Earthquake epicenter positioning method based on distributed acoustic sensing
By preprocessing and network processing the DAS data from the distributed acoustic sensor, the time difference of arrival of the P-wave and S-wave of the seismic signal is extracted. Combined with the azimuth and epicenter distance, the problem of epicenter location in far-field scenes and on straight-line optical fibers is solved, improving the positioning accuracy and real-time performance.
Patent Information
- Application Number
- CN202411058168.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-02
- Publication Date
- 2024-11-22
AI Technical Summary
In existing technologies, the epicenter localization algorithm of distributed acoustic sensors is not applicable in far-field scenarios and on straight-line optical fibers, and the manual extraction of earthquake signal arrival information leads to poor system real-time performance and large filtering errors.
An earthquake epicenter location method based on distributed acoustic sensing is adopted. The method involves acquiring DAS data for preprocessing, using an earthquake event segmentation network and an identification network to extract the arrival time difference of P-waves and S-waves of the earthquake signal, and combining the azimuth and epicenter distance for location.
It achieves accurate positioning in far-field scenarios and on straight-line optical fibers, improving the system's real-time performance and positioning accuracy, and reducing errors.
Smart Images

Figure CN119001842A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of earthquake monitoring, and more specifically, to a method for locating an earthquake epicenter based on distributed acoustic sensing. Background Art
[0002] The epicenter location refers to the vertical projection of the earthquake source on the horizontal surface of the earth, expressed in longitude and latitude. It is an important parameter information for earthquake reporting and early warning. At present, the main monitoring device for epicenter positioning is to use a seismograph station network to estimate the direction through the arrival time difference of seismic wave signals received at different stations. Fiber-optic distributed acoustic sensing DAS is a sensor array with a wide detection range. Due to the different transmission paths of seismic waves to each virtual sensor element on the optical fiber, the arrival time of the detected seismic signal is different. Referring to the positioning method of the seismograph station, the arrival time difference between different virtual sensors on the optical fiber can be used to estimate the position. However, the epicenter positioning algorithms currently used in DAS are only applicable to near-field scenarios, that is, the location of the earthquake is close to the optical fiber, and the trend of the optical fiber is a broken line. It is not applicable to far-field scenarios and optical fibers with a straight trend. In addition, the arrival time information of the seismic signal must be manually extracted, which is not conducive to the real-time performance of the system, and it is believed that the screening error is large. Summary of the invention
[0003] The present invention provides an earthquake epicenter locating method based on distributed acoustic sensing, and provides an epicenter locating method for optical fibers applicable to far-field scenes and straight-line trends.
[0004] In order to solve the above technical problems, the technical solution of the present invention is as follows:
[0005] A method for locating an earthquake epicenter based on distributed acoustic sensing comprises the following steps:
[0006] Acquiring DAS data based on distributed acoustic sensing, and preprocessing the DAS data;
[0007] The preprocessed DAS data is processed using a preset earthquake event segmentation network to obtain segmented earthquake data;
[0008] Based on the segmented seismic data, an azimuth of the earthquake is estimated;
[0009] The pre-processed DAS data is processed using a preset earthquake identification network to obtain the arrival time difference of the P wave and S wave of the earthquake event;
[0010] Calculating the epicenter distance of the earthquake based on the arrival time difference of the P wave and the S wave of the earthquake event;
[0011] The epicenter location of the earthquake is obtained according to the azimuth of the earthquake and the distance from the epicenter of the earthquake.
[0012] In the above-mentioned technical means, the earthquake event segmentation network is used to extract the outline of the earthquake data from the data, and the epicenter azimuth when the theoretical value and the measured value are closest is obtained based on the segmented earthquake data. Then, the earthquake recognition network is used to obtain the arrival time difference of the P wave and the S wave of the earthquake signal to estimate the epicenter distance of the earthquake epicenter. Finally, the azimuth and epicenter distance information are fused to obtain the final estimated longitude and latitude position of the earthquake epicenter.
[0013] Further, the DAS data is preprocessed, including:
[0014] Bandpass filtering the DAS data through a 3-20 Hz bandpass filter;
[0015] The bandpass filtered DAS data were normalized to zero mean;
[0016] The zero-mean normalized DAS data is downsampled in the time domain to obtain the preprocessed DAS data.
[0017] Furthermore, the preset earthquake identification network sequentially sets four first encoders, two two-dimensional convolutional layers, four first decoders and one two-dimensional convolutional layer from input to output, wherein:
[0018] The input is the pre-processed DAS data;
[0019] There is also a jump connection between the nth first encoder and the 5th to nth first decoders, n = 1, 2, 3, 4;
[0020] After the output of the last two-dimensional convolutional layer is combined with the input, the output of the preset earthquake identification network is obtained.
[0021] Furthermore, each of the first encoders includes a BN layer, a residual convolution module and a pooling layer connected in sequence.
[0022] Furthermore, each of the first decoders includes an upsampling layer, a connection layer, a BN layer and a residual convolution module connected in sequence.
[0023] Furthermore, the pre-processed DAS data is processed using a preset seismic event segmentation network to obtain segmented seismic data, including:
[0024] The preprocessed DAS data is input into the preset earthquake event segmentation network, wherein the size of the preprocessed DAS data is (H, W), and the preset earthquake event segmentation network outputs a data of size (H, W, 2), including 2 arrays Mask of size (H, W)seismic and Mask noise , Mask seismic and Mask noise They are the seismic signal weight matrix and the noise signal weight matrix, Mask seismic (i, j) indicates that the i-th row and j-th column of the preprocessed DAS data is the weight of the seismic signal. Similarly, Mask noise (i, j) indicates that the data at the i-th row and j-th column is the weight of the noise signal. seismic (i, j) is greater than Mask noise (i, j), the i-th row and j-th column of the preprocessed DAS data is a seismic signal, which is set to 1. seismic (i, j) is less than Mask noise (i, j), the position of the i-th row and j-th column of the preprocessed DAS data is a noise signal, which is set to 0 to obtain the segmented seismic data.
[0025] Further, based on the segmented seismic data, an azimuth of the earthquake is estimated, including:
[0026] Traversing the segmented seismic data by channel, recording the subscript index of the data that is first 1 in each channel as the wavefront of the seismic signal recorded in the current channel, and finally obtaining the seismic wavefront recorded along each virtual sensor element;
[0027] Performing straight line fitting on the seismic wavefront to obtain a fitted seismic wavefront;
[0028] According to the fitted seismic wavefront, a grid search is used to traverse the solution space, the weight value of each point is calculated, and the azimuth of the earthquake is obtained.
[0029] Furthermore, the preset earthquake identification network is provided with an encoding module, a decoding module and a projection module in sequence from input to output, wherein:
[0030] The encoding module includes six second encoders connected in sequence, each of which includes a BN layer, two two-dimensional convolutional layers, a maximum pooling layer and a dimension matching layer, wherein the output end of the BN layer is respectively connected to the input end of a two-dimensional convolutional layer and the input end of the dimension matching layer, the two two-dimensional convolutional layers are connected, and the output end of the last two-dimensional convolutional layer is merged with the output end of the dimension matching layer and then connected to the input end of the maximum pooling layer;
[0031] The decoding module includes six second decoders connected in sequence, each of which includes an upsampling layer, a BN layer, two one-dimensional convolutional layers and a dimension matching layer, wherein the upsampling layer and the BN layer are connected in sequence, the output end of the BN layer is connected to the input end of a one-dimensional convolutional layer and the input end of the dimension matching layer respectively, the two one-dimensional convolutional layers are connected, and the output end of the last one-dimensional convolutional layer is combined with the output end of the dimension matching layer as the output of the second decoder;
[0032] The projection module includes a residual structure consisting of three one-dimensional convolutional layers and one dimension matching layer.
[0033] Further, calculating the epicenter distance of the earthquake according to the arrival time difference between the P wave and the S wave of the earthquake event includes:
[0034]
[0035] Where Δ is the distance from the epicenter of the earthquake, Δt is the arrival time difference between the P wave and the S wave, and v P is the velocity of P wave in solid medium, v S is the velocity of S waves in solid media.
[0036] Further, obtaining the earthquake epicenter location according to the azimuth of the earthquake and the distance from the epicenter of the earthquake includes:
[0037] The azimuth of the earthquake and the distance from the epicenter of the earthquake are combined to obtain the coordinates of the location where the earthquake occurred through judgment;
[0038] The location coordinates of the earthquake are converted into longitude and latitude coordinates to obtain the epicenter of the earthquake.
[0039] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:
[0040] The present invention realizes the positioning of the epicenter of an earthquake and is not only applicable to near-field environments, but also to far-field scenarios and situations of optical fibers with a straight line. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 A schematic diagram of a flow chart of a method for locating an earthquake epicenter based on distributed acoustic sensing provided by an embodiment of the present invention;
[0042] Figure 2 A schematic diagram of the principle of an earthquake epicenter location method based on distributed acoustic sensing provided by an embodiment of the present invention;
[0043] Figure 3 A schematic diagram of a process for preprocessing DAS data provided by an embodiment of the present invention;
[0044] Figure 4A schematic diagram of the results of preprocessing the DAS data provided by an embodiment of the present invention;
[0045] Figure 5 A schematic diagram of a process for segmenting seismic data provided by an embodiment of the present invention;
[0046] Figure 6 A schematic diagram of the structure of an earthquake event segmentation network provided by an embodiment of the present invention;
[0047] Figure 7 A schematic diagram of a seismic wavefront provided by an embodiment of the present invention;
[0048] Figure 8 A schematic diagram of a seismic wavefront after fitting provided by an embodiment of the present invention;
[0049] Fig. 9 A schematic diagram of theoretical wavefront calculation provided by an embodiment of the present invention;
[0050] Fig.10 A schematic diagram of weight calculation results provided by an embodiment of the present invention;
[0051] Fig.11 A schematic diagram of data for training an earthquake identification network provided by an embodiment of the present invention;
[0052] Fig.12 A schematic diagram of the structure of an earthquake identification network provided by an embodiment of the present invention;
[0053] Fig.13 An example diagram of the identification results of the earthquake identification network provided by the embodiment of the present invention;
[0054] Fig.14 This is an example diagram of the location estimation results of an earthquake event provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0055] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;
[0056] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;
[0057] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.
[0058] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0059] Example 1
[0060] This embodiment provides a method for locating the epicenter of an earthquake based on distributed acoustic sensing. Figure 1 and Figure 2 As shown, the following steps are included:
[0061] Acquiring DAS data based on distributed acoustic sensing, and preprocessing the DAS data;
[0062] The preprocessed DAS data is processed using a preset earthquake event segmentation network to obtain segmented earthquake data;
[0063] Based on the segmented seismic data, an azimuth of the earthquake is estimated;
[0064] The pre-processed DAS data is processed using a preset earthquake identification network to obtain the arrival time difference of the P wave and S wave of the earthquake event;
[0065] Calculating the epicenter distance of the earthquake based on the arrival time difference of the P wave and the S wave of the earthquake event;
[0066] The epicenter location of the earthquake is obtained according to the azimuth of the earthquake and the distance from the epicenter of the earthquake.
[0067] In a further embodiment, the DAS data is preprocessed, such as Figure 3 As shown, the original, includes:
[0068] DSA data such as Figure 4 As shown in (a), the original signal collected by DAS generally has a lot of noise, and it is difficult to see the seismic signal. The seismic signal is generally in the frequency band of 3-20 Hz. The seismic signal can be extracted first by a 3-20 Hz bandpass filter. Therefore, the DAS data is bandpass filtered by a 3-20 Hz bandpass filter, as shown in FIG. Figure 4 (b)
[0069] The DAS data after bandpass filtering is normalized to zero mean, such as Figure 4 (c) is used to enhance the signal components and facilitate subsequent input network processing;
[0070] The zero-mean normalized DAS data is downsampled in the time domain to reduce the data volume and the subsequent processing time, thus obtaining the preprocessed DAS data.
[0071] In a further embodiment, the preset earthquake event segmentation network adopts U-net, which is a typical encoding-decoding model. From input to output, four first encoders, two two-dimensional convolutional layers, four first decoders and one two-dimensional convolutional layer are sequentially arranged, such as Figure 6 As shown, where:
[0072] The input is the pre-processed DAS data;
[0073] There is also a jump connection between the nth first encoder and the 5th to nth first decoders, n = 1, 2, 3, 4;
[0074] After the output of the last two-dimensional convolutional layer is combined with the input, the output of the preset earthquake event segmentation network is obtained.
[0075] Each of the first encoders includes a BN layer, a residual convolution module and a pooling layer connected in sequence.
[0076] Each of the first decoders includes an upsampling layer, a connection layer, a BN layer and a residual convolution module connected in sequence.
[0077] In a specific embodiment, each layer of the earthquake event segmentation network is described as follows:
[0078] (1) Batch Normalization (BN): The function of the BN layer in the neural network is to change the mean of the data to 0 and the variance to 1. This ensures that the distribution of the input data is relatively stable, avoids the problem of overfitting of the network due to differences in the characteristic distribution of the data, and speeds up the training of the network model.
[0079] (2) Residual Convolutional Module: The residual structure is widely used in the field of deep learning
[61] . It superimposes the input and output through a skip connection, which effectively solves the problems of gradient vanishing and model degradation in deep network training.
[0080] (3) Pooling: The pooling strategy used in this embodiment is Max-Pooling, and the window size is (2,2). That is, each time the data passes through the pooling layer, the length and width of the data will be halved. The pooling layer can reduce the dimension of the data, remove redundant information in the data, and extract the main features of the data.
[0081] (4) Encoder (Encoder-N): The encoder consists of a BN layer, a residual convolution module, and a pooling layer. The number N after the encoder represents the number of output feature channels of the residual convolution module. For example, 8 in Encoder-8 means that the feature dimension of the output after the module is 8. Due to the existence of the pooling layer, the length (H) and width (W) of the data processed by the encoder will become half of the original size. The specific parameter configuration of each part of the encoder is shown in Table 1 below;
[0082] Table 1
[0083]
[0084] (5) Skip Connection: There is a skip connection between the encoding module and the decoding module of U-net. As a "fast channel" in the model architecture, it can fuse the low-level features obtained in the encoding stage with the high-level features obtained in the decoding stage, thereby enhancing the network's ability to understand the data;
[0085] (6) Up-Sampling layer: The window size of the up-sampling layer in this embodiment is the same as that of the pooling layer, which is also (2,2). During the encoding stage, the data resolution is reduced and the detail information is lost due to multiple pooling. The data is restored to the data size before pooling by up-sampling, and the original resolution is restored and the detail information is reconstructed;
[0086] (7) Batch Normalization (BN): The function of the BN layer in the neural network is to change the mean of the data to 0 and the variance to 1. This ensures that the distribution of the input data is relatively stable, avoids the problem of overfitting of the network due to differences in the characteristic distribution of the data, and speeds up the training of the network model.
[0087] (8) Decoder (Decoder-N): The decoder consists of an upsampling layer, a connection layer, a BN layer, and a residual convolution module. The connection layer concatenates the information of the same feature dimension in the encoding stage and the decoding stage to achieve the fusion of self-encoding features, so that the decoder can use low-level and high-level feature information at the same time, which helps to improve the model's ability to capture semantic information, thereby improving the performance of image processing tasks. The number N after the decoder represents the number of output feature channels of the residual convolution module. For example, 8 in Decoder-8 means that the feature dimension of the output of this module is 8. Due to the existence of the upsampling layer, the length (H) and width (W) of the data processed by the decoder will become twice the original size. The specific parameter configuration of each part of the decoder is shown in Table 2 below.
[0088] Table 2
[0089]
[0090] In a further embodiment, the pre-processed DAS data is processed using a preset seismic event segmentation network to obtain segmented seismic data, such as Figure 5 As shown, including:
[0091] The preprocessed DAS data is input into the preset earthquake event segmentation network, wherein the size of the preprocessed DAS data is (H, W), and the preset earthquake event segmentation network outputs a data of size (H, W, 2), including 2 arrays Mask of size (H, W) seismic and Masknoise , Mask seismic and Mask noise They are the seismic signal weight matrix and the noise signal weight matrix, Mask seismic (i, j) indicates that the i-th row and j-th column of the preprocessed DAS data is the weight of the seismic signal. Similarly, Mask noise (i, j) indicates that the data at the i-th row and j-th column is the weight of the noise signal. seismic (i, j) is greater than Mask noise (i, j), the i-th row and j-th column of the preprocessed DAS data is a seismic signal, which is set to 1. seismic (i, j) is less than Mask noise (i, j), the position of the i-th row and j-th column of the preprocessed DAS data is a noise signal, which is set to 0 to obtain the segmented seismic data.
[0092] In a further embodiment, based on the segmented seismic data, an estimation is performed to obtain the azimuth of the earthquake, including wavefront extraction, straight line fitting and azimuth estimation, wherein:
[0093] The basic idea of wavefront extraction is that the segmented data is a two-dimensional matrix containing only the numbers 0 and 1. By traversing the channels, the subscript index of the data that is first 1 (earthquake event) in each channel is recorded, which is the wavefront of the seismic signal recorded in the current channel. Finally, the seismic wavefront recorded by each virtual sensor element along the optical fiber is obtained, such as Figure 7 shown.
[0094] But in fact, the optical fiber is locally bent at certain locations due to long-term sea waves and geological effects, and the seismic transmission process is affected by complex geology and produces multiple reflection waves. These factors will cause the wavefront profile of the received seismic signal to be not a straight line, but a locally distorted curve, which will have a certain impact on the final position estimation. Therefore, a straight line fitting is required for the initial wavefront profile.
[0095] The straight line fitting first fits the initial wavefront according to the linear function formula y=kx+b to get fit1. Then the initial wavefront is interpolated and fitted. The main idea is to select the average value of the wavefront information of the first 500m and the last 500m of the optical fiber, and use these two values as the start and end for linear interpolation, and finally get fit2. Finally, the average value of the two fitting methods is calculated as the final straight line fitting result. The first fitting method is to use the entire section of the optical fiber data as much as possible, but due to some external factors, the original layout of the optical fiber is partially changed. Therefore, the second fitting method focuses on reducing the error of the intermediate data caused by the influence of the optical fiber layout, such as Figure 8 As shown;
[0096] At the epicenter coordinates (x,y) Theoretical value The wavefront T(x,t), such as Fig. 9 As shown:
[0097]
[0098] After linear fitting, the wavefront of the earthquake signal is obtained. The grid search method is used to search for the optimal solution in the solution space. Traverse the x and y of the solution space and find the weight value of each point. The larger the weight value, the greater the probability of an earthquake occurring in this direction. Fig.10 The result graph shows that there is a set of mirror-symmetric results. This is because the optical fiber is approximately regarded as a linear array, and the azimuth estimation through the linear array always obtains a pair of angle results that are mirror-symmetric about the optical fiber. (In other words, for the seismic wave signals incident from these two directions, the wavefront extraction results for the received data are approximately the same)
[0099] The weight calculation method is as follows:
[0100]
[0101] In a further embodiment, the preset earthquake identification network, such as Fig.12 As shown, from input to output, the encoding module, the decoding module and the projection module are sequentially arranged, wherein:
[0102] The encoding module includes six second encoders connected in sequence, each of which includes a BN layer, two two-dimensional convolutional layers, a maximum pooling layer and a dimension matching layer, wherein the output end of the BN layer is respectively connected to the input end of a two-dimensional convolutional layer and the input end of the dimension matching layer, the two two-dimensional convolutional layers are connected, and the output end of the last two-dimensional convolutional layer is merged with the output end of the dimension matching layer and then connected to the input end of the maximum pooling layer;
[0103] The decoding module includes six second decoders connected in sequence, each of which includes an upsampling layer, a BN layer, two one-dimensional convolutional layers and a dimension matching layer, wherein the upsampling layer and the BN layer are connected in sequence, the output end of the BN layer is connected to the input end of a one-dimensional convolutional layer and the input end of the dimension matching layer respectively, the two one-dimensional convolutional layers are connected, and the output end of the last one-dimensional convolutional layer is combined with the output end of the dimension matching layer as the output of the second decoder;
[0104] The projection module includes a residual structure consisting of three one-dimensional convolutional layers and one dimension matching layer.
[0105] In a specific embodiment, each layer of the earthquake identification network is described as follows:
[0106] The data set used for training is selected from the data recorded by the DAS deployed on Triangle Island from September 2021 to September 2022. Since a single DAS data contains the signals recorded by all virtual sensors of the entire optical fiber, the data size is too large, which brings a huge computational burden to the computer. Therefore, a DAS file is processed by sliding window according to the channel dimension, with a sliding window length of 32 channels and a step size of 16 channels. In addition, 41s of data are intercepted in the time dimension for a total of 2048 sampling points (the sampling frequency is reduced to 50Hz). The final data size input to the network is (2048,32,1,B), where B represents the size of each Batch in the training process.
[0107] Since this training is supervised training, each input data needs to be calibrated before training. The obtained label is a one-dimensional sequence of length 2048. The sequence contains only 0 or 1, where 0 represents a noise event and 1 represents an earthquake event. Fig.11 (a) represents data containing seismic signals, Fig.11 (b) shows the data without the seismic signal.
[0108] (1) Encoding module: The function of the encoding module is to gradually expand the feature dimension of the input data, so that the network can automatically extract signal features, eliminating the need for manual processing. The encoding module consists of six encoders, each of which consists of a BN layer, two two-dimensional convolutional layers (Conv2D-N), a maximum pooling layer (Maxing-Pooling), and a dimensional matching layer (DimensionalMatching). The specific structure and parameter configuration of the encoder are shown in Table 3.
[0109] Table 3
[0110]
[0111] (2) Decoding module: The main task of the decoding module is to use the data features obtained by the encoder to gradually restore the detail information to the target sequence size. Each decoder consists of a sampling layer, a BN layer, two one-dimensional convolutional layers (Conv1D-N), and a dimension matching layer. The specific structure and parameter configuration of the decoder are shown in Table 4.
[0112] Table 4
[0113]
[0114] (3) Projection module: The essence of the projection module is a residual module consisting of three one-dimensional convolutional layers (including a dimension matching layer). Its function is to map the output of the decoder to the final prediction result, thereby improving the network performance. Its specific structure and parameter configuration are shown in Table 5 below.
[0115] Table 5
[0116]
[0117] Finally, the structure and parameter configuration of the entire recognition network are shown in Table 6, which will realize the end-to-end conversion from a two-dimensional spatiotemporal array to a one-dimensional recognition sequence without any human intervention.
[0118] The recognition result is as follows Fig.13 As shown, Fig.13 The black line segment in (a) is the result after being processed by the recognition network. It can be seen that the recognition network can accurately capture the arrival time of the P wave and S wave of the earthquake event. Fig.13 (b) is the result of data without earthquake events after passing through the denoising network. It can be seen that the recognition network did not identify the earthquake in the data.
[0119] Table 6
[0120]
[0121] In a further embodiment, the epicenter distance of the earthquake is calculated based on the arrival time difference between the P wave and the S wave of the earthquake event. For an earthquake event, two body waves, the P wave and the S wave, are always generated. Since the propagation speeds of the P wave and the S wave are different, their arrival times are different, but the propagation paths are the same. Using their propagation time difference relationship, the following formula can be obtained:
[0122]
[0123] By transforming the above formula, we can get the function expression of epicenter distance Δ:
[0124]
[0125] Where Δ is the distance from the epicenter of the earthquake, Δt is the arrival time difference between the P wave and the S wave, and v P is the velocity of P wave in solid medium, v S is the velocity of S waves in solid media, and the general empirical values are 5.9 km / s and 3.5 km / s. So far, the epicenter distance has been estimated. The earthquake occurs on a circle with the distance from the optical fiber as the center and the epicenter distance as the radius.
[0126] In a further embodiment, obtaining the epicenter location of an earthquake according to the azimuth of the earthquake and the distance from the epicenter of the earthquake comprises:
[0127] By fusing the estimated azimuth and epicenter distance information, we can determine the locations of the two earthquakes. We need to make a decision on these two locations. (The reason why there are two locations is that when determining the azimuth, a set of mirror-symmetrical lines is fitted. After determining the epicenter distance, there will be two points on the circle that intersect with this set of mirror-symmetrical lines.)
[0128] Although the occurrence of earthquake events is random, earthquakes often occur in certain geologically active seismic zones. By statistically analyzing historical earthquake events, we select the results located at earthquake-prone points. Finally, the determined position coordinates are converted from rectangular coordinates to the actual longitude and latitude coordinates of the earth. The conversion of rectangular coordinates to longitude and latitude coordinates is based on the fact that for two points with the same longitude, the latitude differs by 0.001° every 111m, and for two points with the same latitude, the longitude differs by 0.01° every 1000m. Finally, through the entire algorithm, the estimated longitude and latitude of the epicenter of the earthquake is obtained, such as Fig.14 shown.
[0129] The solid white dot in the center of the image indicates the location of Guishan Island, which will serve as the reference point of the entire model, i.e., the coordinate origin. The dotted circle in the figure is a circle drawn with the epicenter distance (obtained by the P-wave and S-wave arrival time difference algorithm) as the radius. The bright radial area indicates the most likely estimated azimuth, and the junction of the circle and the radial area is the estimated epicenter. The red triangle pattern in the figure indicates the epicenter of the earthquake officially reported by the Earthquake Administration, and the green square indicates the result estimated using the algorithm proposed in this article. It can be intuitively seen that the estimated value and the reported value are basically consistent.
[0130] In order to quantitatively evaluate the results, Table 7 below gives the detailed information of these earthquake events reported by the Earthquake Bureau, as well as the errors between the estimated and reported values.
[0131] Table 7
[0132]
[0133] After calculation, the average positioning error of the proposed algorithm is (±0.058°, ±0.063°).
[0134] The same or similar reference numerals correspond to the same or similar components;
[0135] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;
[0136] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.
Claims
1. A method for locating the epicenter of an earthquake based on distributed acoustic sensing, characterized in that: The following steps are involved: Acquiring DAS data based on distributed acoustic sensing, and preprocessing the DAS data; The preprocessed DAS data is processed using a preset earthquake event segmentation network to obtain segmented earthquake data; Based on the segmented seismic data, an azimuth of the earthquake is estimated; The pre-processed DAS data is processed using a preset earthquake identification network to obtain the arrival time difference of the P wave and S wave of the earthquake event; Calculating the epicenter distance of the earthquake based on the arrival time difference of the P wave and the S wave of the earthquake event; The epicenter location of the earthquake is obtained according to the azimuth of the earthquake and the distance from the epicenter of the earthquake.
2. The earthquake epicenter location method based on distributed acoustic sensing according to claim 1, characterized in that: Preprocessing the DAS data includes: Bandpass filtering the DAS data through a 3-20 Hz bandpass filter; The bandpass filtered DAS data were normalized to zero mean; The zero-mean normalized DAS data is downsampled in the time domain to obtain the preprocessed DAS data.
3. The earthquake epicenter location method based on distributed acoustic sensing according to claim 1, characterized in that: The preset earthquake recognition network is provided with four first encoders, two two-dimensional convolutional layers, four first decoders and one two-dimensional convolutional layer in sequence from input to output, wherein: The input is the pre-processed DAS data; There is also a jump connection between the nth first encoder and the 5th to nth first decoders, n = 1, 2, 3, 4; After the output of the last two-dimensional convolutional layer is combined with the input, the output of the preset earthquake identification network is obtained.
4. The earthquake epicenter location method based on distributed acoustic sensing according to claim 3, characterized in that: Each of the first encoders includes a BN layer, a residual convolution module and a pooling layer connected in sequence.
5. The earthquake epicenter location method based on distributed acoustic sensing according to claim 3, characterized in that: Each of the first decoders includes an upsampling layer, a connection layer, a BN layer and a residual convolution module connected in sequence.
6. The earthquake epicenter location method based on distributed acoustic sensing according to any one of claims 1, 3 to 5, characterized in that: The pre-processed DAS data is processed using a preset seismic event segmentation network to obtain segmented seismic data, including: The preprocessed DAS data is input into the preset earthquake event segmentation network, wherein the size of the preprocessed DAS data is (H, W), and the preset earthquake event segmentation network outputs a data of size (H, W, 2), including 2 arrays Mask of size (H, W) seismic and Mask noise , Mask seismic and Mask noise They are the seismic signal weight matrix and the noise signal weight matrix, Mask seismic (i, j) indicates that the i-th row and j-th column of the preprocessed DAS data is the weight of the seismic signal. Similarly, Mask noise (i, j) indicates that the data at the i-th row and j-th column is the weight of the noise signal. seismic (i, j) is greater than Mask noise (i, j), the i-th row and j-th column of the preprocessed DAS data is a seismic signal, which is set to 1. seismic (i, j) is less than Mask noise (i, j), the position of the i-th row and j-th column of the preprocessed DAS data is a noise signal, which is set to 0 to obtain the segmented seismic data.
7. The earthquake epicenter location method based on distributed acoustic sensing according to claim 6, characterized in that: Based on the segmented seismic data, an azimuth of the earthquake is estimated, including: Traversing the segmented seismic data by channel, recording the subscript index of the data that is first 1 in each channel as the wavefront of the seismic signal recorded in the current channel, and finally obtaining the seismic wavefront recorded along each virtual sensor element; Performing straight line fitting on the seismic wavefront to obtain a fitted seismic wavefront; According to the fitted seismic wavefront, a grid search is used to traverse the solution space, the weight value of each point is calculated, and the azimuth of the earthquake is obtained.
8. The earthquake epicenter location method based on distributed acoustic sensing according to claim 1, characterized in that: The preset earthquake identification network is provided with an encoding module, a decoding module and a projection module in sequence from input to output, wherein: The encoding module includes six second encoders connected in sequence, each of which includes a BN layer, two two-dimensional convolutional layers, a maximum pooling layer and a dimension matching layer, wherein the output end of the BN layer is respectively connected to the input end of a two-dimensional convolutional layer and the input end of the dimension matching layer, the two two-dimensional convolutional layers are connected, and the output end of the last two-dimensional convolutional layer is merged with the output end of the dimension matching layer and then connected to the input end of the maximum pooling layer; The decoding module includes six second decoders connected in sequence, each of which includes an upsampling layer, a BN layer, two one-dimensional convolutional layers and a dimension matching layer, wherein the upsampling layer and the BN layer are connected in sequence, the output end of the BN layer is connected to the input end of a one-dimensional convolutional layer and the input end of the dimension matching layer respectively, the two one-dimensional convolutional layers are connected, and the output end of the last one-dimensional convolutional layer is combined with the output end of the dimension matching layer as the output of the second decoder; The projection module includes a residual structure consisting of three one-dimensional convolutional layers and one dimension matching layer.
9. The earthquake epicenter location method based on distributed acoustic sensing according to claim 8, characterized in that: Calculating the epicenter distance of an earthquake based on the arrival time difference between the P wave and the S wave of the earthquake event, including: Where Δ is the distance from the epicenter of the earthquake, Δt is the arrival time difference between the P wave and the S wave, and v P is the velocity of P wave in solid medium, v S is the velocity of S waves in solid media.
10. The earthquake epicenter location method based on distributed acoustic sensing according to claim 9, characterized in that: According to the azimuth of the earthquake and the distance from the epicenter of the earthquake, the epicenter of the earthquake is located, including: The azimuth of the earthquake and the distance from the epicenter of the earthquake are combined to obtain the coordinates of the location where the earthquake occurred through judgment; The location coordinates of the earthquake are converted into longitude and latitude coordinates to obtain the epicenter of the earthquake.
