Seismic data static correction processing method, device, computer equipment and storage medium

Abstract

The present invention provides a method, apparatus, computer device, and storage medium for seismic data static correction processing. The method includes obtaining target seismic data; inputting the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data including first arrival time data; inputting the first arrival picking data including the first arrival time data into a static correction processing neural network for training to obtain static correction processed first arrival picking data and second arrival time data; calculating a static correction amount according to the second arrival time data and the first arrival time data after static correction processing; and correcting the seismic data according to the static correction amount. Automatic first arrival picking and direct static correction calculation are achieved. This method does not require manual picking of first arrival data, avoids the complex near-surface modeling process, and realizes an efficient and accurate static correction processing function.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic data processing, and particularly relates to a method, device, computer device, and storage medium for static correction processing of seismic data. Background Art

[0002] Static correction is an important link in seismic data processing. Whether the static correction processing is accurate directly affects the effects of a series of subsequent processes. The currently widely used static correction processing method is to perform near-surface velocity tomography modeling processing using first arrival data to obtain near-surface velocity model data, and calculate the travel time difference caused by the undulating surface using the velocity model for static correction. Accurate first arrival data and near-surface velocity model data are required. The first arrival picking process is time-consuming and laborious, the near-surface velocity modeling process is complex, and obtaining an accurate near-surface velocity model is a difficult process. Facing the seismic data in complex surface exploration areas such as mountains, it is very difficult to obtain accurate first arrival data and near-surface velocity models, thus affecting the effect of static correction. There are many static correction processing methods for seismic data in complex surfaces, mainly concentrated on first arrival picking and high-precision near-surface velocity modeling. For example, automatic first arrival picking methods and tomographic near-surface velocity modeling methods have been widely studied and applied, and certain processing effects have been achieved, but the static correction problems in complex surfaces such as mountains still cannot be completely solved. How to obtain an accurate static correction processing effect is a difficult problem faced by seismic exploration in mountainous and complex areas. Summary of the Invention

[0003] Based on this, in view of the above technical problems, it is necessary to provide a method, device, computer device, and storage medium for static correction processing of seismic data.

[0004] A method for static correction processing of seismic data includes:

[0005] Obtain target seismic data;

[0006] Input the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data including first arrival time data;

[0007] Input the first arrival picking data including the first arrival time data into a static correction processing neural network for training to obtain static correction processed first arrival picking data and second arrival time data;

[0008] Calculate a static correction amount according to the second arrival time data and the first arrival time data after static correction processing;

[0009] Correct the seismic data according to the static correction amount.

[0010] In one embodiment, before the step of inputting the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data including first arrival times:

[0011] Obtain sample data in a preset format, where the sample data in the preset format is each trace data of a shot gather data, and the format of the seismic trace data includes a file header, trace header data for each trace, and a trace data body, and the trace data body records the amplitude value at each sampling point;

[0012] Input the sample data in the preset format into an initial picking neural network for training to obtain the initial picking neural network model.

[0013] In one embodiment, before the step of obtaining sample data in a preset format, further included is:

[0014] Construct an initial picking neural network including a first input layer, a first intermediate layer, and a first output layer, where the first input layer is used to input one trace of seismic data, and after calculation by the first intermediate layer, the first output layer outputs a data with the same number of sample points as the input seismic trace data, and each sample point value in the output data is a first sample point value or a second sample point value.

[0015] In one embodiment, the step of inputting the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data including first arrival time data includes:

[0016] Input the target seismic data into a pre-trained initial picking neural network model for training, and the first arrival picking neural network model outputs the first arrival picking data including sample point values, where the sample point values include a first sample point value and a second sample point value;

[0017] Extract, through the first arrival picking neural network model, the first arrival picking data with the sample point value being the first sample point value, and obtain the first arrival time data of the first arrival picking data corresponding to the first sample point value.

[0018] In one embodiment, the step of inputting the first arrival picking data including the first arrival time data into a static correction processing neural network for training to obtain static correction processed first arrival picking data and second arrival time data includes:

[0019] Input the first arrival picking data including the first arrival time data into a static correction processing neural network for training;

[0020] Through the static correction processing application network, reset the elevation of static correction for the target seismic data that needs static correction processing, and output the first arrival time at the target elevation.

[0021] In one embodiment, before the step of inputting the first arrival picking data including the first arrival time data into the static correction processing neural network for training to obtain the first arrival picking data after static correction processing and the second arrival time data, the following steps are further included:

[0022] Construct a static correction processing neural network including a second input layer, a second intermediate layer, and a second output layer.

[0023] In one embodiment, the step of calculating the static correction amount according to the second arrival time data and the first arrival time data after static correction processing includes:

[0024] Calculate the difference between the second arrival time data and the first arrival time data after static correction processing to obtain the static correction amount.

[0025] A computer device includes a memory and a processor. The memory stores a computer program. It is characterized in that when the processor executes the computer program, the following steps are implemented:

[0026] Obtain target seismic data;

[0027] Input the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data including first arrival time data;

[0028] Input the first arrival picking data including the first arrival time data into the static correction processing neural network for training to obtain the first arrival picking data after static correction processing and the second arrival time data;

[0029] Calculate the static correction amount according to the second arrival time data and the first arrival time data after static correction processing;

[0030] Correct the seismic data according to the static correction amount.

[0031] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the following steps are implemented:

[0032] Obtain target seismic data;

[0033] Input the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data including first arrival time data;

[0034] Input the first arrival picking data including the first arrival time data into the static correction processing neural network for training to obtain the first arrival picking data after static correction processing and the second arrival time data;

[0035] Calculate the static correction amount based on the second first arrival time data and the first first arrival time data after static correction processing;

[0036] Correct the seismic data according to the static correction amount.

[0037] The above seismic data static correction processing method, device, computer equipment and storage medium achieve automatic first arrival picking and direct static correction calculation. This method eliminates the need for manual first arrival data picking, avoids the complex near-surface modeling process, and realizes efficient and accurate static correction processing functions. Description of the Drawings

[0038] Figure 1 It is a schematic diagram of the application scenario of the seismic data static correction processing method in an embodiment;

[0039] Figure 2 It is a schematic flow diagram of the seismic data static correction processing method in an embodiment;

[0040] Figure 3 It is a structural block diagram of the seismic data static correction processing device in an embodiment;

[0041] Figure 4 It is an internal structure diagram of the computer equipment in an embodiment;

[0042] Figure 5 It is a schematic diagram of the implementation process of the seismic data static correction processing method in an embodiment;

[0043] Figure 6 It is an effect diagram of first arrival picking in an embodiment;

[0044] Figure 7A It is a schematic diagram of first arrival picking before static correction processing in an embodiment;

[0045] Figure 7B It is a schematic diagram of first arrival picking after static correction processing in an embodiment. Detailed Embodiments

[0046] In order to make the objectives, technical solutions and advantages of the present application clearer, the following further elaborates on the present application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] Embodiment 1

[0048] The seismic data static correction processing method provided by the present application can be applied, for example, as Figure 1In the application environment shown. Among them, the computer 102 communicates with the server 104 through the network. Among them, the terminal 102 can be but is not limited to various personal computers, servers, laptops, smartphones, tablets, and portable wearable devices, and the server 104 can be implemented by an independent server or a server cluster composed of multiple servers. The user sends the target seismic data to the server 104 through the terminal 102, and the server 104 obtains the target seismic data; inputs the target seismic data into a pre-trained initial picking neural network model for training to obtain the first arrival picking data containing the first arrival time data; inputs the first arrival picking data containing the first arrival time data into a static correction processing neural network for training to obtain the first arrival picking data after static correction processing and the second arrival time data; calculates the static correction amount according to the second arrival time data and the first arrival time data after static correction processing; corrects the seismic data according to the static correction amount.

[0049] Embodiment 2

[0050] In this embodiment, as Figure 2 shown, a method for static correction processing of seismic data is provided, which includes:

[0051] Step 210, obtain target seismic data.

[0052] Step 220, input the target seismic data into a pre-trained initial picking neural network model for training to obtain the first arrival picking data containing the first arrival time data.

[0053] Specifically, extract the data volume of the sample seismic data from the target seismic data, and remove the file header description information of the sample seismic data and the trace header information of each trace. Verify that the obtained data volume is generated with the same size as the seismic data volume, and the value of each sampling point in the data volume is set to 0, and then the value at the sampling point corresponding to the first arrival time of each trace is set to 1, indicating the first arrival time position of this trace of data.

[0054] Extract one trace of data from the obtained seismic data volume at a time, and use the amplitude value of each sampling point of one trace as the input data and input it into each node of the initial picking neural network model, and use the trace data corresponding to the meaning of the first arrival time of the input trace as the output test data. When the seismic data is being trained, it is not input in sequence, but one trace of data is input at a certain interval, and the cycle is repeated in turn until the training of all input data is finally completed.

[0055] Specifically, the seismic trace data volume is extracted from the seismic data that needs to be picked for the first arrival, and is input into the trace initial picking neural network model in the order of traces. After being calculated by the initial picking neural network model, the sampling positions of the seismic traces corresponding to the nodes with the node value of 1 in the output layer are used as the first arrival times of the target traces. All the seismic data that need to be picked for the first arrival are sequentially input into the trace neural network, and finally the first arrival times of all the data are obtained.

[0056] By inputting a seismic trace data into the initial picking neural network model, after being calculated by the initial picking neural network model, the output layer outputs a data with the same number of sample points as the input seismic trace data, and each sample point value in the data is 1 or 0. 1 indicates that the sample point is the first arrival time point, and 0 indicates that the sample point is not the first arrival time point. In this embodiment, the first arrival time point corresponding to the sampling point is the first arrival time data, and moreover, the first arrival time data is the first arrival time obtained by actually picking the elevation.

[0057] Step 230: Input the first arrival picking data containing the first arrival time data into the static correction processing neural network for training to obtain the first arrival picking data after static correction processing and the second arrival time data.

[0058] The goal of the static correction processing neural network is to establish the relationship between the surface elevation and the first arrival time. The neural network structure is divided into an input layer, an intermediate layer, and an output layer. The input layer is the coordinates and elevation of the shot point position and the coordinates and elevation of the geophone point position of the seismic trace data. Therefore, six nodes are set in the input layer, respectively corresponding to the shot point coordinates and elevation (Sx, Sy, Sz) and the geophone point coordinates and elevation (Rx, Ry, Rz) of each trace data. Two layers are set in the intermediate layer, and the number of nodes in each layer is 50. One node in the input layer outputs the first arrival time. The neural network is a fully connected network.

[0059] In this embodiment, the shot point coordinates and elevation (Sx, Sy, Sz) and the geophone point coordinates and elevation (Rx, Ry, Rz) of each seismic trace are extracted, and the six extracted parameters are input into the input layer of the neural network, and the first arrival time of the input trace is used as the verification data of the output layer. The neural network training is sequentially performed on each trace of the target seismic data.

[0060] Step 240: Calculate the static correction amount according to the second arrival time data and the first arrival time data after static correction processing.

[0061] In one embodiment, the step of calculating the static correction amount according to the second arrival time data and the first arrival time data after static correction processing includes: calculating the difference between the second arrival time data and the first arrival time data after static correction processing to obtain the static correction amount.

[0062] After selecting the completed neural network, the relationship between each elevation data and the first arrival time is established. The elevation of the static correction for the target seismic data that needs to be static corrected is reset and input into the neural network. The value output after the neural network calculation is the first arrival time of the target elevation. The difference obtained by subtracting the first arrival time picked from the actual elevation from the first arrival time of the target elevation is the static correction amount of the target elevation.

[0063] Step 250, correct the seismic data according to the static correction amount.

[0064] Perform the above operations on each trace of data to obtain the static correction of the target elevation and complete the static correction processing.

[0065] In the above embodiments, automatic first arrival picking and direct static correction calculation are realized. This method does not require manual picking of first arrival data, avoids the complex near-surface modeling process, and realizes the function of efficient and accurate static correction processing.

[0066] In one embodiment, before the step of inputting the target seismic data into a pre-trained initial picking neural network model for training to obtain the first arrival picking data including the first arrival time, it includes:

[0067] Obtain sample data in a preset format, where the sample data in the preset format is each trace of the shot gather data. The format of the seismic trace data includes a file header, the trace header data of each trace, and a trace data body, and the trace data body records the amplitude values at each sampling point; input the sample data in the preset format into the initial picking neural network for training to obtain the initial picking neural network model.

[0068] Specifically, after the deep neural network structure design is completed, it is necessary to train the neural network model with sample data to obtain the initial picking neural network model. The sample data uses the accurate first arrival time data obtained by manual picking. The input sample is each trace data of the shot gather data. The format of the seismic trace data includes the file header, the trace header data of each trace, and the trace data body. The trace data body records the amplitude value at each sampling point. The neural network input data only needs the trace data body. Therefore, it is necessary to reconstruct the input seismic data, strip the file header description data and the trace header data of each trace, and retain the trace data body of each trace to form the sample data of the pure data body. The output sample data is constructed based on the input sample data and the first arrival time. According to the size of the input sample data, an empty data body of the same size is generated, and the value of each sampling point in the data body is 0. According to the first arrival time, the value of the corresponding sampling point of each trace data in the output sample data is set to 1 to obtain the output sample data. The training process of the neural network is as follows: The generated input sample data is input one trace at a time in the trace order into the output layer of the input trace neural network. Each sampling point of the input trace data corresponds to a node of the input layer. The output data is the output sample data of the corresponding trace. Each sampling point of each output sample corresponds to a node of the output layer.

[0069] In one embodiment, before the step of obtaining the sample data in the preset format, it further includes:

[0070] Construct an initial picking neural network including a first input layer, a first intermediate layer, and a first output layer, wherein the first input layer is used to input one trace of seismic data. After being calculated by the first intermediate layer, the first output layer outputs one trace of data with the same number of sample points as the input seismic trace data, and the value of each sample point in the output data is the first sample value or the second sample value.

[0071] In this embodiment, according to the characteristics of the seismic data, the initial picking neural network is provided with an input layer, an intermediate layer, and an output layer. The node data of the input layer is the same as the number of sampling points of one trace of seismic data and is used to input one trace of seismic data. The intermediate layer has two layers, and the number of nodes in each layer is the same as that of the input layer. The number of nodes in the output layer is the same as the number of nodes in the input layer. The network is a fully connected network. The input layer inputs one trace of seismic data. After being calculated by the intermediate layer, the output layer outputs one trace of data with the same number of sample points as the input seismic trace data, and the value of each sample point in the data is 1 or 0. 1 indicates that the sample point is the first arrival time point, and 0 indicates that the sample point is a non-first arrival time point.

[0072] Specifically, first, set up the first-arrival picking neural network structure. Design the neural network structure according to one input layer, two intermediate layers, and one output layer. The node data of the input layer is consistent with the number of sampling points of the seismic data for which first-arrival picking is required. The number of nodes in the two intermediate layers and the node data of the output layer are consistent with the node data of the input layer. Subsequently, generate training sample data. Select some of the seismic data that requires first-arrival picking as training sample data. First, manually pick the first arrivals of the training sample seismic data to obtain accurate first-arrival times. According to the neural network structure, extract the data volume of the sample seismic data, and remove the file header description information and the trace header information of each trace of the sample seismic data. The acquisition of the verification data volume is to generate a data volume of the same size as the seismic data volume, and the value of each sampling point in the data volume is set to 0. Then, the value at the sampling point corresponding to the first-arrival time of each trace is set to 1, indicating the position of the first-arrival time of this trace of data. Then, determine the training parameters. The training parameters of the neural network are the key factors determining the training effect. Considering the computational amount and accuracy of the training, control the neural network training through two parameters: the number of loops and the error amount. The training error determines the training accuracy and prevents overfitting. The number of loops controls the computational amount of the training and prevents getting stuck in multiple loops and unable to end normally.

[0073] In one embodiment, the step of inputting the target seismic data into a pre-trained initial picking neural network model for training to obtain first-arrival picking data including first first-arrival time data includes:

[0074] Input the target seismic data into a pre-trained initial picking neural network model for training. The first-arrival picking neural network model outputs the first-arrival picking data including sample values, where the sample values include first sample values and second sample values; extract, through the first-arrival picking neural network model, the first-arrival picking data with the sample value being the first sample value, and obtain the first first-arrival time data of the first-arrival picking data corresponding to the first sample value.

[0075] In this embodiment, using the trained neural network model, generate input data for the target seismic data that requires first-arrival picking processing according to the requirements of the input samples, input it into the trained neural network. After the neural network calculates, extract the sample points with a sampling value of 1 in the output trace, and obtain the position time where the sample point with a value of 1 is located as the first-arrival picking result of this trace.

[0076] In one embodiment, the step of inputting the first-arrival picking data including the first first-arrival time data into a static correction processing neural network for training to obtain the first-arrival picking data after static correction processing and second first-arrival time data includes:

[0077] Input the first arrival picking data containing the first arrival time data into the static correction processing neural network for training; through the static correction processing application network, reset the elevation of the static correction for the target seismic data that needs static correction processing, and output the first arrival time at the target elevation.

[0078] Specifically, for the seismic data with picked first arrivals, extract the shot point coordinates and elevation (Sx, Sy, Sz), and the geophone point coordinates and elevation (Rx, Ry, Rz) for each trace. Input the six data for each obtained trace into the six nodes of the input layer of the neural network, and the output verification data is the input arrival time. Extract the training samples according to the above processing process for all the seismic traces with picked first arrivals and read them into the neural network for training.

[0079] In one embodiment, before the step of inputting the first arrival picking data containing the first arrival time data into the static correction processing neural network for training to obtain the first arrival picking data after static correction processing and the second arrival time data, it further includes:

[0080] Construct a static correction processing neural network including a second input layer, a second intermediate layer, and a second output layer.

[0081] Specifically, the goal of the static correction processing neural network is to establish the relationship between the surface elevation and the first arrival time. The neural network structure is divided into an input layer, an intermediate layer, and an output layer. The input layer is the shot point position coordinates and elevation, and the geophone point position coordinates and elevation of the seismic trace data. Therefore, six nodes are set in the input layer, corresponding to the shot point coordinates and elevation (Sx, Sy, Sz), and the geophone point coordinates and elevation (Rx, Ry, Rz) of each trace data respectively. Two layers are set in the intermediate layer, with 50 nodes in each layer. One node in the input layer outputs the first arrival time. The neural network is a fully connected network.

[0082] In the above embodiment, using a deep neural network, automatically pick the first arrivals of seismic data. After obtaining the first arrival data, use the deep neural network to process the first arrival data to obtain the relationship between the first arrival time and the surface elevation. Utilize the relationship between the first arrival time and the surface elevation, and by setting different target surface elevations, calculate the first arrival time at the corresponding elevation, realizing direct static correction processing.

[0083] It should be understood that although Figure 2 the steps in the flowchart Figure 2At least a part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed and completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in rotation with at least a part of other steps or sub-steps or stages of other steps.

[0084] Embodiment III

[0085] The present invention uses a deep neural network to automatically pick the first arrivals of seismic data. After obtaining the first arrival data, the deep neural network is used to process the first arrival data to obtain the relationship between the first arrival time and the surface elevation. Using the relationship between the first arrival time and the surface elevation, by setting different target surface elevations, the first arrival times corresponding to the respective elevations are calculated, thereby realizing direct static correction processing. The process of direct static correction processing is as follows: a unified surface elevation is set, the deep neural network is used to calculate the first arrival time of the unified elevation, the first arrival time of the unified elevation is subtracted from the actually picked first arrival time to obtain the time correction amount for the target elevation, and the static correction processing is directly performed using the time correction amount.

[0086] In this embodiment, the process of seismic data static correction processing is as follows:

[0087] (1) First arrival picking neural network structure

[0088] According to the characteristics of the seismic data, the initial picking neural network is provided with an input layer, an intermediate layer, and an output layer. The node data of the input layer is the same as the number of sampling points of the seismic data, and is used to input a trace of seismic data. The intermediate layer has two layers, and the number of nodes in each layer is the same as that of the input layer. The number of nodes in the output layer is the same as the number of nodes in the input layer. The network is a fully connected network. The input layer inputs a trace of seismic data. After being calculated by the intermediate layer, the output layer outputs a data with the same number of sample points as the input seismic trace data. Each sample point value in the data is 1 or 0. 1 indicates that the sample point is the first arrival time point, and 0 indicates that the sample point is not the first arrival time point.

[0089] (2) Obtaining of first arrival picking sample data

[0090] After the deep neural network structure is designed, it is necessary to train the neural network model with sample data. The sample data uses the accurate first arrival time data obtained by manual picking. The input sample is each trace data of the shot gather data. The format of the seismic trace data includes the file header, the trace header data of each trace, and the trace data body. The trace data body records the amplitude value at each sampling point. The neural network input data only needs the trace data body. Therefore, it is necessary to reconstruct the input seismic data, strip the file header description data and the trace header data of each trace, and retain the trace data body of each trace to form the sample data of the pure data body. The output sample data is constructed based on the input sample data and the first arrival time. According to the size of the input sample data, an empty data body of the same size is generated, and the value of each sampling point in the data body is 0. According to the first arrival time, the value of the corresponding sampling point of each trace data in the output sample data is set to 1 to obtain the output sample data. The training process of the neural network is as follows: The generated input sample data is input one trace at a time in the trace order into the output layer of the neural network. Each sampling point of the input trace data corresponds to a node in the input layer. The output data is the output sample data of the corresponding trace. Each sampling point of each output sample corresponds to a node in the output layer.

[0091] (3) Acquisition of first arrival picking data

[0092] Using the trained neural network model, the target seismic data that needs to be processed for first arrival picking is generated into input data according to the requirements of the input sample, and is input into the trained neural network. For one trace data output by the neural network calculation, the sample points with a sampling value of 1 in the output trace are extracted, and the position time where the sample points with a value of 1 are located is obtained as the first arrival picking result of this trace.

[0093] (4) Neural network structure setting for static correction processing

[0094] The goal of the neural network for static correction processing is to establish the relationship between the surface elevation and the first arrival time. The neural network structure is divided into an input layer, an intermediate layer, and an output layer. The input layer is the shot point position coordinates and elevation of the seismic trace data, as well as the coordinates and elevation of the geophone point position. Therefore, six nodes are set in the input layer, corresponding to the shot point coordinates and elevation (Sx, Sy, Sz) and the geophone point coordinates and elevation (Rx, Ry, Rz) of each trace data respectively. Two layers are set in the intermediate layer, and the number of nodes in each layer is 50. One node in the input layer outputs the first arrival time. The neural network is a fully connected network.

[0095] (5) Generation of training samples

[0096] Extract the shot point coordinates and elevation (Sx, Sy, Sz), as well as the geophone point coordinates and elevation (Rx, Ry, Rz) for each trace of the seismic data for which the first arrivals have been picked. Input the six data for each trace obtained into the six nodes of the input layer of the neural network, and the output verification data is the input initial time. Extract the training samples for all the seismic traces with picked first arrivals according to the above processing process and read them into the neural network for training.

[0097] (6) Static correction processing

[0098] Select the completed neural network to establish the relationship between the elevation data of each trace and the first arrival time. Reset the elevation of the static correction for the target seismic data that needs static correction processing and input it into the neural network. The value output after the neural network calculation is the first arrival time of the target elevation. Subtract the first arrival time picked at the actual elevation from the first arrival time of the target elevation, and the obtained difference is the static correction amount of the target elevation. Perform the above operations on each trace of data to obtain the static correction of the target elevation and complete the static correction processing.

[0099] The present invention provides a seismic data static correction method based on a deep neural network, which realizes direct static correction processing of seismic data, avoids processing processes such as first arrival picking and near-surface velocity modeling, meets the requirements of static correction processing of seismic data with complex surface conditions, improves the processing effect of seismic data, reduces exploration costs, and improves economic benefits.

[0100] Example 4

[0101] Please combine Figure 5 , Step 1, setting the neural network structure for first arrival picking. Design the neural network structure according to one input layer, two intermediate layers, and one output layer. The number of node data in the input layer is the same as the number of sampling points of the seismic data for which first arrival picking is required. The number of nodes in the two intermediate layers and the number of node data in the output layer are the same as the number of node data in the input layer.

[0102] Step 2, generating training sample data. Select some seismic data that needs first arrival picking as training sample data. First, manually pick the first arrivals of the training sample seismic data to obtain accurate first arrival times. According to the neural network structure, extract the data volume of the sample seismic data and remove the file header description information and the trace header information of each trace of the sample seismic data. The acquisition of the verification data volume is to generate a data volume of the same size as the seismic data volume, set the value of each sampling point in the data volume to 0, and then set the value at the sampling point position corresponding to the first arrival time of each trace to 1, indicating the first arrival time position of this trace of data.

[0103] Step 3: Determine the training parameters. The training parameters of the neural network are the key factors determining the training effect. Considering the computational complexity and accuracy of the training, the neural network training is controlled by two parameters: the number of iterations and the error amount. The training error determines the training accuracy and prevents overfitting. The number of iterations controls the computational complexity of the training and prevents getting stuck in multiple loops and unable to end properly.

[0104] Step 4: Neural network training. According to the neural network structure and parameters, the seismic data volume obtained in Step 2 is used. One trace of data is extracted at a time, and the amplitude value of each sampling point of one trace is used as the input data and input into each node of the input layer of the neural network. The trace data corresponding to the meaning of the arrival time of the input trace is used as the output verification data. When training the seismic data, the data is not input in order, but one trace of data is input at a certain interval, and the process is cycled in turn until the training of all input data is finally completed.

[0105] Step 5: First arrival picking. Using the neural network trained in Step 4, the seismic data that needs to be picked for the first arrival is used to extract the seismic trace data volume according to the requirements of Step 2 and input the traces into the neural network in trace order. After the neural network finishes the calculation, the sampling position of the trace corresponding to the node with a node value of 1 in the output layer is used as the first arrival time of the target trace. All the seismic data that needs to be picked for the first arrival is input into the neural network in turn, and finally the first arrival times of all the data are obtained. The first arrival image after picking is as Figure 6 shown.

[0106] Step 6: Structure setting of the static correction processing neural network. The structure of the static correction neural network is divided into an input layer, an intermediate layer, and an output layer. The input layer is set with six nodes, which respectively correspond to the shot point coordinates and elevation (Sx, Sy, Sz), and the receiver point coordinates and elevation (Rx, Ry, Rz) of each trace of data. The intermediate layer is set with two layers, and the number of nodes in each layer is 50. There is one node in the output layer, and the output is the first arrival time. The neural network is a fully connected network.

[0107] Step 7: Static correction neural network training. The shot point coordinates and elevation (Sx, Sy, Sz), and the receiver point coordinates and elevation (Rx, Ry, Rz) of each trace of the seismic data are extracted, and the six extracted parameters are input into the input layer of the neural network. The first arrival time of the input trace is used as the verification data of the output layer. The neural network training is performed on each trace of the target seismic data in turn.

[0108] Step 8: Static correction processing. Save the neural network parameters after completing the training. According to the requirements of static correction, determine the surface elevation, replace the shot point elevation and the build wave point elevation in the seismic data traces that need static correction processing according to the new surface elevation, obtain the static correction shot point coordinates (Sx, Sy, Sz) and geophone coordinates (Rx, Ry, Rz), input the six parameters of the obtained new elevation into the trace neural network, and calculate and output the first arrival time data of the corresponding trace through the neural network calculation. Subtract the first arrival time calculated from the target elevation from the first arrival time picked from the actual elevation, and the obtained difference is the static correction amount. Calculate all trace data in sequence to obtain the final static correction amount. Correct the target seismic data according to the obtained static correction amount to achieve static correction processing. Please refer to Figure 7A and Figure 7B , which are the first arrival picks before static correction processing and the first arrival picks after static correction processing, respectively.

[0109] Example 5

[0110] In this example, as Figure 3 shown, a seismic data static correction processing device is provided, including:

[0111] A target seismic data acquisition module 310, configured to acquire target seismic data;

[0112] A first neural network training module 320, configured to input the target seismic data into a pre-trained initial pick-up neural network model for training 330 to obtain pick-up data including first arrival time data;

[0113] A second neural network training module 340, configured to input the pick-up data including the first arrival time data into a static correction processing neural network for training to obtain pick-up data after static correction processing and second arrival time data;

[0114] A static correction amount calculation module 350, configured to calculate a static correction amount according to the second arrival time data and the first arrival time data after static correction processing;

[0115] A static correction module 360, configured to correct seismic data according to the static correction amount.

[0116] In one embodiment, the seismic data static correction processing device further includes:

[0117] A sample data acquisition module, configured to acquire sample data in a preset format, where the sample data in the preset format is each trace data of shot gather data, and the format of the seismic trace data includes a file header, trace header data for each trace, and a trace data body, and the trace data body records the amplitude values at each sampling point;

[0118] An initial picking neural network model training and obtaining module, configured to input the sample data in the preset format into an initial picking neural network for training to obtain the initial picking neural network model.

[0119] In one embodiment, the seismic data static correction processing device further includes:

[0120] An initial picking neural network construction module, configured to construct an initial picking neural network including a first input layer, a first intermediate layer, and a first output layer, where the first input layer is used to input a trace of seismic data, and after being calculated by the first intermediate layer, the first output layer outputs a data with the same number of sample points as the input seismic trace data, and each sample point value in the output data is a first sample point value or a second sample point value.

[0121] In one embodiment, the step of inputting the target seismic data into a pre-trained initial picking neural network model for training to obtain the first arrival picking data including the first arrival time data includes:

[0122] Input the target seismic data into a pre-trained initial picking neural network model for training, and the first arrival picking neural network model outputs the first arrival picking data including sample point values, where the sample point values include a first sample point value and a second sample point value;

[0123] Extract, through the first arrival picking neural network model, the first arrival picking data with the sample point value being the first sample point value, and obtain the first arrival time data of the first arrival picking data corresponding to the first sample point value.

[0124] In one embodiment, the second neural network training module includes:

[0125] A first arrival picking data input unit, configured to input the first arrival picking data including the first arrival time data into a static correction processing neural network for training;

[0126] A target elevation output unit, configured to reset the elevation of static correction for the target seismic data that needs static correction processing through the static correction processing application network, and output the first arrival time of the target elevation.

[0127] In one embodiment, the seismic data static correction processing device further includes:

[0128] A static correction processing neural network construction module, configured to construct a static correction processing neural network including a second input layer, a second intermediate layer, and a second output layer.

[0129] In one embodiment, the static correction amount calculation module is further configured to calculate the difference between the second arrival time data after static correction processing and the first arrival time data to obtain the static correction amount.

[0130] For the specific limitations of the seismic data static correction processing device, reference can be made to the limitations of the seismic data static correction processing method in the above text, which will not be elaborated here. Each unit in the above seismic data static correction processing device can be implemented in whole or in part by software, hardware, and their combination. The above units can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each of the above units.

[0131] Embodiment Six

[0132] In this embodiment, a computer device is provided. Its internal structure diagram can be as Figure 4 shown. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program, and a database is deployed on the non-volatile storage medium, and the database is used to store the first arrival picking neural network model and the static correction processing neural network model. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with other computer devices. The computer program, when executed by the processor, implements a seismic data static correction processing method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0133] Those skilled in the art can understand that Figure 4 the structure shown in

[0134] is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have a different component layout.

[0135] Step 210, obtain target seismic data.

[0136] Step 220, input the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data including first arrival time data.

[0137] Specifically, a data volume of sample seismic data is extracted from the target seismic data, and the file header description information of the sample seismic data and the trace header information of each trace are removed. It is verified that the obtained data volume is a data volume of the same size as generated according to the seismic data volume, and the value of each sampling point in the data volume is set to 0, and then the value at the sampling point position corresponding to the first arrival time of each trace is set to 1, indicating the first arrival time position of the data of this trace.

[0138] For the obtained seismic data volume, one trace of data is extracted at a time, and the amplitude value of each sampling point of one trace is used as input data and input into each node of the initial picking neural network model, and the trace data corresponding to the meaning of the first arrival time of the input trace is used as the output verification data. When the seismic data is trained, it is not input in sequence, but one trace of data is input at a certain interval, and the cycle is repeated in turn until the training of all input data is finally completed.

[0139] Specifically, the seismic trace data volume to be picked for the first arrival is extracted, and the trace is input into the initial picking neural network model in trace order. After being calculated by the initial picking neural network model, the sampling position of the trace corresponding to the node with a node value of 1 in the output layer is used as the first arrival time of the target trace. All the seismic data that needs to be picked for the first arrival is input into the trace neural network in turn, and finally the first arrival times of all the data are obtained.

[0140] By inputting one trace of seismic data into the initial picking neural network model, after being calculated by the initial picking neural network model, the output layer outputs a data with the same number of sample points as the input seismic trace data, and the value of each sample point in the data is 1 or 0. 1 indicates that the sample point is the first arrival time point, and 0 indicates that the sample point is not the first arrival time point. In this embodiment, the first arrival time point corresponding to this sampling point is the first arrival time data, and this first arrival time data is the first arrival time obtained by actual elevation picking.

[0141] Step 230, input the first arrival picking data including the first arrival time data into the static correction processing neural network for training to obtain the first arrival picking data after static correction processing and the second arrival time data.

[0142] The goal of the static correction processing neural network is to establish the relationship between the surface elevation and the first arrival time. The neural network structure is divided into an input layer, an intermediate layer, and an output layer. The input layer is the shot point position coordinates and elevation and the receiver point position coordinates and elevation of the seismic trace data. Therefore, six nodes are set in the input layer, corresponding to the shot point coordinates and elevation (Sx, Sy, Sz) and the receiver point coordinates and elevation (Rx, Ry, Rz) of each trace data respectively. Two layers are set in the intermediate layer, and the number of nodes in each layer is 50. One node in the input layer outputs the first arrival time. The neural network is a fully connected network.

[0143] In this embodiment, the shot point coordinates and elevation (Sx, Sy, Sz) as well as the geophone point coordinates and elevation (Rx, Ry, Rz) of each trace of seismic data are extracted. The six extracted parameters are input into the input layer of the trace neural network, and the first arrival time of the input trace is used as the verification data for the output layer. The neural network training is performed on each trace of the target seismic data in sequence.

[0144] Step 240: Calculate the static correction amount based on the second first arrival time data and the first first arrival time data after static correction processing.

[0145] In one embodiment, the step of calculating the static correction amount based on the second first arrival time data and the first first arrival time data after static correction processing includes: calculating the difference between the second first arrival time data and the first first arrival time data after static correction processing to obtain the static correction amount.

[0146] After the neural network is established, the relationship between the elevation data of each trace and the first arrival time is selected. The elevation for static correction of the target seismic data that needs to be static corrected is reset and input into the neural network. The value output after neural network calculation is the first arrival time of the target elevation. Subtracting the first arrival time picked from the actual elevation from the first arrival time of the target elevation, the obtained difference is the static correction amount of the target elevation.

[0147] Step 250: Correct the seismic data according to the static correction amount.

[0148] Perform the above operations on each trace of data to obtain the static correction of the target elevation and complete the static correction processing.

[0149] In the above embodiment, automatic first arrival picking and direct static correction calculation are realized. This method does not require manual picking of first arrival data, avoids the complex near-surface modeling process, and realizes the function of efficient and accurate static correction processing.

[0150] In one embodiment, when the processor executes the computer program, the following steps are also implemented:

[0151] Obtain sample data in a preset format, where the sample data in the preset format is each trace data of the shot gather data. The format of the seismic trace data includes a file header, the trace header data of each trace, and a trace data body, and the trace data body records the amplitude values at each sampling point; input the sample data in the preset format into the initial picking neural network for training to obtain the initial picking neural network model.

[0152] Specifically, after the deep neural network structure design is completed, sample data is required to train the neural network model to obtain the initial pick-up neural network model. The sample data uses the accurate first arrival time data obtained by manual picking. The input sample is each trace data of the shot gather data. The format of the seismic trace data includes the file header, the trace header data of each trace, and the trace data body. The trace data body records the amplitude value at each sampling point. The neural network input data only requires the trace data body. Therefore, it is necessary to reconstruct the input seismic data, strip the file header description data and the trace header data of each trace, and retain the trace data body of each trace to form the sample data of the pure data body. The output sample data is constructed based on the input sample data and the first arrival time. According to the size of the input sample data, an empty data body of the same size is generated, and the value of each sampling point in the data body is 0. According to the first arrival time, the value of the corresponding sampling point of each trace data in the output sample data is set to 1 to obtain the output sample data. The training process of the neural network is as follows: The generated input sample data is input one trace data at a time in the trace order into the output layer of the input trace neural network. Each sampling point of the input trace data corresponds to a node of the input layer. The output data is the output sample data of the corresponding trace. Each sampling point of each output sample corresponds to a node of the output layer.

[0153] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0154] Construct an initial pick-up neural network including a first input layer, a first intermediate layer, and a first output layer. Among them, the first input layer is used to input one trace of seismic data. After being calculated by the first intermediate layer, the first output layer outputs one trace of data with the same number of sample points as the input seismic trace data, and the value of each sample point in the output data is the first sample value or the second sample value.

[0155] In this embodiment, according to the characteristics of the seismic data, the initial pick-up neural network is provided with an input layer, an intermediate layer, and an output layer. The node data of the input layer is the same as the number of sampling points of one trace of seismic data and is used to input one trace of seismic data. The intermediate layer has two layers, and the number of nodes in each layer is the same as that of the input layer. The number of nodes in the output layer is the same as the number of nodes in the input layer. The network is a fully connected network. The input layer inputs one trace of seismic data. After being calculated by the intermediate layer, the output layer outputs one trace of data with the same number of sample points as the input seismic trace data, and the value of each sample point in the data is 1 or 0. 1 indicates that the sample point is the first arrival time point, and 0 indicates that the sample point is not the first arrival time point.

[0156] Specifically, first, set up the first arrival picking neural network structure. Design the neural network structure according to one input layer, two intermediate layers, and one output layer. The node data of the input layer is consistent with the number of sampling points of the seismic data that requires first arrival picking. The number of nodes in the two intermediate layers and the node data of the output layer are consistent with the node data of the input layer. Subsequently, generate training sample data. Select some seismic data that requires first arrival picking as training sample data. First, manually pick the first arrivals of the training sample seismic data to obtain accurate first arrival times. According to the neural network structure, extract the data volume of the sample seismic data, and remove the file header description information and the trace header information of each trace of the sample seismic data. Verify that the obtained data volume generates a data volume of the same size according to the seismic data volume. Each sampling point value in the data volume is set to 0, and then the value at the sampling point corresponding to the first arrival time of each trace is set to 1, indicating the position of the first arrival time of this trace of data. Then, determine the training parameters. The training parameters of the neural network are the key factors determining the training effect. Considering the computational amount and accuracy of the training, control the neural network training through two parameters: the number of loops and the error amount. The training error determines the training accuracy and prevents overfitting. The number of loops controls the computational amount of the training and prevents getting stuck in multiple loops and unable to end normally.

[0157] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0158] Input the target seismic data into a pre-trained initial picking neural network model for training. The first arrival picking neural network model outputs the first arrival picking data including sample values, where the sample values include the first sample value and the second sample value; extract the first arrival picking data with the sample value being the first sample value through the first arrival picking neural network model, and obtain the first first arrival time data of the first arrival picking data corresponding to the first sample value.

[0159] In this embodiment, using the trained neural network model, generate input data according to the requirements of the input samples for the target seismic data that requires first arrival picking processing, input it into the trained neural network. For the data of one trace output after the neural network calculation, extract the sample points with a sampling value of 1 in the output trace, and obtain the position time where the sample point with a value of 1 is located as the first arrival picking result of this trace.

[0160] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0161] Input the first arrival picking data including the first first arrival time data into a static correction processing neural network for training; through the static correction processing application network, reset the elevation of the static correction for the target seismic data that requires static correction processing, and output the first arrival time at the target elevation.

[0162] Specifically, for the seismic data for which the first arrivals have been picked, extract the shot point coordinates and elevation (Sx, Sy, Sz), as well as the geophone point coordinates and elevation (Rx, Ry, Rz) for each trace. Input the six data for each trace obtained into the six nodes of the input layer of the neural network, and the output verification data is the input initial time. Extract training samples from all the seismic traces for which the first arrivals have been picked according to the above processing process and read them into the neural network for training.

[0163] In one embodiment, when the processor executes the computer program, the following steps are further implemented:

[0164] Construct a static correction processing neural network including a second input layer, a second intermediate layer, and a second output layer.

[0165] Specifically, the goal of the static correction processing neural network is to establish the relationship between the surface elevation and the first arrival time. The neural network structure is divided into an input layer, an intermediate layer, and an output layer. The input layer is the shot point position coordinates and elevation of the seismic trace data, as well as the coordinates and elevation of the geophone point position. Therefore, six nodes are set in the input layer, corresponding to the shot point coordinates and elevation (Sx, Sy, Sz) and the geophone point coordinates and elevation (Rx, Ry, Rz) of each trace data respectively. Two layers are set in the intermediate layer, and the number of nodes in each layer is 50. One node in the input layer outputs the first arrival time. The neural network is a fully connected network.

[0166] Embodiment Seven

[0167] In this embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0168] Step 210, obtain target seismic data.

[0169] Step 220, input the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data including first arrival time data.

[0170] Specifically, extract the data body of the sample seismic data from the target seismic data, and remove the file header description information and the trace header information of each trace. The acquisition of the verification data body is to generate a data body of the same size according to the seismic data body, set the value of each sampling point in the data body to 0, and then set the value at the sampling point corresponding to the first arrival time of each trace to 1, indicating the first arrival time position of this trace data.

[0171] For the acquired seismic data volume, extract one trace of data at a time. Use the amplitude value of each sampling point in one trace as input data and input it into each node of the initial picking neural network model. Use the trace data corresponding to the arrival time meaning of the input trace as the output verification data. When training the seismic data, it is not input in sequence, but one trace of data is input at a certain interval, and the cycle continues until the training of all input data is finally completed.

[0172] Specifically, extract the seismic trace data volume of the seismic data that needs to be picked for the first arrival, and input it into the initial picking neural network model in trace order. After being calculated by the initial picking neural network model, the sampling position of the trace corresponding to the node with a node value of 1 in the output layer is used as the first arrival time of the target trace. Input all the seismic data that needs to be picked for the first arrival into the neural network in sequence, and finally obtain the first arrival times of all data.

[0173] By inputting one trace of seismic data into the initial picking neural network model, after being calculated by the initial picking neural network model, the output layer outputs a data with the same number of sample points as the input seismic trace data, and each sample point value in the data is 1 or 0. 1 indicates that the sample point is the first arrival time point, and 0 indicates that the sample point is not the first arrival time point. In this embodiment, the first arrival time point corresponding to the sampling point is the first arrival time data, and this first arrival time data is the first arrival time obtained by actual elevation picking.

[0174] Step 230, input the first arrival picking data containing the first arrival time data into the static correction processing neural network for training to obtain the first arrival picking data after static correction processing and the second arrival time data.

[0175] The goal of the static correction processing neural network is to establish the relationship between the surface elevation and the first arrival time. The neural network structure is divided into an input layer, an intermediate layer, and an output layer. The input layer is the shot point position coordinates and elevation and the geophone point position coordinates and elevation of the seismic trace data. Therefore, six nodes are set in the input layer, corresponding to the shot point coordinates and elevation (Sx, Sy, Sz) and the geophone point coordinates and elevation (Rx, Ry, Rz) of each trace of data respectively. Two layers are set in the intermediate layer, and the number of nodes in each layer is 50. One node in the input layer outputs the first arrival time. The neural network is a fully connected network.

[0176] In this embodiment, extract the shot point coordinates and elevation (Sx, Sy, Sz) and the geophone point coordinates and elevation (Rx, Ry, Rz) of each trace of seismic data, input the six extracted parameters into the input layer of the neural network, and use the first arrival time of the input trace as the verification data of the output layer. Perform neural network training on each trace of the target seismic data in sequence.

[0177] Step 240: Calculate the static correction amount based on the second first arrival time data and the first first arrival time data after static correction processing.

[0178] In one embodiment, the step of calculating the static correction amount based on the second first arrival time data and the first first arrival time data after static correction processing includes: calculating the difference between the second first arrival time data and the first first arrival time data after static correction processing to obtain the static correction amount.

[0179] After the neural network is established, the relationship between the elevation data of each trace and the first arrival time is established. Reset the elevation of the static correction for the target seismic data that needs static correction processing and input it into the neural network. The value output after the neural network calculation is the first arrival time of the target elevation. Subtract the first arrival time picked up from the actual elevation from the first arrival time of the target elevation, and the obtained difference is the static correction amount of the target elevation.

[0180] Step 250: Correct the seismic data according to the static correction amount.

[0181] Perform the above operations on each trace of data to obtain the static correction of the target elevation and complete the static correction processing.

[0182] In the above embodiment, automatic first arrival picking and direct static correction calculation are realized. This method does not require manual picking of first arrival data, avoids the complex near-surface modeling process, and realizes the function of efficient and accurate static correction processing.

[0183] In one embodiment, when the computer program is executed by the processor, the following steps are also realized:

[0184] Obtain sample data in a preset format, where the sample data in the preset format is each trace of shot gather data. The format of the seismic trace data includes a file header, the trace header data of each trace, and a trace data body. The trace data body records the amplitude value at each sampling point; input the sample data in the preset format into the initial picking neural network for training to obtain the initial picking neural network model.

[0185] Specifically, after the deep neural network structure design is completed, sample data is required to train the neural network model to obtain an initial picking neural network model. The sample data uses the accurate first arrival time data obtained by manual picking. The input sample is each trace data of the shot gather data. The format of the seismic trace data includes the file header, the trace header data of each trace, and the trace data body. The trace data body records the amplitude value at each sampling point. The neural network input data only requires the trace data body. Therefore, it is necessary to reconstruct the input seismic data, strip the file header description data and the trace header data of each trace, and retain the trace data body of each trace to form the sample data of the pure data body. The output sample data is constructed based on the input sample data and the first arrival time. According to the size of the input sample data, an empty data body of the same size is generated, and the value of each sampling point in the data body is 0. According to the first arrival time, the value of the corresponding sampling point of each trace data in the output sample data is set to 1 to obtain the output sample data. The training process of the neural network is as follows: The generated input sample data is input one trace data at a time in the trace order, into the output layer of the input trace neural network. Each sampling point of the input trace data corresponds to a node of the input layer. The output data is the output sample data of the corresponding trace. Each sampling point of each output sample corresponds to a node of the output layer.

[0186] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented:

[0187] Construct an initial picking neural network including a first input layer, a first intermediate layer, and a first output layer. Among them, the first input layer is used to input one trace of seismic data. After being calculated by the first intermediate layer, the first output layer outputs one trace of data with the same number of sample points as the input seismic trace data, and the value of each sample point in the output data is the first sample value or the second sample value.

[0188] In this embodiment, according to the characteristics of the seismic data, the initial picking neural network is provided with an input layer, an intermediate layer, and an output layer. The node data of the input layer is the same as the number of sampling points of one trace of seismic data and is used to input one trace of seismic data. The intermediate layer has two layers, and the number of nodes in each layer is the same as that of the input layer. The number of nodes in the output layer is the same as the number of nodes in the input layer. The network is a fully connected network. The input layer inputs one trace of seismic data. After being calculated by the intermediate layer, the output layer outputs one trace of data with the same number of sample points as the input seismic trace data, and the value of each sample point in the data is 1 or 0. 1 indicates that the sample point is the first arrival time point, and 0 indicates that the sample point is a non-first arrival time point.

[0189] Specifically, first, set up the first arrival picking neural network structure. Design the neural network structure according to one input layer, two intermediate layers, and one output layer. The node data of the input layer is consistent with the number of sampling points of the seismic data for which the first arrival needs to be picked. The number of nodes in the two intermediate layers and the node data of the output layer are consistent with the node data of the input layer. Subsequently, generate training sample data. Select some of the seismic data for which the first arrival needs to be picked as the training sample data. First, manually pick the first arrival of the training sample seismic data to obtain the accurate first arrival time. According to the neural network structure, extract the data volume of the sample seismic data, and remove the file header description information of the sample seismic data and the trace header information of each trace. The obtained verification data volume is to generate a data volume of the same size as the seismic data volume, and the value of each sampling point in the data volume is set to 0, and then the value at the sampling point corresponding to the first arrival time of each trace is set to 1, indicating the position of the first arrival time of this trace of data. Then, determine the training parameters. The training parameters of the neural network are the key factors determining the training effect. Considering the computational amount and accuracy of the training, control the neural network training through two parameters: the number of loops and the error amount. The training error determines the training accuracy and prevents overfitting. The number of loops controls the computational amount of the training and prevents getting stuck in multiple loops and unable to end normally.

[0190] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented:

[0191] Input the target seismic data into a pre-trained initial picking neural network model for training. The first arrival picking neural network model outputs the first arrival picking data including sample values, where the sample values include the first sample value and the second sample value; extract the first arrival picking data with the sample value being the first sample value through the first arrival picking neural network model, and obtain the first first arrival time data of the first arrival picking data corresponding to the first sample value.

[0192] In this embodiment, using the trained neural network model, generate input data according to the requirements of the input samples for the target seismic data that needs to be processed for first arrival picking, input it into the trained neural network, calculate the neural network and output a trace of data, extract the samples with a sampling value of 1 in the output trace, and obtain the position time where the sample with a value of 1 is located as the first arrival picking result of this trace.

[0193] In one embodiment, when the computer program is executed by the processor, the following steps are further implemented:

[0194] Input the first arrival picking data including the first first arrival time data into the static correction processing neural network for training; through the static correction processing application network, reset the elevation of the static correction for the target seismic data that needs to be static corrected, and output the first arrival time of the target elevation.

[0195] Specifically, for the seismic data where the first arrivals have been picked up, the shot point coordinates and elevation (Sx, Sy, Sz), as well as the geophone point coordinates and elevation (Rx, Ry, Rz) of each trace are extracted. The six data of each trace obtained are respectively input into six nodes of the input layer of the neural network, and the output verification data is the input initial time. All the seismic traces with picked-up first arrivals are processed according to the above process to extract training samples and then read into the neural network for training.

[0196] In one embodiment, when the computer program is executed by a processor, the following steps are further implemented:

[0197] Construct a static correction processing neural network including a second input layer, a second intermediate layer, and a second output layer.

[0198] Specifically, the goal of the static correction processing neural network is to establish the relationship between the surface elevation and the first arrival time. The neural network structure is divided into an input layer, an intermediate layer, and an output layer. The input layer is the shot point position coordinates and elevation of the seismic trace data, as well as the coordinates and elevation of the geophone point position. Therefore, six nodes are set in the input layer, corresponding respectively to the shot point coordinates and elevation (Sx, Sy, Sz) and the geophone point coordinates and elevation (Rx, Ry, Rz) of each trace data. Two layers are set in the intermediate layer, and the number of nodes in each layer is 50. One node in the input layer outputs the first arrival time. The neural network is a fully connected network.

[0199] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0200] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0201] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. A method for static correction processing of seismic data, characterized in that, it includes: Obtain target seismic data; Input the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data containing first arrival time data; Construct a static correction processing neural network including a second input layer, a second intermediate layer and a second output layer. The goal of the static correction processing neural network is to establish the relationship between surface elevation and first arrival time. The second input layer is the coordinates and elevation of the shot point position and elevation and the geophone point position of the seismic trace data; Input the first arrival picking data containing the first arrival time data into the static correction processing neural network for training to obtain the first arrival picking data after static correction processing and second arrival time data; Calculate the static correction amount according to the second arrival time data and the first arrival time data after static correction processing; Correct the seismic data according to the static correction amount.

2. The method according to claim 1, characterized in that, Before the step of inputting the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data containing first arrival time, it includes: Obtain sample data in a preset format. Among them, the sample data in the preset format is each trace data of the shot gather data. The format of the seismic trace data includes a file header, the trace header data of each trace and the trace data body. The trace data body records the amplitude value at each sampling point; Input the sample data in the preset format into the initial picking neural network for training to obtain the initial picking neural network model.

3. The method according to claim 2, characterized in that, Before the step of obtaining the sample data in the preset format, it further includes: Construct an initial picking neural network including a first input layer, a first intermediate layer and a first output layer. Among them, the first input layer is used to input one trace of seismic data. After calculation by the first intermediate layer, the first output layer outputs a data with the same number of sample points as the input seismic trace data. Each sample point value in the output data is a first sample point value or a second sample point value.

4. The method according to claim 3, characterized in that, The step of inputting the target seismic data into a pre-trained initial picking neural network model for training to obtain first arrival picking data containing first arrival time data includes: Input the target seismic data into a pre-trained initial picking neural network model for training. The first arrival picking neural network model outputs the first arrival picking data containing sample point values. Among them, the sample point values include a first sample point value and a second sample point value; Extract the first arrival picking data with the sample point value being the first sample point value through the first arrival picking neural network model, and obtain the first arrival time data of the first arrival picking data corresponding to the first sample point value.

5. The method according to claim 1, characterized in that, The step of inputting the first arrival picking data containing the first arrival time data into the static correction processing neural network for training to obtain the first arrival picking data after static correction processing and second arrival time data includes: Input the first arrival pick-up data containing the first arrival time data into the static correction processing neural network for training; Through the static correction processing application network, reset the elevation of the static correction for the target seismic data to be statically corrected, and output the first arrival time at the target elevation.

6. The method according to any one of claims 1-5, characterized in that, The step of calculating the static correction amount according to the second arrival time data and the first arrival time data after static correction processing includes: Calculate the difference between the second arrival time data and the first arrival time data after static correction processing to obtain the static correction amount.

7. A device for static correction processing of seismic data, characterized in that, comprising: A target seismic data acquisition module for acquiring target seismic data; A first neural network training module for inputting the target seismic data into a pre-trained initial pick-up neural network model for training to obtain first arrival pick-up data containing first arrival time data; A second neural network training module for constructing a static correction processing neural network including a second input layer, a second intermediate layer and a second output layer. The objective of the static correction processing neural network is to establish the relationship between the surface elevation and the first arrival time. The second input layer is the coordinates and elevations of the shot point position and elevation and the geophone point position of the seismic trace data; input the first arrival pick-up data containing the first arrival time data into the static correction processing neural network for training to obtain the first arrival pick-up data and the second arrival time data after static correction processing; A static correction amount calculation module for calculating the static correction amount according to the second arrival time data and the first arrival time data after static correction processing; A static correction module for correcting the seismic data according to the static correction amount.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images