Passive source surface wave data regular interference suppression method and system

Abstract

The application provides a passive source surface wave data regular interference suppression method and system, constructs a passive source surface wave interference suppression network; according to the set time period and the number of seismic traces, in the seismic data, randomly cut out data containing surface wave signals in different space-time ranges to construct a small-size surface wave interference suppression data set; the constructed data set is used to train the passive source surface wave interference suppression network in a supervised learning manner; a trace direction mask is constructed, and the trace direction mask is used to cover part of the seismic trace data in the input of the trained passive source surface wave interference suppression network; the above data is used for transfer learning training without manual labeling to obtain processed data. The application can realize suppression of regular interference in passive source surface wave data.

Description

Technical Field

[0001] This invention belongs to the field of geophysical exploration technology and relates to a method and system for suppressing regular interference in passive source surface wave data. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In recent years, the development of urban shallow surface high-frequency surface wave detection technology has provided new ideas for urban geological exploration. This type of method extracts the dispersion information of surface wave components in artificially excited vibration signals or urban human environmental noise to image the wave velocity distribution of the underground medium. Due to the high energy and large amplitude of surface waves, this method is less affected by environmental noise. In addition, surface waves are sensitive to changes in wave velocity, and have advantages such as non-invasiveness and strong site applicability. Urban shallow surface high-frequency surface wave detection has become an effective approach for urban underground space exploration.

[0004] Surface wave data is typically used for shallow surface shear wave velocity structure imaging. The basic process involves field data acquisition, data processing, surface wave data dispersion imaging and dispersion curve extraction, and shear wave velocity inversion based on the dispersion curve. With the rapid development of artificial intelligence, deep learning technology, the most representative of these technologies, has achieved breakthroughs in fields such as natural language processing, autonomous driving, and medical imaging, ushering in a "revolutionary era" for AI. Deep learning algorithms possess strong nonlinear function fitting capabilities, leading to their increasing application in geophysics and significantly improving the processing capabilities of algorithms for data signal processing, parameter model construction, inversion, and imaging. In surface wave dispersion curve extraction and inversion, existing research demonstrates that deep learning methods exhibit significant advantages in reducing labor costs and the influence of subjective human factors, while improving algorithm accuracy. This effectively addresses the problems faced by traditional surface wave detection methods in imaging subsurface media.

[0005] For existing urban shallow surface passive source surface wave detection technology, regular interference in the surface wave data restricts the effectiveness of surface wave dispersion curve extraction. The virtual source shot concentrations recovered from passive source data contain regular interference, including some volume wave signal interference; it also includes spurious phase axis interference introduced by the mutual influence of multi-mode surface wave signals during passive source data cross-correlation; and the complex surface detection conditions in urban environments cause discontinuities in the virtual source shot concentrations during tunneling, resulting in inter-track acquisition errors. These data quality issues severely interfere with the subsequent surface wave dispersion curve extraction process. Summary of the Invention

[0006] To address the aforementioned problems, this invention proposes a method and system for suppressing regular interference in passive source surface wave data. This invention constructs a small-size surface wave interference suppression dataset for training the interference suppression network. Furthermore, addressing the issue of trace acquisition errors in measured urban shallow surface passive source surface wave data, it further improves the interference suppression effect of measured surface wave data through transfer learning based on trace masks, thereby achieving the suppression of regular interference in passive source surface wave data.

[0007] According to some embodiments, the present invention adopts the following technical solution:

[0008] A method for suppressing regular interference in passive source surface wave data includes the following steps:

[0009] Construct a passive source surface wave interference suppression network;

[0010] Based on the set time period and number of seismic traces, data containing surface wave signals from different spatiotemporal domains are randomly extracted from the seismic data to construct a small-size surface wave interference suppression dataset. The dataset includes noisy surface wave data as input and corresponding clean surface wave data as labels.

[0011] Using the constructed dataset, a passive source surface wave interference suppression network is trained in a supervised learning manner.

[0012] Construct a trace mask and use it to cover part of the seismic trace data in the input of the passive source surface wave interference suppression network after training.

[0013] Using the masked input data and the direct prediction results of the passive source surface wave interference suppression network obtained after supervised learning training with the unmasked data as input, the passive source surface wave interference suppression network is trained again through the data composed of the above input and label data, i.e., the transfer learning training process without artificial labels, to obtain the processed data.

[0014] As an alternative implementation, the passive source surface wave interference suppression network comprises three parts: an encoder, skip connections, and a decoder. The encoder consists of multiple convolutional network layers, the decoder consists of multiple deconvolutional network layers, and the skip connections are used to connect some of the convolutional layers of the encoder with the deconvolutional layers of the decoder.

[0015] As an alternative implementation, the small-size surface wave interference suppression dataset includes input data and label data, wherein the input data includes clean surface wave signals and known passive source surface wave interference, and the label data includes clean surface wave signals.

[0016] As an alternative implementation, in the small-size surface wave interference suppression dataset, the pure surface wave signal is obtained through Rayleigh wave field finite-difference forward modeling of the active surface source.

[0017] As an alternative implementation, the passive source surface wave interference in the small-size surface wave interference suppression dataset includes volume wave interference and spurious in-phase axis interference.

[0018] As a further limitation, surface wave data containing body wave interference is added to the data by increasing the source depth during Rayleigh wave field finite difference forward modeling to increase the body wave interference signal in the data.

[0019] As a further limitation, surface wave data containing spurious in-phase axis interference is obtained by using passive source Rayleigh wave field finite difference forward modeling, that is, by obtaining passive source simulation data through continuous excitation, and then recovering it into virtual source shot gather data by seismic interferometry.

[0020] As an alternative implementation method, the passive source surface wave interference suppression network uses the mean square error loss function to calculate the objective function, which is expressed as:

[0021]

[0022] In the formula, b represents the number of samples for each training session; n represents the number of seismic traces; t represents the number of time sampling points; w is the updatable network parameter; X represents the virtual source shot gather data to be processed; S represents the label data, i.e., the clean surface wave data; and N represents the passive source surface wave interference suppression network.

[0023] As an alternative implementation, when performing unlabeled transfer learning training on the measured passive source surface wave virtual source shot collection, two adjacent traces are randomly selected within the input seismic trace range, and a random vector of shape t×2 is randomly generated as a trace direction mask to replace the two trace data selected in the original input. A trace direction mask is randomly generated for each training step.

[0024] As an alternative implementation, an objective function is constructed based on the L2 loss function, and the objective function is used for unlabeled transfer learning training on a measured passive source surface wave virtual source shot set. The objective function is:

[0025]

[0026] In the formula, b represents the number of samples trained each time; w is the updatable network parameter; w pre Δw represents the network parameters obtained through supervised learning training; Δw represents the update amount of the network parameters after transfer learning; X represents the virtual source shot set data to be processed; and N represents the passive source surface wave interference suppression network with added noise mask.

[0027] A passive source surface wave data regular interference suppression system includes:

[0028] The network building module is configured to build a passive source surface wave interference suppression network;

[0029] The dataset construction module is configured to randomly extract data containing surface wave signals from seismic data within different spatiotemporal domains according to the set time period and number of seismic traces, and construct a small-size surface wave interference suppression dataset.

[0030] The training module is configured to train the passive source surface wave interference suppression network in a supervised learning manner using the constructed dataset.

[0031] The trace mask module is configured to construct a trace mask, which is used to cover part of the seismic trace data in the input of the trained passive source surface wave interference suppression network.

[0032] The processing module is configured to use the aforementioned masked input data and the direct prediction results of the passive source surface wave interference suppression network obtained after supervised learning training with the unmasked data as input as label data, and then retrain the passive source surface wave interference suppression network using the data composed of the aforementioned input and label data to obtain the processed data.

[0033] A computer-readable storage medium storing a plurality of instructions adapted for loading by a processor of a terminal device and executing steps in the method.

[0034] A terminal device includes a processor and a computer-readable storage medium, the processor being configured to implement instructions; the computer-readable storage medium being configured to store a plurality of instructions adapted to be loaded by the processor and executed in accordance with the steps of the method described therein.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] This invention addresses the common problem in conventional supervised learning methods for suppressing seismic data interference by using relatively complete spatiotemporal simulated seismic data as input and output. This relatively complete simulated data contains a significant amount of wavefield characteristic information unique to simulated data, such as the energy distribution and variation characteristics, and the spatiotemporal distribution location of effective surface wave signals. These information differs between simulated and measured data. If the network captures this information, it will severely impact the generalization ability of the interference suppression method in practical applications. This invention proposes a method that constructs a surface wave regular interference suppression training set using segmented, smaller-sized data, followed by supervised learning training. In the segmented, smaller-sized data, the aforementioned wavefield characteristics are no longer fully present, becoming variables in different spatiotemporal small-scale data. This prevents the network from capturing the features that affect the method's generalization ability; the network can only more easily capture the waveform morphology characteristics of seismic wave data. Surface wave signals have a unique broom-like morphology, clearly distinguishable from other regular interference, and are applicable to both simulated and measured data. Therefore, the dataset construction method proposed in this invention has the potential to train a network with greater generalization ability in practical applications.

[0037] This invention addresses the common problem of inter-trace acquisition errors caused by complex surface environments in real-world urban passive source surface wave data by proposing a transfer learning method for surface wave interference suppression based on trace-direction masks. Since this measured error is difficult to represent through numerical simulation, transfer learning training on measured data is necessary. Based on an interference suppression network obtained through supervised learning training, a transfer learning method for surface wave interference suppression based on trace-direction masks is developed. This allows the network to be directly applied to the transfer learning training process of measured noisy data. Furthermore, by utilizing blind spot network theory, the network automatically fills in the surface wave data obscured by the mask during training, thereby suppressing seismic trace-direction interference and ultimately improving the interference suppression effect of the method on measured surface wave data.

[0038] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0039] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0040] Figure 1 This is a schematic diagram of the network structure of the passive source surface wave interference suppression network in this embodiment;

[0041] Figure 2 This is a schematic diagram illustrating the process of constructing the small-size surface wave interference suppression dataset in this embodiment;

[0042] Figure 3This is a schematic diagram of the method for adding a directional mask during transfer learning;

[0043] Figure 4 This is a schematic diagram of the results of processing measured surface wave data. Detailed Implementation

[0044] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0045] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0046] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0047] Example 1

[0048] This embodiment discloses a method for suppressing rule-based interference in passive source surface wave data based on deep learning, including the following steps:

[0049] Construct a passive source surface wave interference suppression network, the network structure of which is as follows: Figure 1 As shown, the network takes the surface wave virtual source shot set containing interference, recovered from noisy data, as input and outputs interference-free surface wave data. The passive source surface wave interference suppression network consists of three parts: an encoder, skip connections, and a decoder. The encoder is composed of multiple convolutional network layers, the decoder is composed of multiple deconvolutional network layers, and skip connections are used to connect some of the convolutional layers of the encoder with the deconvolutional layers of the decoder.

[0050] In this embodiment, each input and output is a seismic record with 200 time steps and 20 channels. The network consists of three parts: an encoder, a decoder, and skip connections. The encoder and decoder each contain 5 network layers. The two parts are connected by a convolutional layer, and some network layers are directly connected by skip connections.

[0051] A small-scale surface wave interference suppression dataset was constructed for training a passive source surface wave interference suppression network. The dataset, used as network input, should contain clean surface wave signals and common passive source surface wave interference. Additionally, the dataset should include clean surface wave signals as label data. Furthermore, all input and label data in the dataset should be small-scale spatiotemporal domain (TX domain) data, obtained through data cropping. Specifically, from a complete dataset with a certain sampling time and number of seismic traces, smaller regions containing surface wave signals are randomly cropped according to a fixed time and number of seismic traces. The cropped data comprehensively encompasses data from different spatiotemporal domains.

[0052] In some embodiments, in the small-size surface wave interference suppression dataset, the clean surface wave signal is obtained through Rayleigh wave field finite-difference forward modeling of active surface sources.

[0053] In some embodiments, the passive source surface wave interference in the small-size surface wave interference suppression dataset includes volume wave interference and spurious in-phase axis interference.

[0054] In some embodiments, surface wave data containing body wave interference can be increased by increasing the source depth during Rayleigh wave field finite difference forward modeling to increase the body wave interference signal in the data.

[0055] In some embodiments, passive source Rayleigh wave field finite difference forward modeling can be used to obtain passive source simulation data through continuous excitation, and then the virtual source shot gather data can be recovered by seismic interferometry. In this way, surface wave data containing false in-phase axis interference can be obtained.

[0056] In this embodiment, pure surface wave data is generated on a two-dimensional layered wave velocity model using the finite difference method forward modeling of the second-order spatial and second-order temporal two-dimensional elastic wave equations. This process employs an active source with a Ricker wavelet as the source, a time step of 0.2 ms, and the AEA free boundary processing method. The source is positioned at the surface with a dominant wavelet frequency of 15 Hz to 35 Hz and a wavelet delay of 50 ms. When generating surface wave data containing body waves, the source depth is increased to 5 m-30 m. When generating data containing spurious phase axis interference, continuously excited Ricker wavelets are used as the source, and cross-correlation is used to recover the noise source data into a spurious shot gather, thus representing the data containing spurious phase axis interference. The complete data is then randomly cropped, with 200 time steps and 20 traces of seismic records randomly selected from a complete dataset each time. This process is as follows: Figure 2 As shown. Finally, 6000 small-sized data sets were extracted from the 1000 complete data sets generated by forward modeling to form a dataset for training the passive source surface wave interference suppression network.

[0057] Using the aforementioned dataset, a passive source surface wave interference suppression network was trained via supervised learning. 200 training epochs were performed using the Adam optimizer and a learning rate of 2et. -3 The number of training samples per iteration is set to 24. The objective function is calculated using the mean squared error loss function (MSE), which can be expressed as:

[0058]

[0059] In the formula, b represents the number of samples for each training session; n represents the number of seismic traces; t represents the number of time sampling points; w is the updatable network parameter; X represents the virtual source shot gather data to be processed; S represents the label data, i.e., the clean surface wave data; and N represents the passive source surface wave interference suppression network.

[0060] Based on the network trained as described above, further label-free transfer learning training is performed on measured passive surface wave virtual source shot gathers. First, a trace mask is constructed, which covers a portion of the seismic trace data in each input to the network, creating receptive field blind spots. The input for this process uses small-sized, cropped measured data, and the output is the direct prediction result of the passive surface wave interference suppression network trained using supervised learning.

[0061] In this embodiment, it is first necessary to construct a portion of the seismic trace data from each input of the trace-direction masking network. This process is as follows: Figure 3 As shown, values ​​are randomly selected from a uniform distribution between [-0.2, 0.2] to form a 200×2 vector as a channel mask, randomly replacing two channels in the original input. A channel mask is randomly generated for each step of network training. Subsequently, L2 loss function is used for transfer learning training.

[0062]

[0063] In the formula, b represents the number of samples trained each time; w is the updatable network parameter; w pre Δw represents the network parameters obtained through supervised learning training; Δw represents the update amount of the network parameters after transfer learning; X represents the virtual source shot gather data to be processed; and N represents the passive source surface wave interference suppression network with added noise masks. The processed results of the trained network on the measured data are as follows: Figure 4 As shown.

[0064] The objective function does not need to be reduced as much as possible during the training process. Therefore, the number of training rounds should not be too many, around 40-50 rounds is sufficient. Furthermore, the network parameters obtained when the validation set loss function is minimized should not be selected as the final training result. Instead, the network obtained in the last round of training should be selected as the final training result.

[0065] The values ​​of the parameters in the above embodiments are merely exemplary values. In other embodiments, the values ​​of the above parameters can be adjusted according to specific circumstances, which are all things that those skilled in the art can easily conceive of.

[0066] Example 2

[0067] A passive source surface wave data regular interference suppression system includes:

[0068] The network building module is configured to build a passive source surface wave interference suppression network;

[0069] The dataset construction module is configured to randomly extract data containing surface wave signals from seismic data within different spatiotemporal domains according to the set time period and number of seismic traces, and construct a small-size surface wave interference suppression dataset.

[0070] The training module is configured to train the passive source surface wave interference suppression network in a supervised learning manner using the constructed dataset.

[0071] The trace mask module is configured to construct a trace mask, which is used to cover part of the seismic trace data in the input of the trained passive source surface wave interference suppression network.

[0072] The processing module is configured to perform unlabeled transfer learning training using the acquired passive source surface wave virtual source shot set to obtain processed data.

[0073] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0074] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0075] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0076] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0077] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0078] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for suppressing regular interference in passive source surface wave data, characterized in that, Includes the following steps: Construct a passive source surface wave interference suppression network; Based on the set time period and number of seismic traces, data containing surface wave signals from different spatiotemporal domains are randomly extracted from the seismic data to construct a small-size surface wave interference suppression dataset. Using the constructed dataset, a passive source surface wave interference suppression network is trained in a supervised learning manner. Construct a trace mask and use it to cover part of the seismic trace data in the input of the passive source surface wave interference suppression network after training. Using the above data, we performed unlabeled transfer learning training to obtain the processed data; When performing unlabeled transfer learning training on the measured passive source surface wave virtual source shot collection, two adjacent traces are randomly selected within the input seismic trace range, and a random vector of shape t×2 is randomly generated as a trace direction mask to replace the two selected trace data in the original input. In each training step, a trace direction mask is randomly generated, where t represents the number of time sampling points.

2. The passive source surface wave data regular interference suppression method as described in claim 1, characterized in that, The passive source surface wave interference suppression network consists of three parts: an encoder, skip connections, and a decoder. The encoder is composed of multiple convolutional network layers, the decoder is composed of multiple deconvolutional network layers, and the skip connections are used to connect some of the convolutional layers of the encoder and the deconvolutional layers of the decoder.

3. The passive source surface wave data regular interference suppression method as described in claim 1, characterized in that, The small-size surface wave interference suppression dataset includes input data and label data. The input data includes clean surface wave signals and known passive source surface wave interference, and the label data includes clean surface wave signals.

4. The passive source surface wave data regular interference suppression method as described in claim 1, characterized in that, In the small-size surface wave interference suppression dataset, the pure surface wave signal is obtained through Rayleigh wave field finite difference forward modeling of the surface active source.

5. The passive source surface wave data regular interference suppression method as described in claim 1, characterized in that, The passive source surface wave interference in the small-size surface wave interference suppression dataset includes volume wave interference and spurious in-phase axis interference.

6. A passive source surface wave data regular interference suppression method as described in claim 1 or 5, characterized in that, Surface wave data containing body wave interference can be improved by increasing the source depth during Rayleigh wave field finite difference forward modeling to increase the body wave interference signal in the data. Alternatively, surface wave data containing spurious in-phase axis interference can be obtained through passive source Rayleigh wave field finite difference forward modeling, i.e., by obtaining passive source simulation data through continuous excitation, and then recovering it into virtual source shot gather data through seismic interferometry.

7. A passive source surface wave data regular interference suppression method as described in claim 1 or 2, characterized in that, The passive source surface wave interference suppression network uses the mean square error loss function to calculate the objective function, which is expressed as: In the formula b represents the number of samples in each training iteration; n represents the number of seismic traces; w represents the updatable network parameters; X represents the virtual source shot gather data to be processed; S represents the labeled data, i.e., the clean surface wave data; and N represents the passive source surface wave interference suppression network.

8. The passive source surface wave data regular interference suppression method as described in claim 7, characterized in that, Based on the L2 loss function, an objective function is constructed. This objective function is then used for label-free transfer learning training on a measured passive source surface wave virtual source shot set. The objective function is: In the formula, b represents the number of samples per training; w is an updatable network parameter; w pre represents the network parameter obtained by supervised learning; Δw represents the update amount of the network parameter after transfer learning; X represents the virtual source shot gather data to be processed, and N represents the passive source surface wave interference suppression network with added noise masks.

9. A passive source surface wave data regularity interference suppression system, characterized in that, include: The network building module is configured to build a passive source surface wave interference suppression network; The dataset construction module is configured to randomly extract data containing surface wave signals from seismic data within different spatiotemporal domains according to the set time period and number of seismic traces, and construct a small-size surface wave interference suppression dataset. The training module is configured to train the passive source surface wave interference suppression network in a supervised learning manner using the constructed dataset. The trace mask module is configured to construct a trace mask, which is used to cover part of the seismic trace data in the input of the trained passive source surface wave interference suppression network. The processing module is configured to use the above data to perform unlabeled transfer learning training to obtain processed data; When performing unlabeled transfer learning training on the measured passive source surface wave virtual source shot collection, two adjacent traces are randomly selected within the input seismic trace range, and a random vector of shape t×2 is randomly generated as a trace direction mask to replace the two selected trace data in the original input. In each training step, a trace direction mask is randomly generated, where t represents the number of time sampling points.

Images