1.6; Step 6.1.6, output the judgment result: if it is judged that there is a suspected earthquake event in the continuous waveform data, then the spliced ​​continuous waveforms to be tested are output; if it is judged that there is no earthquake event in the continuous waveform data, then the result that no continuous waveform to be tested is obtained is output.

[0014] Furthermore, in step 6.1.5, the specific steps of waveform matching for each P seismic phase waveform and each S seismic phase waveform are as follows: Step a, filtering each P seismic phase waveform and each S seismic phase waveform to obtain a denoised P seismic phase waveform and an S seismic phase waveform; Step b, arranging the S seismic phase waveforms in order from small to large in time to form an S wave sequence, and arranging the P seismic phase waveforms in order from small to large in time to form a P wave sequence; Step c, extract the S seismic phase waveform with the smallest arrival time from the S wave sequence, and then find out from the P wave sequence all P seismic phase waveforms with smaller arrival times than the extracted S seismic phase waveform. If no corresponding P seismic phase waveform is found, proceed to step h; Step d, obtaining the maximum amplitude of the extracted S seismic phase waveform as Hs and the maximum amplitude of each P seismic phase waveform found as Hp, then setting the height of the interception line as 20% Hs, using the interception line with a height of 20% Hs to intercept the extracted S seismic phase waveform and each P seismic phase waveform found, obtaining the distribution time width ΔS of the S seismic phase waveform above the interception line and the distribution time width ΔP of each P seismic phase waveform, and then counting the number of times Ns that the S seismic phase waveform crosses the interception line up and down within the range of the distribution time width ΔS and the number of times Np that the corresponding P seismic phase waveform crosses the interception line up and down within the range of each distribution time width ΔP; Step e, finding a P seismic phase waveform with Hp<Hs in each P seismic phase waveform, and using the found P seismic phase waveform as the P seismic phase waveform after primary screening, if the corresponding P seismic phase waveform is not found, proceeding to step h; Step f, finding each P seismic phase waveform with 0.85<Np / Ns<1.15 in the P seismic phase waveform after the primary screening, and using the found P seismic phase waveform as the P seismic phase waveform after the secondary screening, if the corresponding P seismic phase waveform is not found, proceeding to step h; Step g, finding the distribution time width ΔP with the smallest difference with the distribution time width ΔS in the P seismic phase waveform after the secondary screening, and determining the P seismic phase waveform corresponding to the found distribution time width ΔP as matching the current S seismic phase waveform, and deleting the found P seismic phase waveform from the P wave sequence, and then entering step h; Step h, determining whether there is an S seismic phase waveform in the S wave sequence, if there is an S seismic phase waveform, returning to step c, otherwise proceeding to step i; Step i: output matching P phase waveforms and S phase waveforms in pairs.

[0015] Furthermore, in step 6, the specific steps of matching and calculating the continuous waveform to be detected with the template waveform of each seismic phase of each seismic waveform template in the template seismic waveform library are as follows: Step 6.2.1, determining the seismic station to which the continuous waveform to be tested belongs according to the data source of the continuous waveform to be tested, and then determining the first-level directory in the template seismic waveform library according to the seismic station; Step 6.2.2, calculate the maximum amplitude of the P phase waveform in the continuous waveform to be detected, and then match the corresponding grade classification according to the maximum amplitude of the P phase waveform, so as to determine the secondary directory under the primary directory, and then determine whether there are P waves and S waves in the determined secondary directory. If there are P waves and S waves, go to step 6.2.3, otherwise go to step 6.2.4; Step 6.2.3, determine the maximum amplitude S1 of the S seismic phase waveform in the continuous waveform to be tested and the maximum amplitude S2 of the S wave in the secondary directory, and then calculate the ratio of the maximum amplitude S1 to the maximum amplitude S2. If the ratio is within the range of 0.9 to 1.1, it is determined that the current continuous waveform to be tested is matched successfully, otherwise it is determined that the current continuous waveform to be tested fails to match, and then proceed to step 6.2.6; Step 6.2.4, searching for P waves and S waves in the secondary directories of adjacent levels step by step, and after finding the P waves and S waves in the secondary directories of adjacent levels, proceed to step 6.2.5; Step 6.2.5, determine the maximum amplitude P1 of the P seismic phase waveform and the maximum amplitude S1 of the S seismic phase waveform in the continuous waveform to be tested, then determine the maximum amplitude P2 of the P wave and the maximum amplitude S2 of the S wave in the secondary directory of the adjacent level, then calculate the matching degree G as P1×S2 / P2×S1, and then determine whether the matching degree G is within the matching threshold range of 0.8~1.2. If it is within the matching threshold range, it is determined that the current continuous waveform to be tested is matched successfully, and the P seismic phase waveform and the S seismic phase waveform in the continuous waveform to be tested are stored as P waves and S waves in the secondary directory matching the P seismic phase waveform level in the template seismic waveform library, otherwise it is determined that the current continuous waveform to be tested fails to match, and then enter step 6.2.6; Step 6.2.6, output the matching calculation results.

[0016] Compared with the prior art, the present invention has the following beneficial effects: by merging the machine-detected earthquake waveform template with the manual-detected earthquake waveform template to jointly build a template earthquake waveform library, the template earthquake waveform library can have a more comprehensive database, effectively improving the success rate and reliability of earthquake detection waveform template matching; by matching and calculating the continuous waveform to be detected with the template waveform of each seismic phase of each earthquake waveform template in the template earthquake waveform library, the continuous waveform to be detected can be quickly and accurately identified and judged, thereby quickly judging whether there is an earthquake event in the continuous waveform to be detected; by intercepting and processing each earthquake waveform template in the template earthquake waveform library, the amount of data to be compared can be reduced, the comparison of invalid data can be reduced, the template matching can be made more accurate and reliable, and the template comparison efficiency can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of the method of the present invention; Figure 2 It is a schematic diagram of the removal time interval of the present invention. DETAILED DESCRIPTION

[0018] The technical solution of the present invention is described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the embodiments.

[0019] like Figure 1 As shown, the template matching earthquake detection method based on artificial intelligence disclosed in the present invention comprises the following steps: Step 1: Obtain the original continuous waveform data of historical earthquake monitoring at each seismic station, and use the trained deep learning detection model to extract the effective seismic phase data from the original continuous waveform data; Step 2, correlating the extracted effective seismic phase data to determine the effective seismic waveforms therein, and using the effective seismic waveforms as machine detection seismic waveform templates; Step 3, obtaining the identified artificial earthquake catalog, and intercepting artificial earthquake monitoring waveforms according to the earthquake events of each seismic station, and then using the intercepted artificial earthquake monitoring waveforms as templates for each artificial detection earthquake waveform; Step 4, merging the machine-detected seismic waveform template and the manually-detected seismic waveform template to jointly build a template seismic waveform library; Step 5, intercepting each seismic waveform template in the template seismic waveform library to obtain the template waveform of each seismic phase; Step 6, read the continuous waveform data to be detected, extract the continuous waveform to be detected that is suspected to contain an earthquake event from the continuous waveform data, and match the continuous waveform to be detected with the template waveform of each seismic phase of each seismic waveform template in the template seismic waveform library. If the calculation result shows that the waveform matching is successful, it is determined that an earthquake event exists in the continuous waveform data to be detected. If the calculation result shows that the waveform matching fails, it is determined that no earthquake event exists in the continuous waveform data to be detected.

[0020] By combining machine-detected earthquake waveform templates with manually-detected earthquake waveform templates to jointly build a template earthquake waveform library, the template earthquake waveform library can have a more comprehensive database, effectively improving the success rate and reliability of earthquake detection waveform template matching; by matching and calculating the continuous waveform to be detected with the template waveform of each seismic phase of each earthquake waveform template in the template earthquake waveform library, the continuous waveform to be detected can be quickly and accurately identified and judged, thereby quickly judging whether there is an earthquake event in the continuous waveform to be detected; by intercepting and processing each earthquake waveform template in the template earthquake waveform library, the amount of data to be compared can be reduced, the comparison of invalid data can be reduced, the template matching can be made more accurate and reliable, and the template comparison efficiency can be improved.

[0021] Furthermore, in step 1, the specific steps of extracting effective seismic phase data from the original continuous waveform data using the trained deep learning detection model are as follows: Step 1.1, setting the time window length and step length for segmented interception of the original continuous waveform data, and then sequentially intercepting the original continuous waveform data of each seismic station according to the time window length and step length to obtain each segmented waveform of each seismic station; Step 1.2, input each segmented waveform into the trained deep learning detection model in turn, and use the deep learning detection model to extract and calculate the probability value of the P seismic phase and the S seismic phase in each segmented waveform. The deep learning detection model technology is an existing conventional technology. The deep learning detection model can quickly output the probability value of the segmented waveform belonging to the P seismic phase or the S seismic phase, thereby realizing the rapid intelligent classification of the segmented waveform; Step 1.3, analyzing and judging the probability values ​​of the P phase and the S phase in each segmented waveform: if the probability value of the P phase and the probability value of the S phase of a segmented waveform are both greater than the segment preset threshold, the current segmented waveform is judged to be valid phase data, otherwise it is judged that there is no valid phase data in the current segmented waveform; Step 1.4, repeat step 1.3 until all segmented waveforms of each seismic station are analyzed and judged, and all valid seismic phase data of each seismic station are output respectively, and each valid seismic phase data of each seismic station is arranged in chronological order.

[0022] Furthermore, the deep learning detection model is a PhaseNet convolutional neural network deep learning model. When training the deep learning detection model: First, obtain the existing seismic data training set and seismic data test set, and directly use the existing seismic data to construct the training set and test set, so that the trained model has better reliability; Then the deep learning detection model is trained using the earthquake data training set; Then, the trained deep learning detection model is tested using the earthquake data test set, and finally the training of the deep learning detection model is completed.

[0023] Furthermore, in step 2, the specific steps of correlating the extracted effective seismic phase data to determine the effective seismic waveforms therein are: Step 2.1, obtaining the coordinates of the seismic station's seismic point corresponding to each valid seismic phase data and the acquisition time of each valid seismic phase data, and then setting the source limiting circle for limiting the possible range of the source, the extraction time interval for defining the data extraction time range, and the effective waveform interception time length. The preferred extraction time interval is 180s, and the effective waveform interception length is 120s. By setting the corresponding extraction time interval, each associated valid seismic phase data can be screened in time; Step 2.2, according to the acquisition time sequence, the effective seismic phase data of each seismic station are allocated to the extraction time intervals of the corresponding time in the sequential arrangement, and then the effective seismic phase data of each seismic station in each extraction time interval are combined into a seismic phase data group; Step 2.3, taking out a seismic phase data group in sequence according to the arrangement time sequence, and decomposing the P seismic phase wave and the S seismic phase wave in each valid seismic phase data in the current seismic phase data group; Step 2.4, determine whether there is any valid seismic phase data that has not been taken out in the seismic phase data group. If there is any valid seismic phase data that has not been taken out, take out a valid seismic phase data and go to step 2.5, otherwise go to step 2.6; Step 2.5, determine the correlation between the P phase wave and the S phase wave in the extracted valid phase data: if the S phase wave arrival time in the valid phase data of a certain seismic station is greater than the P phase wave arrival time, then determine that the current valid phase data is possible event data, and then return to step 2.4; otherwise, determine that the current valid phase data is invalid event data, delete the current valid phase data, and then return to step 2.4; Step 2.6, determine whether the number of possible event data in the current seismic phase data group is greater than the event number threshold. The event number threshold is set to 3. The purpose of setting the event number threshold is to avoid accidental errors on the one hand, and to ensure the basic requirements of subsequent source positioning on the other hand. If the number is less than the event number threshold, the current seismic phase data group is determined to be an invalid data group, and then go to step 2.10, otherwise go to step 2.7; Step 2.7, calculate the distance D between each seismic station and the earthquake source according to the arrival time of the S phase wave, the arrival time of the P phase wave, the velocity of the S phase wave and the velocity of the P phase wave in each possible event data, and then draw each earthquake source location circle with the coordinates of the earthquake point of each seismic station as the center and the corresponding distance D as the radius. After drawing the earthquake source location circle corresponding to each possible event data, mark the intersection of each earthquake source location circle as the positioning reference point; Step 2.8, select each positioning reference point as the center of the source limiting circle in turn, and then determine the number of positioning reference points in each source limiting circle respectively, and then compare the determined number of points to find the maximum number of points. If the maximum number of points is greater than the point number threshold, the point number threshold is set to 0.8 times the number of possible event data in the current seismic phase data group, so as to ensure the reliability of the source positioning, then the current seismic phase data group is determined to be a valid data group, and then go to step 2.9, otherwise the current seismic phase data group is determined to be an invalid data group, and then go to step 2.10; Step 2.9, find the source limiting circle corresponding to the maximum number of points, then determine each positioning reference point within the source limiting circle, then find the corresponding source positioning circle based on each positioning reference point, then determine each seismic station based on each source positioning circle, and finally extract the effective seismic waveform from the effective seismic phase data of each seismic station based on the effective waveform interception time length, and the effective seismic waveform includes the complete waveform of the S seismic phase wave and the P seismic phase wave; Step 2.10, determine whether there are any unretrieved seismic phase data groups, if there are any unretrieved seismic phase data groups, return to step 2.3, otherwise output the effective seismic waveforms of each seismic station.

[0024] Furthermore, in step 3, the specific steps of intercepting artificial earthquake monitoring waveforms according to earthquake events at each seismic station are: Step 3.1, determine the earthquake events of each seismic station according to the identified artificial earthquake catalog, and obtain the artificial monitoring data of each earthquake event; Step 3.2, using TAUP ray tracing technology to calculate the arrival time of the S phase wave and the P phase wave of each earthquake event at each seismic station; Step 3.3, set the starting point of the effective waveform interception interval to the first 7 seconds after the arrival of the P seismic phase wave, and then set the end point of the effective waveform interception interval to the last 15 seconds after the arrival of the P seismic phase wave, and then extract the corresponding artificial earthquake monitoring waveform from the manual monitoring data of each seismic station according to the set effective waveform interception interval. By setting the effective waveform interception interval, the data range can be limited and the effective waveform can be retained.

[0025] Furthermore, in step 4, the specific steps of merging the machine-detected seismic waveform template and the manually-detected seismic waveform template to jointly build a template seismic waveform library are: Step 4.1, collecting each machine-detected seismic waveform template and each manually-detected seismic waveform template, and obtaining the seismic stations corresponding to the machine-detected seismic waveform template and the seismic stations corresponding to the manually-detected seismic waveform template; Step 4.2, using the names of the seismic stations as the first-level directories of the template seismic waveform library, and placing the machine-detected seismic waveform templates and the manual-detected seismic waveform templates into the corresponding first-level directories; Step 4.3, extract the P seismic phase wave in the machine-detected seismic waveform template or the manually-detected seismic waveform template under each first-level directory, classify the obtained seismic waveform templates according to the amplitude of the P seismic phase wave, and use the classified levels as the second-level directory. By setting the level classification, the waveform template matching can be more accurate, and two waveforms with similar amplitudes can be matched, which not only has high matching efficiency but also high matching accuracy. Then, each machine-detected seismic waveform template and each manually-detected seismic waveform template are placed in the corresponding second-level directory respectively. If there are multiple seismic waveform templates under a certain level, delete the redundant seismic waveform templates so that only one seismic waveform template is retained for each level, and finally a template seismic waveform library is formed.

[0026] Furthermore, in step 4.3, the specific steps of classifying the obtained seismic waveform templates according to the amplitude of the P seismic phase wave are as follows: First, decompose each machine-detected seismic waveform template and each manually-detected seismic waveform template collected in step 4.1 to obtain the P seismic phase wave in each seismic waveform template; Then, the maximum amplitude value in each P seismic phase wave is screened out and defined as Tmax, and the minimum amplitude value in each P seismic phase wave is screened out and defined as Tmin; Then set 0.8Tmin as the minimum level value, set 1.6Tmax as the maximum level value, and then set 8 level values ​​between 0.8Tmin and 1.6Tmax. Use the minimum level value, maximum level value and the 8 level values ​​in between to divide the amplitude of the P phase wave into 10 level intervals. Set the amplitude value of the P phase wave to T. When 0<T≤0.8Tmin, the amplitude value is judged to be the first level. When 0.8Tmin<T≤0.8Tmin+(1.6Tmax-0 .8Tmin) / 8, the amplitude value is the second level; when 0.8Tmin+(1.6Tmax-0.8Tmin) / 8<T≤0.8Tmin+(1.6Tmax-0.8Tmin) / 4, the amplitude value is the third level; when 0.8Tmin+(1.6Tmax-0.8Tmin) / 4<T≤0.8Tmin+3(1.6Tmax-0.8Tmin) / 8, the amplitude value is the fourth level; when 0.8Tmin+3(1 .6Tmax-0.8Tmin) / 8<T≤0.8Tmin+(1.6Tmax-0.8Tmin) / 2, the amplitude value is judged as the fifth level; when 0.8Tmin+(1.6Tmax-0.8Tmin) / 2<T≤0.8Tmin+5(1.6Tmax-0.8Tmin) / 8, the amplitude value is judged as the sixth level; when 0.8Tmin+5(1.6Tmax-0.8Tmin) / 8<T≤0.8Tmin+3(1 .6Tmax-0.8Tmin) / 4, the amplitude value is judged as the seventh level; when 0.8Tmin+3(1.6Tmax-0.8Tmin) / 4<T≤0.8Tmin+7(1.6Tmax-0.8Tmin) / 8, the amplitude value is judged as the eighth level; when 0.8Tmin+7(1.6Tmax-0.8Tmin) / 8<T≤1.6Tmax, the amplitude value is judged as the ninth level; when T>1.6Tmax, the amplitude value is judged as the tenth level; Finally, the amplitude of the P seismic phase wave is determined according to which level of the 10 level intervals the amplitude of the P seismic phase wave belongs to, which is the classification level of the corresponding seismic waveform template.

[0027] Furthermore, in step 5, the specific steps of intercepting each seismic waveform template in the template seismic waveform library to obtain the template waveform of each seismic phase are: Step 5.1, obtaining a seismic waveform template in a secondary directory in a template seismic waveform library, and separating a P seismic phase wave and an S seismic phase wave from the obtained seismic waveform template; Step 5.2, respectively intercept the separated P seismic phase wave and S seismic phase wave: the waveform in the interval from 1s before to 5s after the P seismic phase wave is used as the template waveform of the P seismic phase wave, and the waveform in the interval from 1s before to 5s after the S seismic phase wave is used as the template waveform of the S seismic phase wave; Step 5.3, using the obtained P phase wave template waveform and S phase wave template waveform to replace the seismic waveform template in the corresponding secondary directory, so as to optimize the original seismic waveform template, and then determine whether there are still seismic waveform templates in the template seismic waveform library that have not been intercepted. If there are still seismic waveform templates that have not been intercepted, return to step 5.1, otherwise go to step 5.4; Step 5.4, update the template seismic waveform library: name the P seismic phase wave template waveform in each secondary directory as P wave, and name the S seismic phase wave template waveform in each secondary directory as S wave.

[0028] Furthermore, in step 6, the specific steps of extracting the continuous waveform to be detected that is suspected to contain an earthquake event from the continuous waveform data are: Step 6.1.1, extracting each data segment of the waveform in the continuous waveform data, and then performing waveform analysis on each of the extracted data segments to obtain a suspected waveform of each data segment; Step 6.1.2, using the deep learning detection model to identify the separated suspected waveforms, and determine whether each suspected waveform is a P seismic phase waveform or an S seismic phase waveform. If the P seismic phase waveform and the S seismic phase waveform are identified, proceed to step 6.1.3; otherwise, it is determined that there is no earthquake event in the continuous waveform data, and then proceed to step 6.1.6; Step 6.1.3, statistically judge the identified P seismic phase waveform and S seismic phase waveform. If there is one P seismic phase waveform and one S seismic phase waveform, proceed to step 6.1.4. If there are multiple P seismic phase waveforms and S seismic phase waveforms, proceed to step 6.1.5. Step 6.1.4, waveform matching of the S seismic phase wave and the P seismic phase wave. If the S seismic phase wave and the P seismic phase wave belong to the same earthquake source and the arrival time of the S seismic phase wave is greater than the arrival time of the P seismic phase wave, it is determined that there is a suspected earthquake event in the continuous waveform data, and then the S seismic phase wave and the P seismic phase wave are spliced ​​into a continuous waveform to be tested in time sequence according to the arrival time of the S seismic phase wave and the arrival time of the P seismic phase wave, and then the process proceeds to step 6.1.6. Otherwise, it is determined that there is no earthquake event in the continuous waveform data, and then the process proceeds to step 6.1.6. Step 6.1.5, waveform matching is performed on each P seismic phase waveform and each S seismic phase waveform, and the matched P seismic phase waveform and S seismic phase waveform are taken as a seismic matching group. If the number of seismic matching groups obtained by matching is ≥ 1, it is determined that there is a suspected earthquake event in the continuous waveform data, and then the S seismic phase waves and P seismic phase waves in each seismic matching group are spliced ​​into each continuous waveform to be tested according to the time sequence of the arrival time of the S seismic phase wave and the arrival time of the P seismic phase wave, and then the process proceeds to step 6.1.6. Otherwise, it is determined that there is no earthquake event in the continuous waveform data, and then the process proceeds to step 6.1.6; Step 6.1.6, output the judgment result: if it is judged that there is a suspected earthquake event in the continuous waveform data, then the spliced ​​continuous waveforms to be tested are output; if it is judged that there is no earthquake event in the continuous waveform data, then the result that no continuous waveform to be tested is obtained is output.

[0029] Furthermore, in step 6.1.5, the specific steps of waveform matching for each P seismic phase waveform and each S seismic phase waveform are as follows: Step a, filtering each P seismic phase waveform and each S seismic phase waveform to obtain a denoised P seismic phase waveform and an S seismic phase waveform; Step b, arranging the S seismic phase waveforms in order from small to large in time to form an S wave sequence, and arranging the P seismic phase waveforms in order from small to large in time to form a P wave sequence; Step c, extract the S seismic phase waveform with the smallest arrival time from the S wave sequence, and then find out from the P wave sequence all P seismic phase waveforms with smaller arrival times than the extracted S seismic phase waveform. If no corresponding P seismic phase waveform is found, proceed to step h; Step d, obtaining the maximum amplitude of the extracted S seismic phase waveform as Hs and the maximum amplitude of each P seismic phase waveform found as Hp, then setting the height of the interception line as 20%Hs, using the interception line with a height of 20%Hs to intercept the extracted S seismic phase waveform and each P seismic phase waveform found, obtaining the distribution time width ΔS of the S seismic phase waveform above the interception line and the distribution time width ΔP of each P seismic phase waveform, using the interception line to intercept the waveform can determine the effective duration of the waveform, the feature used is that the vibration duration of the S seismic phase waveform and the P seismic phase waveform of the same source is close, then counting the number of times Ns that the S seismic phase waveform crosses the interception line up and down within the range of the distribution time width ΔS and the number of times Np that the corresponding P seismic phase waveform crosses the interception line up and down within the range of each distribution time width ΔP, using the number of times the S seismic phase waveform crosses the interception line and combining the distribution time width to judge whether the vibration frequencies of the S seismic phase waveform and the P seismic phase waveform are close, is based on the feature that the vibration frequencies of the S seismic phase waveform and the P seismic phase waveform of the same source are close; Step e, finding a P seismic phase waveform with Hp<Hs in each P seismic phase waveform, and using the found P seismic phase waveform as the P seismic phase waveform after primary screening, if the corresponding P seismic phase waveform is not found, proceeding to step h; Step f, finding each P seismic phase waveform with 0.85<Np / Ns<1.15 in the P seismic phase waveform after the primary screening, and using the found P seismic phase waveform as the P seismic phase waveform after the secondary screening, if the corresponding P seismic phase waveform is not found, proceeding to step h; Step g, finding the distribution time width ΔP with the smallest difference with the distribution time width ΔS in the P seismic phase waveform after the secondary screening, and determining the P seismic phase waveform corresponding to the found distribution time width ΔP as matching the current S seismic phase waveform, and deleting the found P seismic phase waveform from the P wave sequence, and then entering step h; Step h, determining whether there is an S seismic phase waveform in the S wave sequence, if there is an S seismic phase waveform, returning to step c, otherwise proceeding to step i; Step i: output matching P phase waveforms and S phase waveforms in pairs.

[0030] Furthermore, in step 6, the specific steps of matching and calculating the continuous waveform to be detected with the template waveform of each seismic phase of each seismic waveform template in the template seismic waveform library are as follows: Step 6.2.1, determining the seismic station to which the continuous waveform to be tested belongs according to the data source of the continuous waveform to be tested, and then determining the first-level directory in the template seismic waveform library according to the seismic station; Step 6.2.2, calculate the maximum amplitude of the P phase waveform in the continuous waveform to be detected, and then match the corresponding grade classification according to the maximum amplitude of the P phase waveform, so as to determine the secondary directory under the primary directory, and then determine whether there are P waves and S waves in the determined secondary directory. If there are P waves and S waves, go to step 6.2.3, otherwise go to step 6.2.4; Step 6.2.3, determine the maximum amplitude S1 of the S seismic phase waveform in the continuous waveform to be tested and the maximum amplitude S2 of the S wave in the secondary directory, and then calculate the ratio of the maximum amplitude S1 to the maximum amplitude S2. If the ratio is within the range of 0.9 to 1.1, it is determined that the current continuous waveform to be tested is matched successfully, otherwise it is determined that the current continuous waveform to be tested fails to match, and then proceed to step 6.2.6; Step 6.2.4, search for P wave and S wave in the secondary directory of the adjacent level step by step, and enter step 6.2.5 after finding P wave and S wave in the secondary directory of the adjacent level. By searching for P wave and S wave in the secondary directory of the adjacent level, it can be ensured that each input continuous waveform to be tested can complete the template matching; Step 6.2.5, determine the maximum amplitude P1 of the P seismic phase waveform and the maximum amplitude S1 of the S seismic phase waveform in the continuous waveform to be tested, then determine the maximum amplitude P2 of the P wave and the maximum amplitude S2 of the S wave in the secondary directory of the adjacent level, then calculate the matching degree G as P1×S2 / P2×S1, and then judge whether the matching degree G is within the matching threshold range of 0.8~1.2. By setting a relatively large range, the errors found due to different levels can be enhanced, and the inclusiveness of template matching can be enhanced. If it is within the matching threshold range, it is determined that the current continuous waveform to be tested is matched successfully, and the P seismic phase waveform and the S seismic phase waveform in the continuous waveform to be tested are stored as P waves and S waves in the secondary directory matching the P seismic phase waveform level in the template seismic waveform library, otherwise it is determined that the current continuous waveform to be tested fails to match, and then enter step 6.2.6; Step 6.2.6, output the matching calculation results.

[0031] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes in form and details may be made without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A template matching earthquake detection method based on artificial intelligence, characterized in that: The following steps are involved: Step 1: Obtain the original continuous waveform data of historical earthquake monitoring at each seismic station, and use the trained deep learning detection model to extract the effective seismic phase data from the original continuous waveform data; Step 2, correlating the extracted effective seismic phase data to determine the effective seismic waveforms therein, and using the effective seismic waveforms as machine detection seismic waveform templates; Step 3, obtaining the identified artificial earthquake catalog, and intercepting artificial earthquake monitoring waveforms according to the earthquake events of each seismic station, and then using the intercepted artificial earthquake monitoring waveforms as templates for each artificial detection earthquake waveform; Step 4, merging the machine-detected seismic waveform template and the manually-detected seismic waveform template to jointly build a template seismic waveform library; Step 5, intercepting each seismic waveform template in the template seismic waveform library to obtain the template waveform of each seismic phase; Step 6, read the continuous waveform data to be detected, extract the continuous waveform to be detected that is suspected to contain an earthquake event from the continuous waveform data, and match the continuous waveform to be detected with the template waveform of each seismic phase of each seismic waveform template in the template seismic waveform library. If the calculation result shows that the waveform matching is successful, it is determined that an earthquake event exists in the continuous waveform data to be detected. If the calculation result shows that the waveform matching fails, it is determined that no earthquake event exists in the continuous waveform data to be detected.

2. The template matching earthquake detection method based on artificial intelligence according to claim 1 is characterized in that: In step 1, the specific steps of using the trained deep learning detection model to extract effective seismic phase data from the original continuous waveform data are as follows: Step 1.1, setting the time window length and step length for segmented interception of the original continuous waveform data, and then sequentially intercepting the original continuous waveform data of each seismic station according to the time window length and step length to obtain each segmented waveform of each seismic station; Step 1.2, input each segmented waveform into the trained deep learning detection model in sequence, and use the deep learning detection model to extract and calculate the probability values ​​of the P phase and the S phase in each segmented waveform; Step 1.3, analyzing and judging the probability values ​​of the P phase and the S phase in each segmented waveform: if the probability value of the P phase and the probability value of the S phase of a segmented waveform are both greater than the segment preset threshold, the current segmented waveform is judged to be valid phase data, otherwise it is judged that there is no valid phase data in the current segmented waveform; Step 1.4, repeat step 1.3 until all segmented waveforms of each seismic station are analyzed and judged, and all valid seismic phase data of each seismic station are output respectively, and each valid seismic phase data of each seismic station is arranged in chronological order.

3. The template matching earthquake detection method based on artificial intelligence according to claim 2 is characterized in that: In step 2, the specific steps of correlating the extracted effective seismic phase data to determine the effective seismic waveforms therein are: Step 2.1, obtaining the coordinates of the seismic station points corresponding to each effective seismic phase data and the acquisition time of each effective seismic phase data, and then setting the source limiting circle for limiting the possible range of the source, the extraction time interval for defining the data extraction time range, and the effective waveform interception time length; Step 2.2, according to the acquisition time sequence, the effective seismic phase data of each seismic station are allocated to the extraction time intervals of the corresponding time in the sequential arrangement, and then the effective seismic phase data of each seismic station in each extraction time interval are combined into a seismic phase data group; Step 2.3, taking out a seismic phase data group in sequence according to the arrangement time sequence, and decomposing the P seismic phase wave and the S seismic phase wave in each valid seismic phase data in the current seismic phase data group; Step 2.4, determine whether there is any valid seismic phase data that has not been taken out in the seismic phase data group. If there is any valid seismic phase data that has not been taken out, take out a valid seismic phase data and go to step 2.5, otherwise go to step 2.6; Step 2.5, determine the correlation between the P phase wave and the S phase wave in the extracted valid phase data: if the S phase wave arrival time in the valid phase data of a certain seismic station is greater than the P phase wave arrival time, then determine that the current valid phase data is possible event data, and then return to step 2.4; otherwise, determine that the current valid phase data is invalid event data, delete the current valid phase data, and then return to step 2.4; Step 2.6, determine whether the number of possible event data in the current seismic phase data group is greater than the event number threshold. If the number is less than the event number threshold, the current seismic phase data group is determined to be an invalid data group, and then proceed to step 2.10, otherwise proceed to step 2.7; Step 2.7, calculate the distance D between each seismic station and the earthquake source according to the arrival time of the S phase wave, the arrival time of the P phase wave, the velocity of the S phase wave and the velocity of the P phase wave in each possible event data, and then draw each earthquake source location circle with the coordinates of the earthquake point of each seismic station as the center and the corresponding distance D as the radius. After drawing the earthquake source location circle corresponding to each possible event data, mark the intersection of each earthquake source location circle as the positioning reference point; Step 2.8, select each positioning reference point as the center of the source limiting circle in turn, and then determine the number of positioning reference points in each source limiting circle respectively, and then compare the determined number of points to find the maximum number of points. If the maximum number of points is greater than the point number threshold, the current seismic phase data group is determined to be a valid data group, and then proceed to step 2.9; otherwise, the current seismic phase data group is determined to be an invalid data group, and then proceed to step 2.10; Step 2.9, find the source limiting circle corresponding to the maximum number of points, then determine each positioning reference point within the source limiting circle, then find the corresponding source positioning circle based on each positioning reference point, then determine each seismic station based on each source positioning circle, and finally extract the effective seismic waveform from the effective seismic phase data of each seismic station based on the effective waveform interception time length, and the effective seismic waveform includes the complete waveform of the S seismic phase wave and the P seismic phase wave; Step 2.10, determine whether there are any unretrieved seismic phase data groups, if there are any unretrieved seismic phase data groups, return to step 2.3, otherwise output the effective seismic waveforms of each seismic station.

4. The template matching earthquake detection method based on artificial intelligence according to claim 1 is characterized in that: In step 3, the specific steps of intercepting artificial earthquake monitoring waveforms according to earthquake events at each seismic station are: Step 3.1, determine the earthquake events of each seismic station according to the identified artificial earthquake catalog, and obtain the artificial monitoring data of each earthquake event; Step 3.2, using TAUP ray tracing technology to calculate the arrival time of the S phase wave and the P phase wave of each earthquake event at each seismic station; Step 3.3, set the starting point of the effective waveform interception interval to the 7 seconds before the arrival of the P seismic phase wave, and then set the end point of the effective waveform interception interval to the 15 seconds after the arrival of the P seismic phase wave, and then extract the corresponding artificial earthquake monitoring waveform from the artificial monitoring data of each seismic station according to the set effective waveform interception interval.

5. The template matching earthquake detection method based on artificial intelligence according to claim 1 is characterized in that: In step 4, the specific steps of merging the machine-detected seismic waveform template and the manually-detected seismic waveform template to jointly build a template seismic waveform library are: Step 4.1, collecting each machine-detected seismic waveform template and each manually-detected seismic waveform template, and obtaining the seismic stations corresponding to the machine-detected seismic waveform template and the seismic stations corresponding to the manually-detected seismic waveform template; Step 4.2, using the names of the seismic stations as the first-level directories of the template seismic waveform library, and placing the machine-detected seismic waveform templates and the manual-detected seismic waveform templates into the corresponding first-level directories; Step 4.3, extract the P seismic phase wave in the machine-detected seismic waveform template or the manually-detected seismic waveform template under each first-level directory, classify the obtained seismic waveform templates according to the amplitude of the P seismic phase wave, and use the classified levels as the second-level directory, and then put each machine-detected seismic waveform template and each manually-detected seismic waveform template into the corresponding second-level directory respectively. If there are multiple seismic waveform templates under a certain level, delete the redundant seismic waveform templates so that only one seismic waveform template is retained in each level, and finally a template seismic waveform library is formed.

6. The template matching earthquake detection method based on artificial intelligence according to claim 5 is characterized in that: In step 4.3, the specific steps of classifying the obtained seismic waveform templates according to the amplitude of the P seismic phase wave are as follows: First, decompose each machine-detected seismic waveform template and each manually-detected seismic waveform template collected in step 4.1 to obtain the P seismic phase wave in each seismic waveform template; Then, the maximum amplitude value in each P seismic phase wave is screened out and defined as Tmax, and the minimum amplitude value in each P seismic phase wave is screened out and defined as Tmin; Then, 0.8Tmin is set as the minimum level value, 1.6Tmax is set as the maximum level value, and 8 level values ​​are set equally between 0.8Tmin and 1.6Tmax. The amplitude of the P phase wave is divided into 10 level intervals using the minimum level value, the maximum level value, and the 8 level values ​​in between. Finally, the amplitude of the P seismic phase wave is determined according to which level of the 10 level intervals the amplitude of the P seismic phase wave belongs to, which is the classification level of the corresponding seismic waveform template.

7. The template matching earthquake detection method based on artificial intelligence according to claim 5 is characterized in that: In step 5, the specific steps of intercepting each seismic waveform template in the template seismic waveform library to obtain the template waveform of each seismic phase are: Step 5.1, obtaining a seismic waveform template in a secondary directory in a template seismic waveform library, and separating a P seismic phase wave and an S seismic phase wave from the obtained seismic waveform template; Step 5.2, respectively intercept the separated P seismic phase wave and S seismic phase wave: the waveform in the interval from 1s before to 5s after the P seismic phase wave is used as the template waveform of the P seismic phase wave, and the waveform in the interval from 1s before to 5s after the S seismic phase wave is used as the template waveform of the S seismic phase wave; Step 5.3, using the obtained P phase wave template waveform and S phase wave template waveform to replace the seismic waveform template in the corresponding secondary directory, and then judging whether there are still seismic waveform templates in the template seismic waveform library that have not been intercepted, if there are still seismic waveform templates that have not been intercepted, then return to step 5.1, otherwise go to step 5.4; Step 5.4, update the template seismic waveform library: name the P seismic phase wave template waveform in each secondary directory as P wave, and name the S seismic phase wave template waveform in each secondary directory as S wave.

8. The template matching earthquake detection method based on artificial intelligence according to claim 1 is characterized in that: In step 6, the specific steps of extracting the continuous waveform to be detected that is suspected to contain an earthquake event from the continuous waveform data are: Step 6.1.1, extracting each data segment of the waveform in the continuous waveform data, and then performing waveform analysis on each of the extracted data segments to obtain a suspected waveform of each data segment; Step 6.1.2, using the deep learning detection model to identify the separated suspected waveforms, and determine whether each suspected waveform is a P seismic phase waveform or an S seismic phase waveform. If the P seismic phase waveform and the S seismic phase waveform are identified, proceed to step 6.1.3; otherwise, it is determined that there is no earthquake event in the continuous waveform data, and then proceed to step 6.1.6; Step 6.1.3, statistically judge the identified P seismic phase waveform and S seismic phase waveform. If there is one P seismic phase waveform and one S seismic phase waveform, proceed to step 6.1.

4. If there are multiple P seismic phase waveforms and S seismic phase waveforms, proceed to step 6.1.

5. Step 6.1.4, waveform matching of the S seismic phase wave and the P seismic phase wave. If the S seismic phase wave and the P seismic phase wave belong to the same earthquake source and the arrival time of the S seismic phase wave is greater than the arrival time of the P seismic phase wave, it is determined that there is a suspected earthquake event in the continuous waveform data, and then the S seismic phase wave and the P seismic phase wave are spliced ​​into a continuous waveform to be tested in time sequence according to the arrival time of the S seismic phase wave and the arrival time of the P seismic phase wave, and then the process proceeds to step 6.1.

6. Otherwise, it is determined that there is no earthquake event in the continuous waveform data, and then the process proceeds to step 6.1.

6. Step 6.1.5, waveform matching is performed on each P seismic phase waveform and each S seismic phase waveform, and the matched P seismic phase waveform and S seismic phase waveform are taken as a seismic matching group. If the number of seismic matching groups obtained by matching is ≥ 1, it is determined that there is a suspected earthquake event in the continuous waveform data, and then the S seismic phase waves and P seismic phase waves in each seismic matching group are spliced ​​into each continuous waveform to be tested according to the time sequence of the arrival time of the S seismic phase wave and the arrival time of the P seismic phase wave, and then the process proceeds to step 6.1.

6. Otherwise, it is determined that there is no earthquake event in the continuous waveform data, and then the process proceeds to step 6.1.6; Step 6.1.6, output the judgment result: if it is judged that there is a suspected earthquake event in the continuous waveform data, then the spliced ​​continuous waveforms to be tested are output; if it is judged that there is no earthquake event in the continuous waveform data, then the result that no continuous waveform to be tested is obtained is output.

9. The template matching earthquake detection method based on artificial intelligence according to claim 8 is characterized in that: In step 6.1.5, the specific steps for waveform matching of each P seismic phase waveform and each S seismic phase waveform are as follows: Step a, filtering each P seismic phase waveform and each S seismic phase waveform to obtain a denoised P seismic phase waveform and an S seismic phase waveform; Step b, arranging the S seismic phase waveforms in order from small to large in time to form an S wave sequence, and arranging the P seismic phase waveforms in order from small to large in time to form a P wave sequence; Step c, extract the S seismic phase waveform with the smallest arrival time from the S wave sequence, and then find out from the P wave sequence all P seismic phase waveforms with smaller arrival times than the extracted S seismic phase waveform. If no corresponding P seismic phase waveform is found, proceed to step h; Step d, obtaining the maximum amplitude of the extracted S seismic phase waveform as Hs and the maximum amplitude of each P seismic phase waveform found as Hp, then setting the height of the interception line as 20% Hs, using the interception line with a height of 20% Hs to intercept the extracted S seismic phase waveform and each P seismic phase waveform found, obtaining the distribution time width ΔS of the S seismic phase waveform above the interception line and the distribution time width ΔP of each P seismic phase waveform, and then counting the number of times Ns that the S seismic phase waveform crosses the interception line up and down within the range of the distribution time width ΔS and the number of times Np that the corresponding P seismic phase waveform crosses the interception line up and down within the range of each distribution time width ΔP; Step e, finding a P seismic phase waveform with Hp<Hs in each P seismic phase waveform, and using the found P seismic phase waveform as the P seismic phase waveform after primary screening, if the corresponding P seismic phase waveform is not found, proceeding to step h; Step f, finding each P seismic phase waveform with 0.85<Np / Ns<1.15 in the P seismic phase waveform after the primary screening, and using the found P seismic phase waveform as the P seismic phase waveform after the secondary screening, if the corresponding P seismic phase waveform is not found, proceeding to step h; Step g, finding the distribution time width ΔP with the smallest difference with the distribution time width ΔS in the P seismic phase waveform after the secondary screening, and determining the P seismic phase waveform corresponding to the found distribution time width ΔP as matching the current S seismic phase waveform, and deleting the found P seismic phase waveform from the P wave sequence, and then entering step h; Step h, determining whether there is an S seismic phase waveform in the S wave sequence, if there is an S seismic phase waveform, returning to step c, otherwise proceeding to step i; Step i: output matching P phase waveforms and S phase waveforms in pairs.

10. The template matching earthquake detection method based on artificial intelligence according to claim 7, characterized in that: In step 6, the specific steps of matching and calculating the continuous waveform to be detected with the template waveform of each seismic phase of each seismic waveform template in the template seismic waveform library are as follows: Step 6.2.1, determining the seismic station to which the continuous waveform to be tested belongs according to the data source of the continuous waveform to be tested, and then determining the first-level directory in the template seismic waveform library according to the seismic station; Step 6.2.2, calculate the maximum amplitude of the P phase waveform in the continuous waveform to be detected, and then match the corresponding grade classification according to the maximum amplitude of the P phase waveform, so as to determine the secondary directory under the primary directory, and then determine whether there are P waves and S waves in the determined secondary directory. If there are P waves and S waves, go to step 6.2.3, otherwise go to step 6.2.4; Step 6.2.3, determine the maximum amplitude S1 of the S seismic phase waveform in the continuous waveform to be tested and the maximum amplitude S2 of the S wave in the secondary directory, and then calculate the ratio of the maximum amplitude S1 to the maximum amplitude S2. If the ratio is within the range of 0.9 to 1.1, it is determined that the current continuous waveform to be tested is matched successfully, otherwise it is determined that the current continuous waveform to be tested fails to match, and then proceed to step 6.2.6; Step 6.2.4, searching for P waves and S waves in the secondary directories of adjacent levels step by step, and after finding the P waves and S waves in the secondary directories of adjacent levels, proceed to step 6.2.5; Step 6.2.5, determine the maximum amplitude P1 of the P seismic phase waveform and the maximum amplitude S1 of the S seismic phase waveform in the continuous waveform to be tested, then determine the maximum amplitude P2 of the P wave and the maximum amplitude S2 of the S wave in the secondary directory of the adjacent level, then calculate the matching degree G as P1×S2 / P2×S1, and then determine whether the matching degree G is within the matching threshold range of 0.8~1.

2. If it is within the matching threshold range, it is determined that the current continuous waveform to be tested is matched successfully, and the P seismic phase waveform and the S seismic phase waveform in the continuous waveform to be tested are stored as P waves and S waves in the secondary directory matching the P seismic phase waveform level in the template seismic waveform library, otherwise it is determined that the current continuous waveform to be tested fails to match, and then enter step 6.2.6; Step 6.2.6, output the matching calculation results.

Images