A Smart Microseismic First Arrival Picking Method Based on TX-FK Domain Information

Abstract

The present invention discloses a method for intelligently picking up the first arrival of microseismic events based on TX-FK domain information, which relates to the field of geophysical technology, and includes: processing microseismic data to obtain TX domain data, and obtaining FK domain data based on the TX domain data; constructing a combined U-net model, and the combined U-net model is used for intelligently picking up the first arrival of microseismic events; training the combined U-net model based on the TX domain data and the FK domain data to obtain a trained combined U-net model; inputting the processed microseismic data to be picked up into the trained combined U-net model to obtain an output result, processing the output result to obtain a picking result, and completing the picking up of the first arrival of microseismic events. The present invention has good noise resistance, and can pick up microseismic first arrival information with high precision under low signal-to-noise ratio conditions.

Description

Technical Field

[0001] The invention belongs to the field of geophysical technology, and in particular relates to a microseismic first arrival intelligent picking method based on TX-FK domain information. Background Art

[0002] Microseismic first-arrival picking usually refers to identifying and extracting first-arrival waves from microseismic records, and is a key link in microseismic monitoring. At present, the microseismic first-arrival intelligent picking method combined with deep learning technology has received extensive attention and research, and some methods can automatically process complex microseismic data and improve the efficiency and accuracy of first-arrival picking. However, the existing first-arrival picking methods mainly use time-space (TX) domain information and are easily affected by noise. The reason is that this type of method only uses single-domain information and it is difficult to accurately capture enough information related to the first-arrival wave in a noisy environment. For most actual work areas, the signal-to-noise ratio of microseismic data is low, resulting in low accuracy in picking first arrivals by this type of method. Therefore, a microseismic first-arrival picking method that can extract multi-domain information is needed to improve the accuracy of microseismic first-arrival picking under low signal-to-noise ratio conditions. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a microseismic first-arrival intelligent picking method based on TX-FK domain information. By constructing a combined U-net model with dual-channel input, features are extracted from the microseismic data in the TX domain and the FK domain, and the features are fused to obtain accurate first-arrival picking results. The method has good noise resistance and can pick up microseismic first-arrival information with high precision under low signal-to-noise ratio conditions.

[0004] To achieve the above object, the present invention provides a microseismic first arrival intelligent picking method based on TX-FK domain information, comprising:

[0005] Processing the microseismic data to obtain TX domain data, and obtaining FK domain data based on the TX domain data;

[0006] Constructing a combined U-net model, wherein the combined U-net model is used for intelligent picking of microseismic first arrivals;

[0007] Training the combined U-net model based on the TX domain data and the FK domain data to obtain a trained combined U-net model;

[0008] The microseismic data to be picked up are processed and input into the trained combined U-net model to obtain an output result, and the output result is processed to obtain a picking result, thereby completing the microseismic first arrival picking.

[0009] Optionally, processing the microseismic data to obtain TX domain data includes:

[0010] Partial traces are extracted from the microseismic data, and the microseismic data are divided into several small blocks by stepping in a step-by-step manner to obtain TX domain data.

[0011] Optionally, obtaining FK domain data based on the TX domain data includes:

[0012] Perform two-dimensional Fourier transform processing on the TX domain data to obtain FK domain data.

[0013] Optionally, constructing the combined U-net model, comprising: a first feature extraction module, a second feature extraction module and a feature fusion module;

[0014] The first feature extraction module is used to extract TX domain data features according to the TX domain data;

[0015] The second feature extraction module is used to extract FK domain data features according to the FK domain data;

[0016] The feature fusion module is used to fuse the TX domain data features and the FK domain data features to obtain fused feature information, perform convolution operation on the fused feature information to obtain convolution results, and complete the construction of the combined U-net model.

[0017] Optionally, the first feature extraction module includes a first encoder, a first bottleneck layer and a first decoder;

[0018] The first encoder extracts local features of the input TX domain data through a convolution layer, followed by downsampling once, and repeats this operation twice to finally obtain a feature map after two downsamplings;

[0019] The first bottleneck layer further extracts higher-level features from the downsampled feature map of the first encoder through a deep convolutional layer;

[0020] The first decoder upsamples and convolves the bottleneck feature map, uses skip connections to concatenate the feature map of the corresponding layer in the encoder with the current feature map of the decoder, and then further extracts features through the convolution layer. This complete operation is performed twice to finally obtain the TX domain data features.

[0021] Optionally, the second feature extraction module includes a second encoder, a second bottleneck layer and a second decoder;

[0022] The second encoder extracts local features of the input FK domain data through a convolution layer, followed by downsampling once, and repeats this operation twice to finally obtain a feature map after two downsamplings;

[0023] The second bottleneck layer further extracts higher-level features from the downsampled feature map of the second encoder through a deep convolutional layer;

[0024] The second decoder upsamples and convolves the bottleneck feature map, uses skip connections to concatenate the feature map of the corresponding layer in the encoder with the current feature map of the decoder, and then further extracts features through the convolution layer. This complete operation is performed twice to finally obtain the FK domain data features.

[0025] Optionally, obtain a trained combined U-net model including:

[0026] The combined U-net model is trained based on the TX domain data and the FK domain data to obtain sample output, calculate the error between the sample output and the picking result, and update the hyperparameters of the combined U-net model until the error converges to obtain a trained combined U-net model.

[0027] Optionally, processing the output result to obtain a picking result includes:

[0028] The output results are merged to obtain binary key waveform data, and the waveform starting point is selected as the microseismic first arrival point according to the binary key waveform data to obtain the picking result.

[0029] Technical effect of the invention: The present invention discloses a method for intelligently picking up the first arrival of microseismic earthquakes based on TX-FK domain information, using U-net to build a combined U-net model with dual input channels; using TX domain and FK domain microseismic data as sample input, and using key waveforms starting from the first arrival point as sample output, constructing training samples and training the combined U-net model; using the trained combined U-net model to realize intelligent picking up of the first arrival of microseismic earthquakes; the present invention can effectively alleviate the problem that conventional methods are difficult to accurately capture the first arrival information in a noisy environment by simultaneously extracting the characteristic information of the first arrival of microseismic earthquakes in the TX domain and the FK domain, and can significantly improve the accuracy of picking up the first arrival of microseismic earthquakes. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0031] Figure 1 A schematic diagram of a process of a microseismic first arrival intelligent picking method based on TX-FK domain information according to an embodiment of the present invention;

[0032] Figure 2 A schematic diagram of a combined U-net model with dual input channels constructed for an embodiment of the present invention;

[0033] Figure 3 A schematic diagram of the training process of the combined U-net model constructed for an embodiment of the present invention;

[0034] Figure 4 Schematic diagram of the first arrival picking results of the proposed method under different signal-to-noise ratios in the embodiment of the present invention, where (a) is 20dB, (b) is 10dB, (c) is 5dB, and (d) is 2dB;

[0035] Figure 5 Schematic diagram of first arrival picking results of the conventional method under different signal-to-noise ratios according to the embodiment of the present invention, where (a) is 20 dB, (b) is 10 dB, (c) is 5 dB, and (d) is 2 dB;

[0036] Figure 6 This is a schematic diagram of the first arrival picking results of five data tracks when the signal-to-noise ratio is 5 dB in an embodiment of the present invention;

[0037] Figure 7 Schematic diagram of the first arrival picking result of microseismic data in the actual working area of ​​the embodiment of the present invention, wherein (a) is the method proposed by the present invention, and (b) is the conventional method;

[0038] Figure 8 It is a schematic diagram of the first arrival picking results of five data for actual work area data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0039] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0040] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0041] like Figure 1-3 As shown, this embodiment provides a microseismic first arrival intelligent picking method based on TX-FK domain information, including:

[0042] Processing the microseismic data to obtain TX domain data, and obtaining FK domain data based on the TX domain data;

[0043] Constructing a combined U-net model, wherein the combined U-net model is used for intelligent picking of microseismic first arrivals;

[0044] Training the combined U-net model based on the TX domain data and the FK domain data to obtain a trained combined U-net model;

[0045] The microseismic data to be picked up are processed and input into the trained combined U-net model to obtain an output result, and the output result is processed to obtain a picking result, thereby completing the microseismic first arrival picking.

[0046] Furthermore, the microseismic data is processed to obtain TX domain data including:

[0047] Extract some traces from the microseismic data, divide the microseismic data into several small blocks by stepping, and obtain TX domain data. This can capture the local features in each sub-area more carefully, increase the diversity of data, and improve the resolution, and prevent the problem of frequency domain wave number feature dispersion that may be caused when directly extracting features over a large range.

[0048] Further, obtaining FK domain data based on the TX domain data includes:

[0049] Perform two-dimensional Fourier transform processing on the TX domain data to obtain FK domain data.

[0050] The corresponding TX domain data is converted into the corresponding FK domain data, so that the features of the small block in the TX domain and the FK domain remain consistent. The features are more concentrated after being converted into the FK domain, which improves the stability of the frequency wavenumber distribution and makes the features clearer and more identifiable.

[0051] Further, the combined U-net model is constructed, which includes: a first feature extraction module, a second feature extraction module and a feature fusion module;

[0052] The first feature extraction module is used to extract TX domain data features according to the TX domain data;

[0053] The second feature extraction module is used to extract FK domain data features according to the FK domain data;

[0054] The feature fusion module is used to fuse the TX domain data features and the FK domain data features to obtain fused feature information, perform convolution operation on the fused feature information to obtain convolution results, and complete the construction of the combined U-net model.

[0055] The TX domain features reflect the temporal and spatial changes of seismic signals, while the FK domain features reveal the frequency and wave number information of the signal. Fusion of the features of these two domains helps to make up for the deficiencies that may exist in a single domain, making the fused features more complete and diverse, and improving the model's ability to capture complex seismic signals.

[0056] Further, the first feature extraction module includes a first encoder, a first bottleneck layer and a first decoder;

[0057] The first encoder extracts local features of the input TX domain data through a convolution layer, followed by downsampling once, and repeats this operation twice to finally obtain a feature map after two downsamplings;

[0058] The first bottleneck layer further extracts higher-level features from the downsampled feature map of the first encoder through a deep convolutional layer;

[0059] The first decoder upsamples and convolves the bottleneck feature map, uses skip connections to concatenate the feature map of the corresponding layer in the encoder with the current feature map of the decoder, and then further extracts features through the convolution layer. This complete operation is performed twice to finally obtain the TX domain data features.

[0060] The second feature extraction module includes a second encoder, a second bottleneck layer and a second decoder;

[0061] The second encoder extracts local features of the input FK domain data through a convolution layer, followed by downsampling once, and repeats this operation twice to finally obtain a feature map after two downsamplings;

[0062] The second bottleneck layer further extracts higher-level features from the downsampled feature map of the second encoder through a deep convolutional layer;

[0063] The second decoder upsamples and convolves the bottleneck feature map, uses skip connections to concatenate the feature map of the corresponding layer in the encoder with the current feature map of the decoder, and then further extracts features through the convolution layer. This complete operation is performed twice to finally obtain the FK domain data features.

[0064] Specifically, the first feature extraction module and the second feature extraction module both adopt the following structure; first, the encoder extracts features from the input data through two 3x3 16-channel convolutional layers, and then uses the maximum pooling layer to downsample to reduce the spatial dimension of the feature map; then, the data passes through two 3x3 32-channel convolutional layers and is downsampled again through the pooling layer; the bottleneck layer uses two 3x3 32-channel convolution operations to further extract high-level features; the decoder first performs upsampling (for example, using UpSampling2D) and passes through a 3x3 32-channel convolutional layer, then splices the second-layer feature map in the encoder through a jump connection, and then further extracts and fuses features through two 3x3 32-channel convolutional layers. Next, upsampling is performed again, and after being processed by a 3x3 16-channel convolutional layer, it is spliced ​​with the first-layer feature map of the encoder, and finally the features are further fused through two 3x3 16-channel convolution operations, and finally a 1x1 convolutional layer is used to output the features, using a linear activation function. The output TX domain and FK domain features are fused, and the fused features are reduced in dimension through a 1x1 convolution layer. Finally, the sigmoid activation function is used to output a binary classification probability value.

[0065] Through multi-layer convolution and downsampling, the model can effectively extract local features and high-level features in the TX domain and FK domain, thereby improving the recognition ability of complex structures. In addition, through the mechanism of upsampling and jump connection, the model can restore the spatial resolution of the feature map, retain detail information, and reduce information loss during downsampling. This structure helps to improve the accuracy of the model, especially when processing complex data, and enhances the feature extraction ability.

[0066] Furthermore, the trained combined U-net model is obtained including:

[0067] The combined U-net model is trained based on the TX domain data and the FK domain data to obtain sample output, calculate the error between the sample output and the picking result, and update the hyperparameters of the combined U-net model until the error converges to obtain a trained combined U-net model.

[0068] In practical applications, seismic data are usually interfered by various noises. However, by fusing the information features of the TX domain and the FK domain, when faced with complex or noisy data, the model can still extract effective features from the information of the two domains, significantly improving the accuracy and stability of picking.

[0069] Further, processing the output result to obtain the picking result includes:

[0070] The output results are merged to obtain binary key waveform data, and the waveform starting point is selected as the microseismic first arrival point according to the binary key waveform data to obtain the picking result.

[0071] Since block processing is used in prediction, the output results of each block are local. By merging these blocks, the complete waveform data can be restored to ensure the spatiotemporal continuity of the entire seismic record. This process effectively avoids information loss or discontinuity caused by block cutting. The use of a binary classification strategy can significantly highlight the key feature areas in the seismic waveform, which helps to identify and determine the first arrival point more quickly and accurately, and improve the accuracy and reliability of the picking results.

[0072] The present invention proposes a method for intelligently picking the first arrival of microseismic earthquakes. To verify its effectiveness, synthetic data and actual data are used for testing, and the method is compared with a conventional microseismic first arrival picking method that only uses TX domain information. In the model data test, 100 microseismic data are synthesized, and 20 of them are selected to make training samples. The sample input is sent to the constructed combined U-net model for training. After 20 iterations, the error between the sample output and the picking result converges, and the training of the combined U-net model is completed. Then, different levels of random noise are gradually added to the synthetic microseismic data to form synthetic microseismic data with different signal-to-noise ratios. The above-mentioned trained combined U-net model and conventional methods are used to perform first arrival picking on microseismic data with different signal-to-noise ratios, and the results are as follows. Figure 4 (a) is 20dB, (b) is 10dB, (c) is 5dB, and (d) is 2dB, which are the first arrival picking results of the proposed method under the same signal-to-noise ratio; Figure 5 (a) is 20dB, (b) is 10dB, (c) is 5dB, and (d) is 2dB. These are the first arrival picking results of the conventional method under different signal-to-noise ratios. Figure 6 The first arrival picking results of five data tracks when the signal-to-noise ratio is 5dB are shown. As can be seen from the figure, as the signal-to-noise ratio decreases, the conventional method that only uses TX domain information gradually shows discontinuous and inaccurate picking results, while this method can still accurately pick up the first arrival information under low signal-to-noise ratio conditions, showing good noise resistance and advancement.

[0073] In the actual data test, 20 data from the actual microseismic data are first selected to make training samples, and then the sample input is sent to the combined U-net model for training, and finally the trained combined U-net model is used to perform first arrival picking on the actual microseismic data. Figure 7 (a) is the method proposed by the present invention, and (b) is the conventional method, which is the result of picking the first arrival of microseismic data in the actual work area; Figure 8As shown in the figure, the first arrival picking results of five data tracks for the actual work area data. The first arrival information picked by the conventional method using only TX domain information has discontinuities, and the root mean square error between it and the true first arrival is 201.66; while the first arrival of microseismic earthquakes picked by the proposed method is more continuous, and the root mean square error between it and the true first arrival is 5.525, which has higher accuracy, further proving the advanced nature of this method.

[0074] The present invention discloses a method for intelligently picking up the first arrival of microseismic earthquakes based on TX-FK domain information. A combined U-net model with dual input channels is constructed by using U-net. Microseismic data in TX domain and FK domain are used as sample inputs, and key waveforms starting from the first arrival point are used as sample outputs to construct training samples and train the combined U-net model. The trained combined U-net model is used to realize intelligent picking up of the first arrival of microseismic earthquakes. The present invention can effectively alleviate the problem that conventional methods are difficult to accurately capture the first arrival information in a noisy environment by simultaneously extracting characteristic information of the first arrival of microseismic earthquakes in the TX domain and the FK domain, and can significantly improve the accuracy of picking up the first arrival of microseismic earthquakes.

[0075] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A microseismic first arrival intelligent picking method based on TX-FK domain information, It is characterized in that include: Processing the microseismic data to obtain TX domain data, and obtaining FK domain data based on the TX domain data; Constructing a combined U-net model, wherein the combined U-net model is used for intelligent picking of microseismic first arrivals; Constructing the combined U-net model, which includes: a first feature extraction module, a second feature extraction module and a feature fusion module; The first feature extraction module is used to extract TX domain data features according to the TX domain data; The second feature extraction module is used to extract FK domain data features according to the FK domain data; The feature fusion module is used to fuse the TX domain data features and the FK domain data features to obtain fused feature information, perform convolution operation processing on the fused feature information to obtain a convolution result, and complete the construction of the combined U-net model; The first feature extraction module includes a first encoder, a first bottleneck layer and a first decoder; The first encoder extracts local features of the input TX domain data through a convolution layer, followed by downsampling once, and repeats this operation twice to finally obtain a feature map after two downsamplings; The first bottleneck layer further extracts higher-level features from the downsampled feature map of the first encoder through a deep convolutional layer; The first decoder upsamples and convolves the bottleneck feature map, concatenates the feature map of the corresponding layer in the encoder with the current feature map of the decoder using a skip connection, and then further extracts features through a convolutional layer. This complete operation is performed twice to finally obtain TX domain data features; The second feature extraction module includes a second encoder, a second bottleneck layer and a second decoder; The second encoder extracts local features of the input FK domain data through a convolution layer, followed by downsampling once, and repeats this operation twice to finally obtain a feature map after two downsamplings; The second bottleneck layer further extracts higher-level features from the downsampled feature map of the second encoder through a deep convolutional layer; The second decoder upsamples and convolves the bottleneck feature map, concatenates the feature map of the corresponding layer in the encoder with the current feature map of the decoder using a skip connection, and then further extracts features through a convolutional layer. This complete operation is performed twice to finally obtain FK domain data features; Training the combined U-net model based on the TX domain data and the FK domain data to obtain a trained combined U-net model; The microseismic data to be picked up are processed and input into the trained combined U-net model to obtain an output result, and the output result is processed to obtain a picking result, thereby completing the microseismic first arrival picking.

2. The microseismic first arrival intelligent picking method based on TX-FK domain information according to claim 1, It is characterized in that Processing microseismic data to obtain TX domain data includes: Partial traces are extracted from the microseismic data, and the microseismic data are divided into several small blocks by stepping in a step-by-step manner to obtain TX domain data.

3. The microseismic first arrival intelligent picking method based on TX-FK domain information according to claim 1, It is characterized in that Obtaining FK domain data based on the TX domain data includes: Perform two-dimensional Fourier transform processing on the TX domain data to obtain FK domain data.

4. The microseismic first arrival intelligent picking method based on TX-FK domain information according to claim 1, It is characterized in that Obtaining the trained combined U-net model includes: The combined U-net model is trained based on the TX domain data and the FK domain data to obtain sample output, calculate the error between the sample output and the picking result, and update the hyperparameters of the combined U-net model until the error converges to obtain a trained combined U-net model.

5. The microseismic first arrival intelligent picking method based on TX-FK domain information according to claim 1, It is characterized in that Processing the output result to obtain the picking result includes: The output results are merged to obtain binary key waveform data, and the waveform starting point is selected as the microseismic first arrival point according to the binary key waveform data to obtain the picking result.

Images