Seismic exploration node data quality detection method and device

Abstract

The invention relates to a seismic exploration node data quality detection method and device which are used in the field of quality control of seismic exploration data, and the method comprises the steps: obtaining to-be-detected seismic time sequence collection data of a plurality of data collection nodes; preprocessing the to-be-detected seismic time sequence acquisition data to obtain to-be-detected time sequence data with reserved data features and reduced data volume; calculating the similarity among the to-be-detected time series data of the plurality of data acquisition nodes under different synchronization lengths to obtain a to-be-detected multi-scale similarity matrix set; the step length represents the number of adjacent data points in the to-be-detected time series data; and inputting the to-be-detected multi-scale similarity matrix set into a pre-constructed time sequence anomaly detection model, and outputting a detection result of whether an abnormal data point exists or not. High-efficiency and high-accuracy anomaly detection is realized, and the manual workload is reduced.

Description

Technical Field

[0001] The present disclosure relates to the technical field of geological exploration and data quality inspection, and in particular to a method and device for detecting the quality of seismic exploration node data. Background Art

[0002] In geological exploration technology, seismic exploration technology is usually used to analyze geological structures and potential oil and gas therein. Seismic exploration is the most important and effective method in geophysical exploration to solve oil and gas exploration problems. It is an important means of surveying oil and natural gas resources before drilling, and is also widely used in coalfield and engineering geological exploration, regional geological research and crustal research. Seismic exploration refers to a geophysical exploration method that uses the differences in elasticity and density of underground media caused by artificially excited elastic waves to observe and analyze the propagation laws of seismic waves generated by artificial earthquakes underground to infer the properties and morphology of underground rock formations.

[0003] In the process of realizing the concept of the present disclosure, the inventors found that there are at least the following technical problems in the related technology: In the seismic exploration and acquisition link, with the complexity of the exploration area, node instruments have been widely used in recent years; however, since the node instruments cannot transmit the collected seismic data in real time, nor can they provide the results of quality inspection of the node collected data in real time, in the node instrument acquisition project, for problems such as abnormal nodes and strong noise interference, usually only waiting for the node to recover and download the data for inspection. Moreover, in the process of quality inspection, most of them need to rely on the judgment and experience of experienced engineers, the operation is cumbersome and prone to manual errors. Summary of the invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, an embodiment of the present disclosure provides a method and device for detecting the quality of seismic exploration node data.

[0005] In the first aspect, the embodiments of the present disclosure provide a method for detecting the data quality of seismic exploration nodes. The above method includes: obtaining the seismic time series acquisition data to be detected of multiple data acquisition nodes; preprocessing the above seismic time series acquisition data to be detected to obtain the time series data to be detected with the data features retained and the data volume reduced; calculating the similarity between the time series data to be detected of multiple data acquisition nodes under different step lengths to obtain a set of multi-scale similarity matrices to be detected; the above step length represents the number of adjacent data points in the above time series data to be detected; inputting the above multi-scale similarity matrix set to be detected into a pre-constructed time series anomaly detection model, and outputting the detection result of whether there are abnormal data points.

[0006] According to an embodiment of the present disclosure, the above-mentioned time series anomaly detection model is a multi-scale convolutional recursive encoder and decoder MSCRED network model based on an attention mechanism; the above-mentioned time series anomaly detection model includes: an encoding module, multiple LSTM models based on an attention mechanism, a decoding module and a comparison module. The above-mentioned encoding module includes multiple convolutional layers, and the above-mentioned multiple convolutional layers are used to perform layer-by-layer encoding processing on the above-mentioned multi-scale similarity matrix set to be detected to obtain a multi-layer encoding feature map. In the above-mentioned multiple LSTM models based on the attention mechanism, each LSTM model includes: an LSTM network, an attention module and an output module, and the above-mentioned LSTM network is used to perform time series context learning on each layer of the encoding feature map in the above-mentioned multi-layer encoding feature map to obtain the hidden layer vector at each moment within the preset backtracking time length; the above-mentioned attention module is used to determine the attention weights corresponding to the hidden layer vector at the current moment and the hidden layer vector at the previous moment within the above-mentioned preset backtracking step length; the above-mentioned output module is used to weight and calculate the hidden layer vectors at each previous moment and the current moment within the above-mentioned preset backtracking step length and the corresponding attention weights, obtain intermediate variables and output them. The decoding module is used to perform decoding processing according to the intermediate variables and each layer of the coding feature map in the multi-layer coding feature map to obtain a reconstructed similarity matrix set. The comparison module is used to compare the gap between the corresponding data points in the reconstructed similarity matrix set and the multi-scale similarity matrix set to be detected, and in the case of a target data point whose gap exceeds a set threshold, the target data point is determined to be an abnormal data point.

[0007] According to an embodiment of the present disclosure, the multi-layer convolutional layers include: m convolutional layers, where m≥3 and is a positive integer; the encoding module performs layer-by-layer encoding processing on the multi-scale similarity matrix set to be detected, including: based on the first convolutional layer, feature extraction is performed on the input multi-scale similarity matrix set to be detected to obtain a first-layer encoding feature map; based on the second convolutional layer, downsampling is performed on the first-layer encoding feature map to obtain a second-layer encoding feature map; based on the m-th convolutional layer, downsampling is performed on the m-1th layer encoding feature map to obtain an m-th layer encoding feature map; wherein the first-layer encoding feature map to the m-th layer encoding feature map are respectively input into m LSTM models based on the attention mechanism for processing to obtain corresponding m intermediate variables.

[0008] According to an embodiment of the present disclosure, in the above-mentioned decoding module, decoding processing is performed according to the above-mentioned intermediate variables and each layer of the encoding feature maps in the above-mentioned multi-layer encoding feature maps to obtain a reconstructed similarity matrix set, including: upsampling the above-mentioned m-th intermediate variable to obtain a 0th layer decoding feature map with the same dimension as the m-1th intermediate variable; splicing the 0th layer decoding feature map with the above-mentioned m-1th intermediate variable to obtain a 1st layer decoding feature map; upsampling the above-mentioned 1st layer decoding feature map to obtain a 2nd layer decoding feature map with the same dimension as the m-2th intermediate variable; splicing the 2nd layer decoding feature map with the above-mentioned m-2th intermediate variable to obtain a 3rd layer decoding feature map; and performing feature reconstruction on the last layer decoding feature map to obtain a reconstructed similarity matrix set with the same dimension as the above-mentioned multi-scale similarity matrix set to be detected.

[0009] According to an embodiment of the present disclosure, the similarity between the time series data to be detected of multiple data acquisition nodes under different step sizes is calculated to obtain a set of multi-scale similarity matrices to be detected, including: numbering the data acquisition nodes according to the spatial arrangement order; for each data acquisition node, determining the time series segment data under multiple step segments of the time series data to be detected corresponding to each step size; for the time series segment data in each step segment of each data acquisition node corresponding to each step size, performing the following steps to obtain a triangular first similarity matrix corresponding to each step segment under each step size: calculating the first similarity between the current data acquisition node and its own time series segment data in the current step segment, the first similarity being the diagonal element of the first similarity matrix at time t, the row number and column number of the diagonal element being equal to the number of the current data acquisition node; time t being the last number in the current step segment According to the corresponding moment; calculate the second similarity between the time segment data of the current data acquisition node and the subsequent data acquisition node under the current step segment, the above second similarity is used as the non-diagonal element of the above first similarity matrix, the row number of the above non-diagonal element is equal to the number of the above current data acquisition node, and the column number of the above non-diagonal element is equal to the number of the above subsequent data acquisition node; the number of the above subsequent data acquisition node is greater than the number of the above current data acquisition node; based on the matrix symmetry algorithm, the above first similarity matrix in triangular form is processed to obtain the second similarity matrix corresponding to each of the multiple step segments under each step, and the above second similarity matrix is ​​a symmetric matrix; the above second similarity matrix is ​​combined according to the two dimensions of the time sequence of the step period and the step size to obtain a set of multi-scale similarity matrices to be detected, and the above multi-scale similarity matrix set to be detected is in the form of a tensor.

[0010] According to an embodiment of the present disclosure, the above-mentioned seismic time series acquisition data to be detected is preprocessed to obtain the time series data to be detected with the data features retained and the data volume reduced, including: determining the target neighborhood range corresponding to the working condition of the above-mentioned data acquisition node according to the preset correspondence between the exploration condition and the neighborhood range; dividing the above-mentioned seismic time series acquisition data to be detected according to the above-mentioned target neighborhood range, taking the mean value of the data divided into the same target neighborhood range, and using each mean value result as the data value in the corresponding target neighborhood range; arranging the mean value results in each target neighborhood range in chronological order to obtain the above-mentioned time series data to be detected.

[0011] According to an embodiment of the present disclosure, the above method also includes: pre-building a time series anomaly detection model. The above pre-building of the time series anomaly detection model includes: obtaining seismic time series acquisition data from multiple data acquisition nodes; pre-processing the above seismic time series acquisition data to obtain time series data with data features retained and data volume reduced; calculating the similarity between the time series data of multiple data acquisition nodes under different step lengths to obtain a multi-scale similarity matrix set; the above step length represents the number of adjacent data points in the above time series data; the above step length is the same in the model training stage and the model use stage; using the first part of the above multi-scale similarity matrix set as the training set data and the second part as the test set data; based on the above training set data, the time series The anomaly detection model is trained, wherein the parameters of the above-mentioned time series anomaly detection model are updated during the training process until the training reaches a set number of times or the loss function is lower than a set value, and the above-mentioned loss function represents the gap between the input multi-scale similarity matrix set and the reconstructed similarity matrix set output by the decoding module; based on the above-mentioned test set data, the trained time series anomaly detection model is tested, and the parameters of the above-mentioned time series anomaly detection model are fine-tuned during the test process, so that the prediction accuracy of the adjusted time series anomaly detection model meets the set accuracy requirement, and the adjusted time series anomaly detection model is used as the constructed time series anomaly detection model.

[0012] In the second aspect, an embodiment of the present disclosure provides a device for detecting the data quality of seismic exploration nodes. The above-mentioned device includes: a first data acquisition module, a first preprocessing module, a first similarity calculation module and an anomaly detection module. The above-mentioned first data acquisition module is used to acquire the seismic time series acquisition data to be detected of multiple data acquisition nodes. The above-mentioned first preprocessing module is used to preprocess the above-mentioned seismic time series acquisition data to be detected, and obtain the time series data to be detected after retaining the data characteristics and reducing the data volume. The above-mentioned first similarity calculation module is used to calculate the similarity between the time series data to be detected of multiple data acquisition nodes under different step lengths, and obtain a set of multi-scale similarity matrices to be detected; the above-mentioned step length represents the number of adjacent data points in the above-mentioned time series data to be detected. The above-mentioned anomaly detection module is used to input the above-mentioned multi-scale similarity matrix set to be detected into a pre-constructed time series anomaly detection model, and output the detection result of whether there are abnormal data points.

[0013] In a third aspect, an embodiment of the present disclosure provides a device for constructing a time series anomaly detection model. The above-mentioned time series anomaly detection model is used for data quality detection of seismic exploration nodes, and the above-mentioned device includes: a second data acquisition module, a second preprocessing module, a second similarity calculation module, a data partitioning module, a training module and a testing module. The above-mentioned second data acquisition module is used to acquire seismic time series acquisition data of multiple data acquisition nodes. The above-mentioned second preprocessing module is used to preprocess the above-mentioned seismic time series acquisition data to obtain time series data with data features retained and data volume reduced. The above-mentioned second similarity calculation module is used to calculate the similarity between the time series data of multiple data acquisition nodes under different step lengths to obtain a multi-scale similarity matrix set; the above-mentioned step length represents the number of adjacent data points in the above-mentioned time series data; the above-mentioned step length is the same in the model training stage and the model use stage. The above-mentioned data partitioning module is used to use the first part of the above-mentioned multi-scale similarity matrix set as the training set data and the second part as the test set data. The training module is used to train the time series anomaly detection model using the training set data, wherein the parameters of the time series anomaly detection model are updated during the training process until the training reaches a set number of times or the loss function is lower than a set value, and the loss function represents the gap between the input multi-scale similarity matrix set and the reconstructed similarity matrix set output by the decoding module. The test module is used to test the trained time series anomaly detection model using the test set data, and fine-tune the parameters of the time series anomaly detection model during the test process, so that the prediction accuracy of the adjusted time series anomaly detection model meets the set accuracy requirements. The trained time series anomaly detection model is used as the constructed time series anomaly detection model; or, the adjusted time series anomaly detection model is used as the constructed time series anomaly detection model.

[0014] In a fourth aspect, an embodiment of the present disclosure provides an electronic device. The electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; and the processor is used to implement the method for detecting the quality of seismic exploration node data as described above when executing the program stored in the memory.

[0015] In a fifth aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the method for detecting the quality of seismic exploration node data as described above is implemented.

[0016] The above technical solution provided by the embodiments of the present disclosure has at least some or all of the following advantages:

[0017] By acquiring the time series acquisition data of earthquakes to be detected from multiple data acquisition nodes; preprocessing the above-mentioned time series acquisition data of earthquakes to be detected to obtain the time series data to be detected with the data features retained and the data volume reduced; effectively reducing the amount of data used for subsequent calculations and improving the operation efficiency; considering that the correlation between the corresponding time series data of different data acquisition nodes is closely related to the seismic echo, for the same set of seismic data acquisition nodes, based on the similarity between the data of data acquisition nodes in spatially adjacent areas and the performance under different time scales (corresponding to different step lengths), the similarity between the time series data to be detected of multiple data acquisition nodes under different step lengths is calculated to obtain a set of multi-scale similarity matrices to be detected; the above-mentioned multi-scale similarity matrix set to be detected is input into the pre-built MSCRED network model based on the attention mechanism to realize efficient and high-accuracy anomaly detection, which can quickly locate the anomalies in the data, shorten the data quality control time of the seismic data acquisition nodes, and reduce the manual workload. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0019] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0020] Figure 1 The flowchart of the method for detecting the quality of seismic exploration node data according to the embodiment of the present disclosure is schematically shown;

[0021] Figure 2 A schematic diagram schematically shows a scenario of similarity calculation of data collected by 30 data collection nodes and pre-processed according to an embodiment of the present disclosure;

[0022] Figure 3 An example of calculating the similarity of time series segment data of multiple data acquisition nodes within a certain step length segment at a certain backtracking scale to obtain a first similarity matrix in a triangular form according to an embodiment of the present disclosure is schematically shown;

[0023] Figure 4 Schematically illustrates an example of obtaining a second similarity matrix through matrix symmetry operation processing according to an embodiment of the present disclosure;

[0024] Figure 5 The following schematically shows a process diagram of anomaly detection performed by a time series anomaly detection model during training and testing phases according to an embodiment of the present disclosure;

[0025] Figure 6 A schematic diagram schematically shows hidden layer vectors at various moments within a preset lookback time obtained by an LSTM model based on an attention mechanism according to an embodiment of the present disclosure;

[0026] Figure 7 The structural block diagram of the electronic device provided by the embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0028] The first exemplary embodiment of the present disclosure provides a method for detecting data quality of a seismic exploration node. The method of this embodiment can be executed by an electronic device with computing capability, on which a pre-built time series anomaly detection model is stored; or a pre-built time series anomaly detection model can be called. The above-mentioned electronic device can be a computing device independent of the seismic exploration node; or the above-mentioned electronic device is a seismic exploration node, and a built-in processing module needs to be newly added in a conventional seismic exploration node, and the above-mentioned processing module can implement the method provided by the embodiment of the present disclosure.

[0029] Figure 1 A flow chart of a method for detecting seismic exploration node data quality according to an embodiment of the present disclosure is schematically shown.

[0030] Reference Figure 1 As shown, the method for detecting the quality of seismic exploration node data provided by the embodiment of the present disclosure includes the following steps: S110, S120, S130 and S140.

[0031] In step S110, the time series acquisition data of earthquakes to be detected of multiple data acquisition nodes are acquired.

[0032] The above data acquisition nodes are seismic exploration nodes, and multiple seismic exploration nodes are placed in a spatial arrangement order and used to collect seismic echo signals. The data collected by each data acquisition node is time series data, so it is described as seismic time series acquisition data. In order to distinguish the training stage and the use stage of the time series anomaly detection model, the seismic time series acquisition data to be detected is used for description in the use stage of the model; the seismic time series acquisition data is used for description in the training stage of the model.

[0033] The time series data of the earthquake to be detected of the data acquisition node (referred to as node) is a data body. According to the length T (time length) of the node record data and the sampling interval n, it is concluded that the data length includes k = T / n + 1 sample point values, that is, a single node contains k data, and the data of a single node is defined as d = [d 1 ,d 2 ,...,d k ], the superscript of d indicates the data serial number corresponding to a single node. The earthquake time series data to be monitored of all data acquisition nodes are loaded into the electronic device in a specific format to form a data set D = [d1, d2, ..., d n ], n represents the total number of data collection nodes.

[0034] For example, each sample point in the time series data of the earthquake to be detected is two-dimensional data, including the time dimension and the amplitude dimension. For example, the data collected by a single node is represented as {(t1, A1), (t2, A2), (t3, A3), … (t k ,A k )}. In an exemplary practical application scenario Y1, for example, the time series acquisition data of earthquakes to be detected collected by 30 seismic exploration nodes are loaded, the data format is segy (a standard tape data format for data acquisition and recording in geological exploration), the data record length is 32.95 hours, the sampling interval is 4 milliseconds, and the total number of sample point values ​​contained in each seismic exploration node is 29,655,000.

[0035] In step S120, the above-mentioned earthquake time series acquisition data to be detected is preprocessed to obtain the time series data to be detected with the data features retained and the data volume reduced.

[0036] According to an embodiment of the present disclosure, in the above-mentioned step S120, the above-mentioned seismic time series acquisition data to be detected is preprocessed to obtain the time series data to be detected with the data features retained and the data amount reduced, including: determining the target neighborhood range corresponding to the working condition of the above-mentioned data acquisition node according to the preset correspondence between the exploration condition and the neighborhood range; dividing the above-mentioned seismic time series acquisition data to be detected according to the above-mentioned target neighborhood range, taking the mean value of the data divided into the same target neighborhood range, and using each mean value result as the data value in the corresponding target neighborhood range; arranging the mean value results in each target neighborhood range in chronological order to obtain the above-mentioned time series data to be detected.

[0037] Among them, when calculating the mean of data within the same target neighborhood, the average is taken after the absolute value of the data.

[0038] In some embodiments, the above-mentioned exploration conditions include but are not limited to: the objects of actual geological exploration (different geological structures, different geological environments, etc.), the layout information of seismic exploration nodes in geological exploration (layout spacing, layout range, layout number, layout shape, etc.). Since the characteristics of the time series data collected by each data acquisition node corresponding to different exploration conditions are different, the corresponding neighborhood ranges are also different, and the data performance within the same neighborhood range is almost the same (the characteristics are close). Therefore, the preset corresponding relationship can be generated based on actual experience values, or the neighborhood range under each exploration condition can be set by experts, or the optimal neighborhood range can be selected according to the simulated performance of the seismic time series data under different neighborhood range processing results to construct the above-mentioned preset corresponding relationship.

[0039] Figure 2 A schematic diagram schematically shows a scenario of similarity calculation for data collected by 30 data collection nodes and pre-processed according to an embodiment of the present disclosure.

[0040] Reference Figure 2 As shown, in the above scenario Y1, the absolute value of the sample value of each node is taken and the neighborhood average processing is adopted. 1500 is used as the target neighborhood range, and the 1500 adjacent data are averaged as a group to obtain a matrix Ms containing 30 rows (corresponding to a total of 30 nodes) and 19770 data in each row (the specific data corresponds to the average result within each target neighborhood range). For example, Ms is expressed as follows:

[0041]

[0042] According to the example of scenario Y1 above, in the actual geological exploration process, the amount of data at each node is quite large. Therefore, the time series acquisition data of the earthquake to be detected of multiple data acquisition nodes are obtained; the time series acquisition data of the earthquake to be detected is preprocessed to obtain the time series data to be detected with the data features retained and the data amount reduced; the amount of data used for subsequent calculations is effectively reduced, thereby improving computing efficiency.

[0043] In step S130, the similarities between the time series data to be detected of multiple data acquisition nodes under different step sizes are calculated to obtain a set of multi-scale similarity matrices to be detected; the above step size represents the number of adjacent data points in the above time series data to be detected.

[0044] According to an embodiment of the present disclosure, in the above step S130, the similarity between the time series data to be detected of multiple data acquisition nodes under different step lengths is calculated to obtain a set of multi-scale similarity matrices to be detected, including:

[0045] The data acquisition nodes are numbered according to the spatial arrangement order; for example, 30 seismic exploration nodes are numbered from 1 to 30 in the order of adjacent arrangement;

[0046] For each data acquisition node, determine the time series segment data under multiple step length segments of the time series data to be detected corresponding to each step length; for example, divided into 3 step lengths, namely 10, 30 and 60, taking the step length of 10 as an example, the corresponding step length segments correspond to data serial numbers such as: 1-10, 11-20, 21-30, 31-40, ..., 19761-197770, a total of 19777 step length segments;

[0047] For the time series segment data in each step segment of each data acquisition node corresponding to each step size, the following steps are performed to obtain the first similarity matrix in a triangular form corresponding to each step segment under each step size: the first similarity between the current data acquisition node and its own time series segment data in the current step segment is calculated, the first similarity is used as the diagonal element of the first similarity matrix at time t, and the row number and column number of the diagonal element are both equal to the number of the current data acquisition node; time t is the time corresponding to the last data in the current step segment; the second similarity between the time series segment data of the current data acquisition node and the subsequent data acquisition node in the current step segment is calculated, the second similarity is used as the non-diagonal element of the first similarity matrix, the row number of the non-diagonal element is equal to the number of the current data acquisition node, and the column number of the non-diagonal element is equal to the number of the subsequent data acquisition node; the number of the subsequent data acquisition node is greater than the number of the current data acquisition node;

[0048] Based on a matrix symmetry algorithm, the first similarity matrix in triangular form is processed to obtain a second similarity matrix corresponding to each of the multiple step length segments under each step length, wherein the second similarity matrix is ​​a symmetric matrix;

[0049] The second similarity matrix is ​​combined according to the time sequence of the step period and the step length, to obtain a set of multi-scale similarity matrices to be detected, wherein the set of multi-scale similarity matrices to be detected is in a tensor form.

[0050] Define the multi-scale (referring to the time scale) set as L = [W1, W2, ..., W x ], at each scale, traverse the data in Ms, backtrack with each scale as the step length, x represents the total number of multiple steps, calculate the inner product of the time series fragment data of each row in Ms at each scale and the time series fragment data of other rows at the corresponding scale, the calculation formula is as follows:

[0051]

[0052] in, is the data of the i-th row in Ms, with the data at the current time t is the starting point and the data set formed by tracing back a total length of w. is the data of row j in Ms, with the data at the current time t As the starting point, trace back the data set formed by length w, and Calculate the inner product to get the similarity between the time series fragment data

[0053] Traverse all i and j in turn to obtain the first similarity matrix in triangular form.

[0054] Then, through the symmetric matrix algorithm, the second similarity matrix (in the form of a symmetric matrix) at the current time t is obtained. This matrix is ​​the similarity matrix corresponding to a certain time series segment data (for example, the 1st to 10th data) under a certain step size (for example, 10).

[0055] With a backtracking step length w, all the time series fragment data in Ms are traversed, that is, multiple step length segments are traversed. The number of step length segments R can be expressed as the following expression: R = k / c, k is the total number of sample data contained in a single node; c is the target neighborhood range adopted by the node; the similarity matrix corresponding to each step length segment under the same step length (for example, 10) is taken as an element, and by traversing R step length segments, a similarity matrix set Mw corresponding to a certain backtracking step length (for example, 10) is obtained. 1 ,Mw 2 ,..,Mw R]. Each similarity matrix is ​​obtained according to the implementation method of obtaining the symmetric matrix form above.

[0056] Then, for different backtracking steps, a multi-scale similarity matrix set M to be detected containing multiple similarity matrix sets can be obtained: d For example, for a multi-scale feature set L = [10, 30, 60], a multi-scale similarity matrix set M to be detected containing three similarity matrix sets Mw can be obtained: d , in the form of a 30×30×3 tensor.

[0057] Figure 3 An example of calculating the similarity of time series segment data of multiple data acquisition nodes within a certain step length segment at a certain backtracking scale to obtain a first similarity matrix in a triangular form according to an embodiment of the present disclosure is schematically shown.

[0058] As an example, define the multi-scale feature set as L = [10, 30, 60], use the multi-scale feature as the backtracking step, extract a row of data in Ms as the calculation row, perform inner product calculation with other rows of data, and obtain the similarity matrix.

[0059] For example, combined with Figure 2 and Figure 3 As shown in FIG. 1 , with 10 in the multi-scale feature set as the backtracking step, first extract the first step segment of the 19770 data in the first row of Ms, including the 1st to 10th data; use formula (1) to calculate the similarity between the 1st to 10th data and the 1st to 10th data in the first row itself, and obtain the elements of the 1st row and 1st column of the similarity matrix with a step length of 10; use formula (1) to calculate the similarity between the 1st to 10th data in the second row, and obtain the elements of the 1st row and 2nd column of the similarity matrix with a step length of 10, and so on, and use formula (1) to calculate the similarity between the 1st to 10th data in the 30th row, and obtain the elements of the 1st row and 2nd column of the similarity matrix with a step length of 10. 0 data are similarly calculated using formula (1) to obtain the element of the 1st row and 30th column of the similarity matrix with a step size of 10; similarly, for the 1st step size segment of the second row, the similarity with itself and the subsequent 3rd row, 4th row to 30th row can be calculated in the above manner, and the elements of the 2nd row and 2nd column, the 2nd row and 3rd column, the 2nd row and 4th column to the 2nd row and 30th column in the similarity matrix can be obtained respectively; recursively along the number of rows, the element of the 30th row and 30th column in the similarity matrix can be obtained; thus, the first similarity matrix in triangular form can be obtained, referring to Figure 3 shown.

[0060] Figure 4 An example of obtaining a second similarity matrix through matrix symmetry operation processing according to an embodiment of the present disclosure is schematically shown.

[0061] The symmetric matrix algorithm is used to perform symmetric processing on the first similarity matrix in the form of a triangular matrix to obtain a second similarity matrix of 30×30. Figure 4 As shown, it can be understood that in the second similarity matrix, the data value of the second row and the first column is the same as the data value of the first row and the second column, and the same can be applied to other positions, and the overall structure is a symmetric matrix.

[0062] In step S140, the above-mentioned multi-scale similarity matrix set to be detected is input into a pre-built time series anomaly detection model, and a detection result of whether there is an abnormal data point is output. Figure 5 The diagram schematically shows a process diagram of anomaly detection performed by a time series anomaly detection model during training and testing phases according to an embodiment of the present disclosure. Figure 6 A schematic diagram schematically shows hidden layer vectors at various moments within a preset lookback time obtained by an LSTM model based on an attention mechanism according to an embodiment of the present disclosure.

[0063] According to an embodiment of the present disclosure, the above-mentioned time series anomaly detection model is an MSCRED network model based on an attention mechanism; the above-mentioned time series anomaly detection model includes: an encoding module, multiple LSTM models based on an attention mechanism, a decoding module and a comparison module.

[0064] The encoding module comprises multiple layers of convolutional layers, and the multiple layers of convolutional layers are used to perform layer-by-layer encoding processing on the set of multi-scale similarity matrices to be detected to obtain a multi-layer encoding feature map.

[0065] Combination Figure 5 and Figure 6 As shown, in the above-mentioned multiple LSTM models based on the attention mechanism, each LSTM model includes: an LSTM network, an attention module and an output module. The above-mentioned LSTM network is used to perform temporal context learning on each layer of the encoding feature map in the above-mentioned multi-layer encoding feature map to obtain the hidden layer vector at each moment within the preset backtracking time, such as Figure 6 As shown, taking the preset backtracking time h as 5 as an example, the hidden layer vectors at each moment are The dimensions of each hidden layer vector are exemplified as 30×30×32; l represents the layer number of the corresponding encoding feature map, Vec in front of the hidden layer vector symbol in the following formula (3) represents the vector, T in the upper right corner represents the transpose of the vector, and x represents the parameter used when calculating the similarity between two hidden layer vectors, which can be the product of the two norms of the two vectors; the above-mentioned attention module is used to determine the attention weights corresponding to the hidden layer vector at the current moment and the hidden layer vector at the previous moment within the above-mentioned preset backtracking step; the above-mentioned output module is used to weight and calculate the hidden layer vectors at each previous moment and the current moment within the above-mentioned preset backtracking step and their corresponding attention weights, obtain intermediate variables and output them, and the process of outputting intermediate variables refers to Figure 5 shown.

[0066] The process of calculating intermediate variables is shown in the following expression:

[0067]

[0068] For each moment H i and H t (Last moment) Calculate the attention coefficient α i (3)

[0069] The decoding module is used to perform decoding processing according to the intermediate variables and each layer of the coding feature map in the multi-layer coding feature map to obtain a reconstructed similarity matrix set, referring to Figure 5 shown.

[0070] The comparison module is used to compare the gap between the reconstructed similarity matrix set and the corresponding data points in the multi-scale similarity matrix set to be detected. If there is a target data point whose gap exceeds a set threshold, the target data point is determined to be an abnormal data point. Figure 5 The usage phase is shown in .

[0071] According to an embodiment of the present disclosure, the multi-layer convolutional layer includes: m convolutional layers, m ≥ 3 and is a positive integer. For example, it can be 4 or 5 convolutional layers. The number of convolutional layers in a specific model is a hyperparameter, and the optimal number of layers can be selected by comparing the anomaly detection effects of different convolutional layers.

[0072] The encoding module performs layer-by-layer encoding processing on the multi-scale similarity matrix set to be detected, including:

[0073] Based on the first convolution layer, feature extraction is performed on the input multi-scale similarity matrix set to be detected (for example, 30×30×3) to obtain the first-layer encoding feature map. For example, the stride (sliding step) of the convolution kernel used in the first convolution layer is 1×1. During the convolution operation, while the convolution kernel slides, the product of the weight matrix and the scanned data matrix is ​​calculated, and then the result is summarized into an output; the corresponding first-layer encoding feature map is an output tensor, for example, with a dimension of 30×30×32;

[0074] Based on the second convolution layer, the first-layer encoding feature map is downsampled to obtain the second-layer encoding feature map; for example, the stride of the convolution kernel used in the second convolution layer is 2×2, and the second-layer encoding feature map is also in tensor form, for example, with a dimension of 15×15×64;

[0075] Based on the m-th convolutional layer, the m-1-th layer encoding feature map is downsampled to obtain the m-th layer encoding feature map; wherein, the 1st layer encoding feature map to the m-th layer encoding feature map are respectively input into m LSTM models based on the attention mechanism for processing to obtain corresponding m intermediate variables;

[0076] Taking m=4 as an example, for example, the stride of the convolution kernel used by the third convolution layer is 2×2, and the third layer encoding feature map is also in tensor form, for example, with a dimension of 8×8×128; the stride of the convolution kernel used by the fourth convolution layer is 2×2, and the fourth layer encoding feature map is also in tensor form, for example, with a dimension of 4×4×256.

[0077] According to an embodiment of the present disclosure, in the above-mentioned decoding module, decoding processing is performed according to the above-mentioned intermediate variables and each layer of the encoding feature maps in the above-mentioned multi-layer encoding feature maps to obtain a reconstructed similarity matrix set, including: upsampling the above-mentioned m-th intermediate variable to obtain a 0th layer decoding feature map with the same dimension as the m-1th intermediate variable; splicing the 0th layer decoding feature map with the above-mentioned m-1th intermediate variable to obtain a 1st layer decoding feature map; upsampling the above-mentioned 1st layer decoding feature map to obtain a 2nd layer decoding feature map with the same dimension as the m-2th intermediate variable; splicing the 2nd layer decoding feature map with the above-mentioned m-2th intermediate variable to obtain a 3rd layer decoding feature map; and performing feature reconstruction on the last layer decoding feature map to obtain a reconstructed similarity matrix set with the same dimension as the above-mentioned multi-scale similarity matrix set to be detected.

[0078] According to an embodiment of the present disclosure, the above method further includes: pre-building a time series anomaly detection model.

[0079] Reference Figure 5The training phase and the previous process description of the model use phase, the above-mentioned pre-built time series anomaly detection model includes: obtaining seismic time series acquisition data from multiple data acquisition nodes; preprocessing the above-mentioned seismic time series acquisition data to obtain time series data with preserved data features and reduced data volume; calculating the similarity between the time series data of multiple data acquisition nodes under different step lengths to obtain a multi-scale similarity matrix set; the above-mentioned step length represents the number of adjacent data points in the above-mentioned time series data; the above-mentioned step length is the same in the model training phase and the model use phase; using the first part of the above-mentioned multi-scale similarity matrix set as the training set data, and the second part as the test set data; based on The training set data is used to train the time series anomaly detection model, wherein the parameters of the time series anomaly detection model are updated during the training process until the training reaches a set number of times or the loss function is lower than a set value, and the loss function represents the gap between the input multi-scale similarity matrix set and the reconstructed similarity matrix set output by the decoding module; based on the test set data, the trained time series anomaly detection model is tested, and the parameters of the time series anomaly detection model are fine-tuned during the test process so that the prediction accuracy of the adjusted time series anomaly detection model meets the set accuracy requirement, and the adjusted time series anomaly detection model is used as the constructed time series anomaly detection model.

[0080] In the process of building a time series anomaly detection model, the seismic time series acquisition data of multiple data acquisition nodes are obtained, the seismic time series acquisition data are preprocessed, the similarity between the time series data of multiple data acquisition nodes under different step lengths is calculated, and a multi-scale similarity matrix set is obtained. The above processes can refer to the previous details of the corresponding steps in the model usage stage.

[0081] In an embodiment including the above steps S110 to S140, the time series acquisition data of earthquakes to be detected of multiple data acquisition nodes are obtained; the above time series acquisition data of earthquakes to be detected are preprocessed to obtain the time series data to be detected with the data features retained and the data volume reduced; the amount of data used for subsequent calculations is effectively reduced, and the operation efficiency is improved; considering that the correlation between the corresponding time series data of different data acquisition nodes is closely related to the seismic echo, for the same set of seismic data acquisition nodes, based on the similarity between the data of data acquisition nodes in spatially adjacent areas and the performance under different time scales (corresponding to different step lengths), the similarity between the time series data to be detected of multiple data acquisition nodes under different step lengths is calculated to obtain a set of multi-scale similarity matrices to be detected; the above multi-scale similarity matrix set to be detected is input into a pre-constructed MSCRED network model based on the attention mechanism, so as to realize efficient and high-accuracy anomaly detection, quickly locate the anomalies in the data, shorten the data quality control time of the seismic data acquisition nodes, and reduce the manual workload.

[0082] A second exemplary embodiment of the present disclosure provides a device for detecting data quality of a seismic exploration node.

[0083] The above-mentioned device for detecting the data quality of seismic exploration nodes comprises: a first data acquisition module, a first preprocessing module, a first similarity calculation module and an anomaly detection module.

[0084] The first data acquisition module is used to acquire the time series acquisition data of the earthquake to be detected from multiple data acquisition nodes.

[0085] The first preprocessing module is used to preprocess the time series data of the earthquake to be detected, so as to obtain the time series data to be detected with the data features retained and the data volume reduced.

[0086] The first similarity calculation module is used to calculate the similarity between the time series data to be detected of multiple data acquisition nodes under different step sizes to obtain a set of multi-scale similarity matrices to be detected; the step size represents the number of adjacent data points in the time series data to be detected.

[0087] The above-mentioned anomaly detection module is used to input the above-mentioned multi-scale similarity matrix set to be detected into a pre-built time series anomaly detection model, and output a detection result of whether there are abnormal data points.

[0088] In some embodiments, the above-mentioned time series anomaly detection model is an MSCRED network model based on the attention mechanism.

[0089] According to an embodiment of the present disclosure, the above-mentioned device also includes: a time series anomaly detection model construction module, which is used to pre-construct a time series anomaly detection model.

[0090] The above-mentioned pre-built time series anomaly detection model includes: obtaining seismic time series acquisition data of multiple data acquisition nodes; preprocessing the above-mentioned seismic time series acquisition data to obtain time series data with retained data features and reduced data volume; calculating the similarity between the time series data of multiple data acquisition nodes under different step lengths to obtain a multi-scale similarity matrix set; the above-mentioned step length represents the number of adjacent data points in the above-mentioned time series data; the above-mentioned step length is the same in the model training stage and the model use stage; the first part of the above-mentioned multi-scale similarity matrix set is used as the training set data, and the second part is used as the test set data; based on the above-mentioned training set data, the time series The anomaly detection model is trained, wherein the parameters of the above-mentioned time series anomaly detection model are updated during the training process until the training reaches a set number of times or the loss function is lower than a set value, and the above-mentioned loss function represents the gap between the input multi-scale similarity matrix set and the reconstructed similarity matrix set output by the decoding module; based on the above-mentioned test set data, the trained time series anomaly detection model is tested, and the parameters of the above-mentioned time series anomaly detection model are fine-tuned during the test process, so that the prediction accuracy of the adjusted time series anomaly detection model meets the set accuracy requirement, and the adjusted time series anomaly detection model is used as the constructed time series anomaly detection model.

[0091] According to an embodiment of the present disclosure, the above-mentioned time series anomaly detection model includes: an encoding module, multiple LSTM models based on the attention mechanism, a decoding module and a comparison module. The above-mentioned encoding module includes multiple convolutional layers, and the above-mentioned multi-layer convolutional layers are used to perform layer-by-layer encoding processing on the above-mentioned multi-scale similarity matrix set to be detected to obtain a multi-layer encoding feature map. In the above-mentioned multiple LSTM models based on the attention mechanism, each LSTM model includes: an LSTM network, an attention module and an output module. The above-mentioned LSTM network is used to perform temporal context learning on each layer of the encoding feature map in the above-mentioned multi-layer encoding feature map to obtain the hidden layer vector at each moment within the preset backtracking time; the above-mentioned attention module is used to determine the attention weights corresponding to the hidden layer vector at the current moment and the hidden layer vector at the previous moment within the above-mentioned preset backtracking step length; the above-mentioned output module is used to weight and calculate the hidden layer vectors at each previous moment and the current moment within the above-mentioned preset backtracking step length and the corresponding attention weights, obtain intermediate variables and output them. The above-mentioned decoding module is used to perform decoding processing according to the above-mentioned intermediate variables and each layer of the encoding feature map in the above-mentioned multi-layer encoding feature map to obtain a reconstructed similarity matrix set. The comparison module is used to compare the gap between the reconstructed similarity matrix set and the corresponding data points in the multi-scale similarity matrix set to be detected, and if there is a target data point whose gap exceeds a set threshold, the target data point is determined to be an abnormal data point.

[0092] According to an embodiment of the present disclosure, the multi-layer convolutional layers include: m convolutional layers, where m≥3 and is a positive integer; the encoding module performs layer-by-layer encoding processing on the multi-scale similarity matrix set to be detected, including: based on the first convolutional layer, feature extraction is performed on the input multi-scale similarity matrix set to be detected to obtain a first-layer encoding feature map; based on the second convolutional layer, downsampling is performed on the first-layer encoding feature map to obtain a second-layer encoding feature map; based on the m-th convolutional layer, downsampling is performed on the m-1th layer encoding feature map to obtain an m-th layer encoding feature map; wherein the first-layer encoding feature map to the m-th layer encoding feature map are respectively input into m LSTM models based on the attention mechanism for processing to obtain corresponding m intermediate variables.

[0093] According to an embodiment of the present disclosure, in the above-mentioned decoding module, decoding processing is performed according to the above-mentioned intermediate variables and each layer of the encoding feature maps in the above-mentioned multi-layer encoding feature maps to obtain a reconstructed similarity matrix set, including: upsampling the above-mentioned m-th intermediate variable to obtain a 0th layer decoding feature map with the same dimension as the m-1th intermediate variable; splicing the 0th layer decoding feature map with the above-mentioned m-1th intermediate variable to obtain a 1st layer decoding feature map; upsampling the above-mentioned 1st layer decoding feature map to obtain a 2nd layer decoding feature map with the same dimension as the m-2th intermediate variable; splicing the 2nd layer decoding feature map with the above-mentioned m-2th intermediate variable to obtain a 3rd layer decoding feature map; and performing feature reconstruction on the last layer decoding feature map to obtain a reconstructed similarity matrix set with the same dimension as the above-mentioned multi-scale similarity matrix set to be detected.

[0094] According to an embodiment of the present disclosure, the similarity between the time series data to be detected of multiple data acquisition nodes under different step sizes is calculated to obtain a set of multi-scale similarity matrices to be detected, including: numbering the data acquisition nodes according to the spatial arrangement order; for each data acquisition node, determining the time series segment data under multiple step segments of the time series data to be detected corresponding to each step size; for the time series segment data in each step segment of each data acquisition node corresponding to each step size, performing the following steps to obtain a triangular first similarity matrix corresponding to each step segment under each step size: calculating the first similarity between the current data acquisition node and its own time series segment data in the current step segment, the first similarity being the diagonal element of the first similarity matrix at time t, the row number and column number of the diagonal element being equal to the number of the current data acquisition node; time t being the last number in the current step segment According to the corresponding moment; calculate the second similarity between the time segment data of the current data acquisition node and the subsequent data acquisition node under the current step segment, the above second similarity is used as the non-diagonal element of the above first similarity matrix, the row number of the above non-diagonal element is equal to the number of the above current data acquisition node, and the column number of the above non-diagonal element is equal to the number of the above subsequent data acquisition node; the number of the above subsequent data acquisition node is greater than the number of the above current data acquisition node; based on the matrix symmetry algorithm, the above first similarity matrix in triangular form is processed to obtain the second similarity matrix corresponding to each of the multiple step segments under each step, and the above second similarity matrix is ​​a symmetric matrix; the above second similarity matrix is ​​combined according to the two dimensions of the time sequence of the step period and the step size to obtain a set of multi-scale similarity matrices to be detected, and the above multi-scale similarity matrix set to be detected is in the form of a tensor.

[0095] According to an embodiment of the present disclosure, the above-mentioned seismic time series acquisition data to be detected is preprocessed to obtain the time series data to be detected with the data features retained and the data volume reduced, including: determining the target neighborhood range corresponding to the working condition of the above-mentioned data acquisition node according to the preset correspondence between the exploration condition and the neighborhood range; dividing the above-mentioned seismic time series acquisition data to be detected according to the above-mentioned target neighborhood range, taking the mean value of the data divided into the same target neighborhood range, and using each mean value result as the data value in the corresponding target neighborhood range; arranging the mean value results in each target neighborhood range in chronological order to obtain the above-mentioned time series data to be detected.

[0096] It is understandable that more details of this embodiment can refer to the content of the first embodiment, which will not be repeated here.

[0097] The third exemplary embodiment of the present disclosure provides a device for constructing a time series anomaly detection model. The time series anomaly detection model is used for seismic exploration node data quality detection.

[0098] The above-mentioned device for constructing a time series anomaly detection model includes: a second data acquisition module, a second preprocessing module, a second similarity calculation module, a data partitioning module, a training module and a testing module.

[0099] The second data acquisition module is used to acquire seismic time series acquisition data from multiple data acquisition nodes.

[0100] The second preprocessing module is used to preprocess the seismic time series acquisition data to obtain time series data with the data features retained and the data volume reduced.

[0101] The above-mentioned second similarity calculation module is used to calculate the similarity between the time series data of multiple data acquisition nodes under different step sizes to obtain a multi-scale similarity matrix set; the above-mentioned step size represents the number of adjacent data points in the above-mentioned time series data; the above-mentioned step size is the same in the model training stage and the model usage stage.

[0102] The data partitioning module is used to use the first part of the multi-scale similarity matrix set as training set data and the second part as test set data.

[0103] The above-mentioned training module is used to use the above-mentioned training set data to train the time series anomaly detection model, wherein the parameters of the above-mentioned time series anomaly detection model are updated during the training process until the training reaches a set number of times or the loss function is lower than a set value, and the above-mentioned loss function represents the gap between the input multi-scale similarity matrix set and the reconstructed similarity matrix set output by the decoding module.

[0104] The above-mentioned testing module is used to test the trained time series anomaly detection model using the above-mentioned test set data, and fine-tune the parameters of the above-mentioned time series anomaly detection model during the test process so that the prediction accuracy of the adjusted time series anomaly detection model meets the set accuracy requirements.

[0105] The trained time series anomaly detection model is used as the constructed time series anomaly detection model; or the adjusted time series anomaly detection model is used as the constructed time series anomaly detection model.

[0106] For other implementation details of this embodiment, reference may be made to the contents of the first embodiment, which will not be repeated here.

[0107] Any multiple of the functional modules included in the above-mentioned device for detecting data quality of seismic exploration nodes or the device for building a time series anomaly detection model can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. At least one of the functional modules included in the device for detecting data quality of seismic exploration nodes or the device for building a time series anomaly detection model can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in any appropriate combination of any of them. Alternatively, at least one of the functional modules included in the device for detecting data quality of seismic exploration nodes or the device for building a time series anomaly detection model can be at least partially implemented as a computer program module, and when the computer program module is run, the corresponding function can be executed.

[0108] A fourth exemplary embodiment of the present disclosure provides an electronic device.

[0109] Figure 7 The structural block diagram of the electronic device provided by the embodiment of the present disclosure is schematically shown.

[0110] Reference Figure 7 As shown, the electronic device 700 provided by the embodiment of the present disclosure includes a processor 701, a communication interface 702, a memory 703 and a communication bus 704, wherein the processor 701, the communication interface 702 and the memory 703 communicate with each other through the communication bus 704; the memory 703 is used to store computer programs; the processor 701 is used to implement the method for seismic exploration node data quality detection as described above when executing the program stored in the memory.

[0111] The fifth exemplary embodiment of the present disclosure further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for detecting the quality of seismic exploration node data as described above is implemented.

[0112] The computer-readable storage medium may be included in the device or apparatus described in the above embodiment; or it may exist independently without being assembled into the device or apparatus. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0113] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0114] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0115] The foregoing is merely a specific embodiment of the present disclosure, which enables those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features applied herein.

Claims

1. A method for detecting the quality of seismic exploration node data, characterized in that: include: Acquire time series acquisition data of earthquakes to be detected from multiple data acquisition nodes; Preprocessing the seismic time series data to be detected to obtain the time series data to be detected with the data features retained and the data volume reduced; Calculate the similarity between the time series data to be detected of multiple data collection nodes under different step lengths to obtain a set of multi-scale similarity matrices to be detected; The step size represents the number of adjacent data points in the time series data to be detected; The multi-scale similarity matrix set to be detected is input into a pre-built time series anomaly detection model, and a detection result of whether there is an abnormal data point is obtained as an output.

2. The method according to claim 1, characterized in that The temporal anomaly detection model is a multi-scale convolutional recursive encoder and decoder MSCRED network model based on an attention mechanism; the temporal anomaly detection model includes: The encoding module comprises a plurality of convolutional layers, wherein the plurality of convolutional layers are used to perform layer-by-layer encoding processing on the set of multi-scale similarity matrices to be detected to obtain a multi-layer encoding feature map; A plurality of LSTM models based on the attention mechanism, each LSTM model comprises: an LSTM network, an attention module and an output module, wherein the LSTM network is used to perform temporal context learning on each layer of the coding feature map in the multi-layer coding feature map to obtain the hidden layer vector at each moment within the preset backtracking time length; the attention module is used to determine the attention weights corresponding to the hidden layer vector at the current moment and the hidden layer vector at the previous moment within the preset backtracking step length; the output module is used to weight and calculate the hidden layer vectors at each previous moment and the current moment within the preset backtracking step length and the corresponding attention weights, obtain the intermediate variable and output it; A decoding module, used for performing decoding processing according to the intermediate variable and each layer of the coding feature map in the multi-layer coding feature map to obtain a reconstructed similarity matrix set; A comparison module is used to compare the gap between the reconstructed similarity matrix set and the corresponding data points in the multi-scale similarity matrix set to be detected, and in the case where there is a target data point whose gap exceeds a set threshold, determine that the target data point is an abnormal data point.

3. The method according to claim 2, characterized in that The multi-layer convolutional layer comprises: m convolutional layers, where m≥3 and is a positive integer; The encoding module performs layer-by-layer encoding processing on the set of multi-scale similarity matrices to be detected, including: Based on the first convolutional layer, feature extraction is performed on the input multi-scale similarity matrix set to be detected to obtain a first-layer encoding feature map; Based on the second convolutional layer, downsampling the first-layer encoding feature map is performed to obtain a second-layer encoding feature map; Based on the m-th convolutional layer, downsample the m-1-th layer encoding feature map to obtain the m-th layer encoding feature map; Among them, the first layer encoding feature map to the mth layer encoding feature map are respectively input into m LSTM models based on the attention mechanism for processing to obtain corresponding m intermediate variables.

4. The method according to claim 3, characterized in that In the decoding module, decoding processing is performed according to the intermediate variable and each layer of the multi-layer coding feature map to obtain a reconstructed similarity matrix set, including: Upsampling the mth intermediate variable to obtain a 0th layer decoding feature map having the same dimension as the m-1th intermediate variable; Concatenate the 0th layer decoding feature map with the m-1th intermediate variable to obtain the 1st layer decoding feature map; Upsampling the first layer decoding feature map to obtain a second layer decoding feature map having the same dimension as the m-2th intermediate variable; Concatenate the second layer decoding feature map with the m-2th intermediate variable to obtain the third layer decoding feature map; Feature reconstruction is performed on the last layer of decoded feature maps to obtain a reconstructed similarity matrix set having the same dimension as the multi-scale similarity matrix set to be detected.

5. The method according to claim 1, characterized in that Calculate the similarity between the time series data to be detected of multiple data collection nodes under different step lengths to obtain a set of multi-scale similarity matrices to be detected, including: Number the data collection nodes according to the spatial arrangement order; For each data collection node, determine the time series segment data under multiple step length segments of the time series data to be detected corresponding to each step length; For the time series segment data in each step length segment of each data collection node corresponding to each step length, the following steps are performed to obtain a first similarity matrix of a triangular form corresponding to each step length segment under each step length: Calculate the first similarity between the current data acquisition node and its own time series segment data in the current step segment, the first similarity is used as the diagonal element of the first similarity matrix at time t, the row number and column number of the diagonal element are equal to the number of the current data acquisition node; time t is the time corresponding to the last data in the current step segment; Calculating a second similarity between the time series segment data of the current data acquisition node and the subsequent data acquisition node under the current step length segment, the second similarity is used as a non-diagonal element of the first similarity matrix, the row number of the non-diagonal element is equal to the number of the current data acquisition node, and the column number of the non-diagonal element is equal to the number of the subsequent data acquisition node; the number of the subsequent data acquisition node is greater than the number of the current data acquisition node; Based on a matrix symmetry algorithm, the first similarity matrix in triangular form is processed to obtain second similarity matrices corresponding to a plurality of step length segments under each step length, wherein the second similarity matrix is ​​a symmetric matrix; The second similarity matrix is ​​combined according to the time sequence of the step period and the step size to obtain a set of multi-scale similarity matrices to be detected, and the set of multi-scale similarity matrices to be detected is in a tensor form.

6. The method according to claim 1, characterized in that Preprocessing the seismic time series data to be detected to obtain the time series data to be detected with the data features retained and the data volume reduced, including: Determine the target neighborhood range corresponding to the working condition of the data acquisition node according to the preset corresponding relationship between the exploration working condition and the neighborhood range; Dividing the seismic time series data to be detected according to the target neighborhood range, performing a mean operation on the data divided into the same target neighborhood range, and using each mean result obtained as the data value in the corresponding target neighborhood range; The mean results within each target neighborhood are arranged in chronological order to obtain the time series data to be detected.

7. The method according to any one of claims 1 to 6, characterized in that Also includes: Pre-built time series anomaly detection models; The pre-built time series anomaly detection model includes: Acquire seismic time series acquisition data of multiple data acquisition nodes; Preprocessing the seismic time series acquisition data to obtain time series data with preserved data features and reduced data volume; Calculate the similarity between the time series data of multiple data collection nodes under different step lengths to obtain a multi-scale similarity matrix set; the step length represents the number of adjacent data points in the time series data; the step length is the same in the model training stage and the model use stage; Using the first part of the multi-scale similarity matrix set as training set data, and the second part as test set data; Based on the training set data, the time series anomaly detection model is trained, wherein the parameters of the time series anomaly detection model are updated during the training process until the training reaches a set number of times or the loss function is lower than a set value, wherein the loss function represents the gap between the input multi-scale similarity matrix set and the reconstructed similarity matrix set output by the decoding module; Based on the test set data, the trained time series anomaly detection model is tested. During the test, the parameters of the time series anomaly detection model are fine-tuned so that the prediction accuracy of the adjusted time series anomaly detection model meets the set accuracy requirement. The adjusted time series anomaly detection model is used as the constructed time series anomaly detection model.

8. A device for detecting the quality of seismic exploration node data, characterized in that: include: A first data acquisition module is used to acquire the time series acquisition data of the earthquake to be detected from multiple data acquisition nodes; A first preprocessing module is used to preprocess the seismic time series data to be detected to obtain the time series data to be detected with the data features retained and the data volume reduced; The first similarity calculation module is used to calculate the similarity between the time series data to be detected of multiple data collection nodes under different step lengths to obtain a set of multi-scale similarity matrices to be detected; The step size represents the number of adjacent data points in the time series data to be detected; The anomaly detection module is used to input the multi-scale similarity matrix set to be detected into a pre-built time series anomaly detection model, and output a detection result of whether there are abnormal data points.

9. A device for constructing a time series anomaly detection model, characterized in that: The time series anomaly detection model is used for seismic exploration node data quality detection, and the device includes: A second data acquisition module is used to acquire seismic time series acquisition data of multiple data acquisition nodes; A second preprocessing module is used to preprocess the seismic time series acquisition data to obtain time series data with preserved data features and reduced data volume; A second similarity calculation module is used to calculate the similarity between the time series data of multiple data collection nodes under different step lengths to obtain a multi-scale similarity matrix set; the step length represents the number of adjacent data points in the time series data; the step length is the same in the model training stage and the model use stage; A data partitioning module, used to use the first part of the multi-scale similarity matrix set as training set data and the second part as test set data; A training module, used to train the time series anomaly detection model using the training set data, wherein the parameters of the time series anomaly detection model are updated during the training process until the training reaches a set number of times or the loss function is lower than a set value, wherein the loss function represents the gap between the input multi-scale similarity matrix set and the reconstructed similarity matrix set output by the decoding module; A testing module, used to test the trained time series anomaly detection model using the test set data, and to fine-tune the parameters of the time series anomaly detection model during the test so that the prediction accuracy of the adjusted time series anomaly detection model meets the set accuracy requirement; The trained time series anomaly detection model is used as the constructed time series anomaly detection model; or the adjusted time series anomaly detection model is used as the constructed time series anomaly detection model.

10. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; A processor, for implementing the method according to any one of claims 1 to 7 when executing a program stored in a memory.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images