Methods, apparatus, electronic equipment and media for suppressing multimodal dispersive surface waves

Abstract

This disclosure relates to a method, apparatus, electronic device, and medium for suppressing multimodal dispersive surface waves, belonging to the field of data processing technology. The method includes: selecting key feature curves from a first part of the dispersive spectrum in a dataset corresponding to seismic data based on manual selection instructions, obtaining dispersive curve selection results for each order; using the first part of the dispersive spectrum as training data and the dispersive curve selection results for each order as training labels, performing supervised training on a dispersive curve processing model to be trained; based on the trained dispersive curve processing model, automatically selecting a second part of the dispersive spectrum to obtain probability distribution maps of dispersive curves for each order, and obtaining candidate dispersive curve point sets for each order after thresholding and non-maximum suppression; and performing dispersive surface wave prediction and suppression processing for each order based on the dispersive curve selection results and the candidate dispersive curve point sets for each order, obtaining the suppression processing results. This method can improve processing efficiency and accuracy.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing, and particularly relates to a multi-modal dispersion surface wave suppression method and device, electronic equipment and medium. BACKGROUND

[0002] Surface wave is a very common noise, which is characterized by low frequency, low speed and strong energy, and has frequency dispersion and multi-modal. In order to highlight the effective signals such as reflection wave and refraction wave, it is necessary to suppress and eliminate the surface wave noise. Suppression of surface wave noise and improvement of signal-to-noise ratio are one of the important links in seismic exploration data processing in complex areas. Strengthening the fine processing requirements such as denoising ability and amplitude preservation performance is particularly important in seismic data processing in complex structure area. How to effectively, accurately and quickly perform surface wave denoising is a key problem in seismic data processing.

[0003] In the process of implementing the present disclosure, the inventors have found that at least the following technical problems exist in the related art: With the successful application of wide-azimuth high-density exploration acquisition and efficient mixed acquisition technology, the current seismic exploration data has become a mass data body. The traditional dispersion imaging method cannot effectively improve the imaging resolution and high-order modal imaging quality, and cannot provide reliable data basis for the picking of multi-modal surface wave dispersion curves. The traditional surface wave suppression method can only process the dispersion energy of the strongest base modal, and cannot completely process the dispersion energy of each modal. The processing accuracy needs to be improved, and the processing efficiency is low. SUMMARY

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, embodiments of the present disclosure provide a multi-modal dispersion surface wave suppression method, device, electronic equipment and medium.

[0005] In a first aspect, embodiments of the present disclosure provide a multi-modal dispersion surface wave suppression method. The above method comprises: generating a dispersion spectrum data set according to seismic data collected in a target work area; picking a key feature curve based on a manual picking instruction for a first part of dispersion spectrum in the dispersion spectrum data set, to obtain a picking result of each order dispersion curve; taking the first part of dispersion spectrum as training data and taking the picking result of each order dispersion curve as training label, and performing supervised training on a dispersion curve processing model to be trained, the dispersion curve processing model being used to output a probability distribution map of each order dispersion curve; based on the trained dispersion curve processing model, automatically picking a second part of dispersion spectrum in the dispersion spectrum data set to obtain a corresponding probability distribution map of each order dispersion curve, and after thresholding and non-maximum suppression, obtaining a candidate dispersion curve point set of each order; based on the picking result of each order dispersion curve and the candidate dispersion curve point set of each order, performing each order dispersion surface wave prediction and suppression processing to obtain a suppression processing result.

[0006] In some embodiments, the dispersion curve processing model comprises an encoder network and a decoder network. The encoder network comprises a plurality of cascaded first convolutional layers, first non-linear activation layers, and a dimension reduction processing layer. The dimension reduction processing layer comprises at least one of a pooling layer and a bottleneck layer, and is configured to perform layer-by-layer feature extraction, non-linear activation processing, and feature dimension reduction processing on the dispersion spectrum grayscale image corresponding to the first part of the dispersion spectrum, and output an encoded dispersion feature map. The decoder network comprises a plurality of cascaded second convolutional layers, second non-linear activation layers, and an up-sampling layer, and is configured to perform image reconstruction processing, non-linear activation processing, and up-sampling processing on the encoded dispersion feature map, respectively, to obtain a dispersion curve probability distribution map of each order restored to the original resolution.

[0007] In some embodiments, the dispersion curve processing model is supervised trained based on a training label. The generation process of the training label comprises: for a training label with a label dimension of [label size, maximum order, frequency sampling point number], the speed value of each curve in the dispersion curve picking result in the training label is discretely mapped to the corresponding pixel position and a Gaussian weight is applied for expansion, a pixel mask with the same resolution as each image in the dispersion spectrum dataset is generated, and then bilinear interpolation resampling is performed to the original resolution to obtain a first supervision signal of the pixel graph type.

[0008] In some embodiments, the generation process further comprises: generating a second supervision signal according to a preset pixel space region in which the dispersion curve picking result is located, and the second supervision signal is used to limit the curve output range of the decoder network. In the process of supervised training, the binary cross-entropy loss is used as the loss function, and the output of the decoder network is supervised and trained at the pixel level based on the first supervision signal and the second supervision signal.

[0009] In some embodiments, the dispersion spectrum dataset is generated based on the seismic data collected in the target work area, comprising: preprocessing the seismic data collected in the target work area to obtain preprocessed seismic data; the preprocessing comprises one or more of the following: decoding, setting an observation system, static correction, amplitude recovery, single-frequency noise suppression, harmonic suppression, band-pass filtering, and normalization processing; selecting the seismic trace data corresponding to the target geophone as the sampling data for the preprocessed seismic data; performing Fourier transform on each seismic trace data in the sampling data to obtain the frequency domain result corresponding to each seismic trace data; and performing Bessel function transformation integration on the frequency domain result to obtain the dispersion spectrum dataset corresponding to each target geophone position.

[0010] In some embodiments, the method further comprises: determining the target receiver according to a set parameter, topographic information of the target work area, distribution information of the receivers, and current computing power; wherein the set parameter comprises at least one of the following: a set offset distance, a set azimuth angle, and a set maximum total sampling number.

[0011] In some embodiments, the seismic data is shot gather data; based on the picking results of the dispersion curves of different orders and the candidate dispersion curve point sets of different orders, dispersion surface wave prediction and suppression processing are performed to obtain a suppression processing result, including: performing interpolation processing on the candidate dispersion point sets of different orders according to the relative positions of the target receivers corresponding to the samples in the seismic data in the target work area and the grid setting information of the target work area, to obtain the modal dispersion curves of different orders; in order from low order to high order, the following prediction and suppression processing operations are performed in a loop to obtain the suppression processing result:

[0012] For seismic trace data in any spatial range or any azimuth angle range in the shot gather data in the target work area, dispersion correction of the current order is performed based on the picking results of the corresponding dispersion curves of different orders or the modal dispersion curves to obtain dispersion-corrected shot gather data;

[0013] The dispersion-corrected shot gather data is used to predict the surface wave data corresponding to the positions of each seismic trace, to obtain the dispersion surface wave model data of the current order;

[0014] The original shot gather data is subtracted from the predicted dispersion surface wave model data of the current order to obtain the best suppression processing result of the dispersion surface wave of the current order;

[0015] The best suppression processing result of the current order surface wave is used as the input of the next loop to perform prediction and suppression processing of the dispersion surface wave of a higher order, until the prediction and suppression processing of the dispersion surface wave of the highest order are completed, and the suppression processing result of all orders is obtained.

[0016] In a second aspect, embodiments of the present disclosure provide a device for multi-modal dispersive surface wave suppression. The device comprises a dataset generation module, an instruction picking module, a training module, an automatic picking module, and a suppression processing module. The dataset generation module is configured to generate a dispersion spectrum dataset based on seismic data collected from a target work area. The instruction picking module is configured to pick key feature curves based on artificial picking instructions for a first part of the dispersion spectrum in the dispersion spectrum dataset, to obtain picking results of dispersion curves of each order. The training module is configured to supervise training of a dispersion curve processing model to be trained by taking the first part of the dispersion spectrum as training data and taking the picking results of the dispersion curves of each order as training labels, and the dispersion curve processing model is configured to output probability distribution maps of dispersion curves of each order. The automatic picking module is configured to automatically pick a second part of the dispersion spectrum in the dispersion spectrum dataset based on the trained dispersion curve processing model, to obtain corresponding probability distribution maps of dispersion curves of each order, and to obtain candidate dispersion curve point sets of each order after thresholding and non-maximum suppression. The suppression processing module is configured to perform dispersion surface wave prediction and suppression processing based on the picking results of the dispersion curves of each order and the candidate dispersion curve point sets of each order, to obtain a suppression processing result.

[0017] In a third aspect, embodiments of the present disclosure provide an electronic device. The electronic device comprises a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is configured to store a computer program; and the processor is configured to execute the program stored in the memory to implement the method for multi-modal dispersive surface wave suppression.

[0018] In a fourth aspect, embodiments of the present disclosure provide a computer readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the method for multi-modal dispersive surface wave suppression.

[0019] The above technical solutions provided by the embodiments of the present disclosure have at least some or all of the following advantages:

[0020] The frequency dispersion curve processing model is trained based on a small amount of dispersion spectrum (corresponding to the first part of the dispersion spectrum) and manually picked labels, and automatic picking processing of a large amount of dispersion spectrum (corresponding to the second part of the dispersion spectrum) is performed based on the trained frequency dispersion curve processing model; based on the picking results of the above-mentioned frequency dispersion curves and the above-mentioned candidate frequency dispersion curve point sets, the surface wave prediction and suppression processing of each order is performed, the suppression processing result can be obtained efficiently and relatively accurately, the signal-to-noise ratio of the seismic data is improved, through the surface wave suppression technology, the energy of the surface wave can be effectively weakened, and the interference degree on the seismic record is reduced, so that the effective signal is more prominent; the resolution of the seismic record is improved, the surface wave suppression can eliminate the waveform distortion caused by the surface wave, and the waveform characteristics of the reflection wave are clearer and more accurate; the data interpretability can also be enhanced, after the surface wave is suppressed, the quality of the seismic profile is obviously improved, and the continuity and stability of the reflection wave are better; this enables the geology interpretation personnel to more intuitively observe and analyze the seismic reflection characteristics, so as to more accurately infer the underground geological structure and lithology change condition. BRIEF DESCRIPTION OF DRAWINGS

[0021] The accompanying drawings, which are incorporated into and form a part of the specification, illustrate an embodiment consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure.

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the accompanying drawings needed to be used in the embodiments or the related description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without creative labor.

[0023] Figure 1 A flowchart of a method for multi-modal dispersion surface wave suppression according to an embodiment of the present disclosure is schematically shown.

[0024] Figure 2 The structure and input-output flow diagram of a frequency dispersion curve processing model according to an embodiment of the present disclosure are schematically shown.

[0025] Figure 3 A schematic diagram of the training and use stages of a frequency dispersion curve processing model according to an embodiment of the present disclosure is schematically shown.

[0026] Figure 4 A schematic diagram of original single shot gather data according to an embodiment of the present disclosure is schematically shown.

[0027] Figure 5 A schematic diagram of high-precision multi-modal dispersion spectrum corresponding to the dispersion spectrum data set generated by processing the original single shot gather data according to an embodiment of the present disclosure is schematically shown.

[0028] Figure 6 Fig. 6 schematically shows a diagram of picking results of candidate dispersion curve points of each order according to an embodiment of the present disclosure.

[0029] Figure 7 Fig. 7 schematically shows a result of predicting a base-order modal surface wave from one single shot data using dispersion curves at each spatial position in the interpolated work area according to an embodiment of the present disclosure.

[0030] Figure 8 Fig. 8 schematically shows a result of suppressing the base-order modal surface wave model according to an embodiment of the present disclosure.

[0031] Figure 9 Fig. 9 schematically shows a result of predicting a first-order modal surface wave based on the base-order modal surface wave suppression according to an embodiment of the present disclosure.

[0032] Figure 10 Fig. 10 schematically shows a result of suppressing the first-order modal surface wave according to an embodiment of the present disclosure.

[0033] Figure 11 Fig. 11 schematically shows a structural block diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0034] To make the objectives, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be described clearly and completely below with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present disclosure.

[0035] A first exemplary embodiment of the present disclosure provides a method for multi-modal dispersion surface wave suppression.

[0036] Figure 1 Fig. 13 schematically shows a flowchart of the method for multi-modal dispersion surface wave suppression according to an embodiment of the present disclosure.

[0037] Reference Figure 1 As shown in Fig. 13, the method for multi-modal dispersion surface wave suppression provided by the embodiments of the present disclosure includes the following steps: S110-S150.

[0038] In step S110, dispersion spectrum data set is generated according to seismic data collected from a target work area.

[0039] The seismic data collected in the target area is shot gather data. In some embodiments, the seismic data is shot gather data. A shot gather data represents a set of signals recorded by all receivers after a single shot (source) is excited. Any shot gather data can cover the shot gather data corresponding to a single shot; or cover the results corresponding to multiple shots, i.e. cover multiple shot gather data.

[0040] In some embodiments, a dispersion spectrum dataset is generated according to the seismic data collected in the target area, including:

[0041] The seismic data collected in the target area is preprocessed to obtain preprocessed seismic data; the preprocessing includes one or more of the following: decoding, setting an observation system, static correction, amplitude recovery, single-frequency noise suppression, harmonic suppression, band-pass filtering, normalization processing;

[0042] For the preprocessed seismic data, seismic trace data corresponding to the target receiver is selected as sampling data;

[0043] For each seismic trace data in the sampling data, a Fourier transform is performed to obtain a frequency domain result corresponding to each seismic trace data; for example, the frequency domain result is frequency domain seismic data, which is represented as , represents the offset, represents the angular frequency;

[0044] The frequency domain result is integrated by a Bessel function transform to obtain a dispersion spectrum dataset corresponding to each target receiver position.

[0045] For example, the integration process is represented as follows:

[0046] , (1)

[0047] wherein, represents the dispersion spectrum dataset corresponding to each target receiver position; represents the spatial position; represents the wave number; represents the frequency domain seismic data; represents the offset, represents the angular frequency; represents the first type of zero-order Bessel function.

[0048] In some embodiments, the method further includes the step of determining the target receiver: according to the set parameters, the topographic information of the target area, the receiver distribution information and the current computing power, the target receiver is determined.

[0049] The set parameters include at least one of the following: a set offset distance, a set azimuth angle, and a set maximum total sampling number.

[0050] For example, under the condition of meeting the set parameters, a certain number (less than the set maximum total sampling number) of geophones are randomly selected as target geophones in the range interval of the set offset distance and the set azimuth angle, and these target geophones are adapted to the topographic information of the target work area, the geophone distribution information, and the current computing power. For example, the target work area range covers the edge area and the sparse area, and the target geophone number can be as large as possible in the terrain area with a sharp change in the ground surface if the computing power is sufficient; for the terrain area with a relatively flat ground surface, the target geophone number can be appropriately reduced if the computing power is limited.

[0051] In this embodiment, since the spatial position wave dispersion characteristics in the target work area are completely based on the near-surface medium, the dispersion characteristics have direction characteristics and spatial variation characteristics, and the selection of the seismic data at each control point is controlled by limiting the offset distance range, the azimuth angle range, and the number of seismic traces, which can reflect the stratum characteristics and facilitate subsequent dispersion analysis.

[0052] In step S120, the key feature curves are picked based on the artificial picking instruction for the first part of the dispersion spectrum in the dispersion spectrum data set, and the picking results of the dispersion curves of each order are obtained. For example, the picking process is realized through human-computer interaction, which is a manual picking process.

[0053] In step S130, the first part of the dispersion spectrum is used as training data, the picking results of the dispersion curves of each order are used as training labels, and the dispersion curve processing model to be trained is supervisedly trained, and the dispersion curve processing model is used to output the probability distribution map of the dispersion curves of each order.

[0054] Figure 2 The structure and input-output flow diagram of the dispersion curve processing model according to an embodiment of the present disclosure are schematically shown. Figure 3 The schematic diagram of the training and use stages of the dispersion curve processing model according to an embodiment of the present disclosure is schematically shown.

[0055] Referring to Figure 2 and Figure 3 The dispersion curve processing model includes an encoder network and a decoder network.

[0056] In some embodiments, the encoder network comprises a plurality of cascaded first convolutional layers, first non-linear activation layers, and dimension reduction processing layers, the dimension reduction processing layers comprise at least one of a pooling layer and a bottleneck layer, and are configured to perform layer-by-layer feature extraction, non-linear activation processing, and feature dimension reduction processing on the first part of the frequency dispersion spectrum gray image, and output an encoded frequency dispersion feature map. In the same cascade layer, the first convolutional layer and the dimension reduction processing layer have a sequence, and the dimension reduction processing layer is after the first convolutional layer. The first non-linear activation layer can be after the first convolutional layer or after the dimension reduction processing layer.

[0057] For example, the encoder network receives a normalized and bilinear interpolation resampled frequency dispersion spectrum gray image with a resolution of 512x512 pixels and a channel number of 1. Through four layers of convolutional layer-bottleneck layer (or pooling layer) cascaded processing, the spatial size is down-sampled by 16 times layer by layer, and a 32x32x128-dimensional high-order abstract feature map is output at the bottleneck layer, which is the encoded frequency dispersion feature map.

[0058] In some embodiments, the decoder network comprises a plurality of cascaded second convolutional layers, second non-linear activation layers, and up-sampling layers, and is configured to perform image reconstruction processing, non-linear activation processing, and up-sampling processing on the encoded frequency dispersion feature map to obtain a restored original resolution of each order of frequency curve probability distribution map. In the same cascade layer, the second convolutional layer and the up-sampling layer have a sequence, and the up-sampling layer is after the second convolutional layer. The second non-linear activation layer can be after the second convolutional layer or after the up-sampling layer.

[0059] For example, in the decoder network, based on four layers of convolution-up-sampling cascaded processing, the spatial resolution is gradually restored to 512x512, the channel number is 1, and the output end is activated by Sigmoid to generate a frequency curve probability distribution map with the same size as the input, i.e., the value in the final obtained frequency curve probability distribution map is limited in the interval [0, 1].

[0060] By setting the activation function after the convolutional layer (including the first convolutional layer and the second convolutional layer) as ReLU (non-linear activation layer), non-linear characteristics are introduced, so that the model can learn more complex features.

[0061] In some embodiments, the dispersion curve processing model is supervised trained, and the corresponding supervision signal is generated based on the training label, and the generation process includes: for the training label with the label dimension of [label size label size, maximum order max order (the general order value can be a natural number less than or equal to the maximum order, and the order 0 represents the base order mode), frequency sampling point number nf], the speed value of each curve in the picking result of each order dispersion curve in the training label is discretely mapped to the corresponding pixel position and is expanded by applying Gaussian weight, a pixel mask with the same resolution as each graph of the dispersion spectrum data set is generated, and then bilinear interpolation resampling is performed to the original resolution (for example, 512x512) to obtain a first supervision signal of the pixel graph type.

[0062] In some embodiments, the generation process further includes: generating a second supervision signal according to a preset pixel space region where the picking result of each order dispersion curve is located, and the second supervision signal is used to limit the curve output range of the decoder network. In the process of supervised training, the binary cross-entropy loss is used as the loss function, and the output of the decoder network is supervised and trained at the pixel level based on the first supervision signal and the second supervision signal.

[0063] In some embodiments, only the first supervision signal is used for pixel-level supervised training, which helps to improve the prediction accuracy of the dispersion curve processing model in image generation prediction. In other embodiments, the first supervision signal is used for pixel-level supervised training and the second supervision signal is used for region restriction constraint, which helps to improve the prediction accuracy and training iteration efficiency of the dispersion curve processing model in image generation prediction.

[0064] In some embodiments, the learning rate of the parameters is adjusted based on the Adam optimizer (a widely used adaptive learning rate optimization algorithm in deep learning), and the encoder network and the decoder network in the dispersion curve processing model are trained together based on the binary cross-entropy loss as the loss function.

[0065] In step S140, based on the trained dispersion curve processing model, the second part of the dispersion spectrum in the dispersion spectrum data set is automatically picked to obtain the corresponding probability distribution graph of each order dispersion curve, and after thresholding and non-maximum suppression, the candidate dispersion curve point set of each order is obtained.

[0066] In the dispersion spectrum data set, a small part (the first part of the dispersion data) is used as training data and model verification data, and the remaining large amount of data (for example, the second part of the dispersion data) is used as to-be-processed data for subsequent automatic picking. The ratio of the first part of the dispersion data to the second part of the dispersion data can be controlled between 1:100 and 1:1000, or dynamically adjusted according to the actual training effect of the model, etc.

[0067] Reference Figure 3 As shown in the usage phase (the lower half of the data flow), after the dispersion curve processing model is trained, it can automatically pick up a large number of second-part dispersion spectra in the above dispersion spectrum dataset, which greatly improves the probability of extracting the modal dispersion curves corresponding to the probability distribution maps of each order dispersion curve (this curve is not the direct output of the above model, but is obtained after interpolation processing of the model output).

[0068] The purpose of thresholding includes: extracting significant features (such as high-frequency noise, tomographic response, etc.) from the spectrum or time-frequency energy distribution, and filtering noise or weak signals by setting a threshold.

[0069] The purpose of nonmaximum suppression includes: suppressing nonmaximum points in the time-frequency domain or spatial domain, and preserving local extrema (such as peak values ​​of the same phase axis and fault edges).

[0070] In step S150, based on the above-mentioned dispersion curve picking results and the above-mentioned candidate dispersion curve point sets, dispersion surface wave prediction and suppression processing are performed to obtain the suppression processing results.

[0071] The following example illustrates the processing procedure for a single-shot cluster.

[0072] In some embodiments, in step S150 above, based on the above-mentioned dispersion curve picking results and the above-mentioned candidate dispersion curve point sets, dispersion surface wave prediction and suppression processing of each order are performed to obtain suppression processing results, including:

[0073] Based on the relative positions of the target receiver points sampled in the above seismic data within the above target work area and the grid setting information of the above target work area, interpolation processing is performed on the above candidate frequency dispersion point sets of each order to obtain the mode dispersion curves of each order.

[0074] The prediction and suppression operations are performed cyclically from low to high order to obtain the suppression result.

[0075] As an example, the prediction and suppression processing operations are performed cyclically from low to high order, including:

[0076] For the seismic trace data within any spatial range or any azimuth range in the shot gather data of the above target work area, the current order dispersion correction is performed based on the corresponding dispersion curve picking results or modal dispersion curves of each order to obtain the dispersion-corrected shot gather data.

[0077] By utilizing the coherence characteristics of the waveform, the surface wave data corresponding to the location of each seismic trace in the above dispersion-corrected shot gather data are predicted to obtain the dispersion surface wave model data of the current order.

[0078] Subtracting the predicted dispersion surface wave model data of the current order from the original shot gather data, a best suppression processing result of the current order dispersion surface wave is obtained.

[0079] Wherein, the best suppression processing result of the current order dispersion surface wave is taken as the input of the next cycle to perform the prediction and suppression processing operation of the higher order dispersion surface wave until the prediction and suppression processing of the highest order dispersion surface wave is completed, and the suppression processing result of all orders is obtained.

[0080] Figure 4 A schematic diagram of the original single shot gather data according to an embodiment of the present disclosure is shown, and the unit of time is second. Figure 5 A schematic diagram of the high-precision multi-modal dispersion spectrum corresponding to the dispersion spectrum data set generated by processing the original single shot gather data according to an embodiment of the present disclosure is shown. Figure 6 A schematic diagram of the picking result of the candidate dispersion curve point set of each order obtained after step S140 based on the trained dispersion curve processing model according to an embodiment of the present disclosure is shown. Figure 7 A result of predicting a single shot data by using the dispersion curve at each spatial position in the interpolated work area according to an embodiment of the present disclosure is shown. Figure 8 A result of suppressing the base order modal surface wave model according to an embodiment of the present disclosure is shown. Figure 9 A result of predicting the first order modal surface wave based on the base order modal surface wave suppression according to an embodiment of the present disclosure is shown. Figure 10 A result of suppressing the first order modal surface wave according to an embodiment of the present disclosure is shown.

[0081] Referring to Figure 4 It is shown that the original single shot gather data of a certain actual work area is selected, and for the original single shot gather data, 10,000 sampling points (i.e. target geophones) are selected to generate a dispersion spectrum data set, and a high-precision dispersion spectrum is obtained. Figure 5 The horizontal coordinate is frequency and the vertical coordinate is phase velocity. Then, in step S120, about 150 dispersion spectrums are manually picked as labels, and two modal dispersion curves in each spectrum are picked. In step S130, based on the labels, the dispersion curve processing model is supervised trained to obtain the trained dispersion curve processing model. In step S140, based on the trained dispersion curve processing model, the remaining dispersion spectrums are automatically picked to obtain the corresponding dispersion curve probability distribution diagram of each order, and after thresholding and non-maximum suppression, the candidate dispersion curve point set of each order is obtained, and the picking result is shown in Figure 6As shown. In step S150, interpolation processing is performed on the candidate dispersion point sets of each order to obtain the mode dispersion curves of each order. Then, in order from low to high order, prediction and suppression processing operations are performed cyclically to obtain the suppression processing results, as shown in the figure. Figure 7 The diagram illustrates the results of predicting the fundamental mode surface wave from a single-shot dataset using the interpolated dispersion curves at various spatial locations within the work area; (Refer to...) Figure 8 The diagram illustrates the optimal suppression result for the fundamental-order dispersive surface wave. Suppression of higher-order dispersive surface waves is then performed, referring to... Figure 9 The diagram illustrates the results of predicting a first-order modal surface wave based on the suppression of the fundamental modal surface wave; refer to... Figure 10 The figure shows the result after suppressing the first-order modal surface wave.

[0082] contrast Figure 10 and Figure 4 As can be seen, the surface wave suppression method provided in this disclosure can efficiently and relatively accurately obtain suppression processing results, improve the signal-to-noise ratio and resolution of seismic data, and also enhance the interpretability of the data.

[0083] In summary, in the embodiment including steps S110-S150, a dispersion curve processing model is trained based on a small amount of dispersion spectrum (corresponding to the first part of the dispersion spectrum) and manually picked labels. Simultaneously, based on the trained dispersion curve processing model, a large amount of dispersion spectrum (corresponding to the second part of the dispersion spectrum) is automatically picked. Thus, based on the picking results of each order of dispersion curves and the set of candidate dispersion curve points, surface wave prediction and suppression processing of each order can be performed, resulting in efficient and relatively accurate suppression processing. This improves the signal-to-noise ratio of seismic data. Surface wave suppression technology can effectively weaken the energy of surface waves, reducing their interference with seismic records and making the effective signal more prominent. It also improves the resolution of seismic records, as surface wave suppression can eliminate waveform distortion caused by surface waves, making the waveform characteristics of reflected waves clearer and more accurate. Furthermore, it enhances the interpretability of the data; after suppressing surface waves, the quality of the seismic profile is significantly improved, and the continuity and stability of reflected waves are better. This allows geological interpreters to more intuitively observe and analyze seismic reflection characteristics, thereby more accurately inferring underground geological structures and lithological changes.

[0084] A second exemplary embodiment of this disclosure provides an apparatus for suppressing multimodal dispersion surface waves.

[0085] The aforementioned device includes: a dataset generation module, an instruction picking module, a training module, an automatic picking module, and a compression processing module.

[0086] The dataset generation module described above is used to generate a dispersion spectrum dataset based on the seismic data collected from the target work area.

[0087] The instruction picking module is configured to pick up the key feature curve based on the artificial picking instruction for a first part of the dispersion spectrum in the dispersion spectrum dataset, and obtain a picking result of each order dispersion curve.

[0088] The training module is configured to supervise the training of the dispersion curve processing model to be trained by taking the first part of the dispersion spectrum as training data and taking the picking result of each order dispersion curve as a training label, and the dispersion curve processing model is configured to output a probability distribution map of each order dispersion curve.

[0089] The automatic picking module is configured to automatically pick up a second part of the dispersion spectrum in the dispersion spectrum dataset based on the trained dispersion curve processing model, obtain a corresponding probability distribution map of each order dispersion curve, and obtain a candidate point set of each order dispersion curve after thresholding and non-maximum suppression.

[0090] The suppression processing module is configured to perform each order dispersion wave prediction and suppression processing based on the picking result of each order dispersion curve and the candidate point set of each order dispersion curve, and obtain a suppression processing result.

[0091] In some embodiments, the dispersion curve processing model comprises an encoder network and a decoder network. The encoder network comprises a plurality of cascaded first convolutional layers, first non-linear activation layers, and dimension reduction processing layers. The dimension reduction processing layers comprise at least one of a pooling layer and a bottleneck layer, are configured to perform layer-by-layer feature extraction, non-linear activation processing, and feature dimension reduction processing on a dispersion spectrum grayscale image corresponding to the first part of the dispersion spectrum, and output an encoded dispersion feature map. The decoder network comprises a plurality of cascaded second convolutional layers, second non-linear activation layers, and up-sampling layers, is configured to perform image reconstruction processing, non-linear activation processing, and up-sampling processing on the encoded dispersion feature map, and obtain a probability distribution map of each order dispersion curve restored to an original resolution.

[0092] The training module comprises a supervision signal generation sub-module.

[0093] In some embodiments, the supervision signal generated based on the training label is used for supervised training of the dispersion curve processing model.

[0094] The supervision signal generation sub-module is configured to: for a training label with a label dimension of [label size, maximum order, frequency sampling point number], discretely map a velocity value of each curve in the picking result of each order dispersion curve in the training label to a corresponding pixel position and apply a Gaussian weight expansion, generate a pixel mask with the same resolution as each image of the dispersion spectrum dataset, and then perform bilinear interpolation resampling to an original resolution to obtain a first supervision signal of a pixel graph type.

[0095] The supervision signal generation submodule is further configured to generate a second supervision signal according to a preset pixel space region in which the picked result of each order dispersion curve is located, and the second supervision signal is used to limit the curve output range of the decoder network. In the process of supervised training, a binary cross-entropy loss is taken as a loss function, and pixel-level supervised training is performed on the output of the decoder network based on the first supervision signal and the second supervision signal.

[0096] In some embodiments, the data set generation module includes a preprocessing submodule, a sampling submodule, a frequency domain transformation submodule, and an integration submodule.

[0097] The preprocessing submodule is configured to preprocess the seismic data collected in the target work area to obtain preprocessed seismic data, and the preprocessing includes one or more of the following: decoding, setting an observation system, static correction, amplitude recovery, single-frequency noise suppression, harmonic suppression, band-pass filtering, and normalization processing.

[0098] The sampling submodule is configured to select, for the preprocessed seismic data, seismic trace data corresponding to a target geophone as sampling data.

[0099] The frequency domain transformation submodule is configured to perform Fourier transform on each seismic trace data in the sampling data to obtain a frequency domain result corresponding to each seismic trace data.

[0100] The integration submodule is configured to perform Bessel function transformation integration on the frequency domain result to obtain a dispersion spectrum data set corresponding to each target geophone position.

[0101] In some embodiments, the sampling submodule is further configured to determine the target geophone according to a set parameter, topographic information of the target work area, geophone distribution information, and current computing power, and the set parameter includes at least one of the following: a set offset distance, a set azimuth angle, and a set maximum total sampling number.

[0102] In some embodiments, the suppression processing module includes an interpolation processing submodule and a cyclic suppression processing submodule.

[0103] The interpolation processing submodule is configured to perform interpolation processing on each order candidate dispersion point set according to the relative position of the sampling corresponding target geophone in the target work area in the seismic data and grid setting information of the target work area to obtain each order modal dispersion curve.

[0104] The cyclic suppression processing submodule is configured to perform prediction and suppression processing operations in a cyclic manner in an order from low order to high order to obtain a suppression processing result.

[0105] In order from low order to high order, the prediction and suppression processing operations are performed in a loop, including:

[0106] For seismic trace data in any spatial range or any azimuth angle range in the shot gather data in the target work area, based on the picking result of the corresponding dispersion curve of each order or the modal dispersion curve of each order, the current order dispersion correction is performed to obtain the dispersion corrected shot gather data;

[0107] Using the coherence feature of the waveform, the surface wave data corresponding to the position of each seismic trace in the dispersion corrected shot gather data is predicted to obtain the dispersion surface wave model data of the current order;

[0108] The original shot gather data is subtracted from the dispersion surface wave model data of the current order obtained by prediction to obtain the best suppression processing result of the dispersion surface wave of the current order;

[0109] Wherein, the best suppression processing result of the current order surface wave is taken as the input of the next cycle to perform the prediction and suppression processing operation of the dispersion surface wave of the higher order, until the prediction and suppression processing of the dispersion surface wave of the highest order is completed, and the suppression processing result of all orders is obtained.

[0110] More details of the embodiment can be referred to the related description of the first embodiment, which will not be repeated here.

[0111] The device provided by the embodiment can adapt to its real-time computing power to efficiently suppress surface waves, and the suppression processing result obtained has higher accuracy, which can effectively improve the signal-to-noise ratio and resolution of seismic data, and also can enhance the interpretability of data.

[0112] Any multiple of the functional modules included in the above device can be combined in one module, or any one of the modules can be split into multiple modules. Alternatively, at least part of the function of one or more of the modules can be combined with at least part of the function of the other modules, and implemented in one module. At least one of the functional modules included in the above device can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on chip, a system on board, a system on package, an application specific integrated circuit (ASIC), or any other reasonable way of hardware or firmware by integrating or packaging the circuit, or in any one of software, hardware and firmware or in any appropriate combination of any of them. Alternatively, at least one of the functional modules included in the above device can be at least partially implemented as a computer program module which can perform corresponding functions when running.

[0113] A third exemplary embodiment of the present disclosure provides an electronic device.

[0114] Figure 11 A structural block diagram of an electronic device is shown schematically.

[0115] Referring to Figure 11 As shown in the figure, the electronic device 1100 provided by the embodiment of the present disclosure includes a processor 1101, a communication interface 1102, a memory 1103 and a communication bus 1104, wherein the processor 1101, the communication interface 1102 and the memory 1103 complete the communication among each other through the communication bus 1104; the memory 1103 is used for storing a computer program; the processor 1101 is used for executing the program stored on the memory, and realizes the method for suppressing multi-modal dispersive surface waves as described above.

[0116] The fourth exemplary embodiment of the present disclosure also provides a computer readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the method for suppressing multi-modal dispersive surface waves as described above.

[0117] The computer readable storage medium can be included in the device or apparatus described in the above embodiments; or it can exist separately and is not assembled into the device or apparatus. The computer readable storage medium carries one or more programs, and when the one or more programs are executed, the method according to the embodiment of the present disclosure is realized.

[0118] According to the embodiment of the present disclosure, the computer readable storage medium can be a non-volatile computer readable storage medium, which can include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer readable storage medium can be any tangible medium containing or storing a program, which can be used by or in conjunction with an instruction execution system, device or apparatus.

[0119] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0120] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method of multimodal dispersion surface wave suppression, characterized by, The method comprises the following steps: According to the seismic data collected in the target work area, a dispersion spectrum data set is generated; For the first part of the dispersion spectrum data set, the key feature curve is picked up based on the artificial picking instruction, and the dispersion curve picking result of each order is obtained; The first part of the dispersion spectrum is taken as the training data, and the dispersion curve picking result of each order is taken as the training label. The dispersion curve processing model is supervised trained, and the dispersion curve processing model is used to output the dispersion curve probability distribution graph of each order. Based on the trained dispersion curve processing model, the second part of the dispersion spectrum in the dispersion spectrum data set is automatically picked up, and the corresponding dispersion curve probability distribution graph of each order is obtained. After thresholding and non-maximum suppression, the dispersion curve point set of each order is obtained. Based on the dispersion curve picking result of each order and the dispersion curve point set of each order, the dispersion surface wave prediction and suppression processing are carried out, and the suppression processing result is obtained. The dispersion curve processing model comprises: The encoder network comprises a plurality of cascaded first convolutional layers, first nonlinear activation layers and dimension reduction processing layers. The dimension reduction processing layer comprises at least one of the following: a pooling layer and a bottleneck layer, which is used for layer-by-layer feature extraction, nonlinear activation processing and feature dimension reduction processing on the dispersion spectrum gray image corresponding to the first part of the dispersion spectrum, and outputs the encoded dispersion feature map; The decoder network comprises a plurality of cascaded second convolutional layers, second nonlinear activation layers and up-sampling layers, which are used for image reconstruction processing, nonlinear activation processing and up-sampling processing on the encoded dispersion feature map, respectively, to obtain the dispersion curve probability distribution graph of each order restored to the original resolution; The corresponding supervision signal of the supervised training of the dispersion curve processing model is generated based on the training label. The generation process comprises: For the training label with the label dimension of [label size, maximum order number, frequency sampling point number], the velocity value of each curve in the dispersion curve picking result in the training label is discretely mapped to the corresponding pixel position and is expanded with Gaussian weight. A pixel mask with the same resolution as each image of the dispersion spectrum data set is generated. After bilinear interpolation resampling to the original resolution, a pixel graph type first supervision signal is obtained. According to the preset pixel space region where the dispersion curve picking result is located, a second supervision signal is generated. The second supervision signal is used to limit the curve output range of the decoder network.

2. The method of claim 1, wherein, In the process of supervised training, the binary cross entropy loss is taken as the loss function, and the output of the decoder network is supervised trained based on the first supervision signal and the second supervision signal. According to the seismic data collected in the target work area, a dispersion spectrum data set is generated, comprising:

3. The method of claim 1, wherein, The seismic data collected in the target work area is preprocessed to obtain preprocessed seismic data. The preprocessing comprises one or more of the following: decoding, setting the observation system, static correction, amplitude recovery, single-frequency noise suppression, harmonic suppression, band-pass filtering, normalization processing; For the preprocessed seismic data, the seismic data corresponding to the target geophone is selected as the sampling data; For each seismic data in the sampling data, Fourier transform is performed to obtain the frequency domain result corresponding to each seismic data. ​ The Bessel function transformation integral is performed on the frequency domain result to obtain a dispersion spectrum data set corresponding to each target geophone position.

4. The method of claim 3, wherein, Further comprising: The target geophone is determined according to the set parameters, the topographic information of the target work area, the geophone distribution information and the current computing power, wherein the set parameters include at least one of the following: set offset, set azimuth, set maximum total sampling number.

5. The method according to any one of claims 1 to 4, characterized in that, The seismic data is shot gather data; Based on the picking results of the dispersion curves of different orders and the candidate dispersion point sets of different orders, dispersion surface wave prediction and suppression processing are performed to obtain a suppression processing result, including: According to the relative position of the target geophone corresponding to the sampling in the target work area and the grid setting information of the target work area, the interpolation processing is performed on the candidate dispersion point sets of different orders to obtain the modal dispersion curves of different orders; In the order from low order to high order, the following prediction and suppression processing operations are performed in a loop to obtain the suppression processing result: For the seismic trace data in any spatial range or any azimuth range in the shot gather data in the target work area, based on the picking results of the corresponding dispersion curves of different orders or the modal dispersion curves of different orders, the current order dispersion correction is performed to obtain the dispersion corrected shot gather data; The coherent features of the waveforms are used to predict the surface wave data corresponding to the positions of each seismic trace in the dispersion corrected shot gather data to obtain the dispersion surface wave model data of the current order; The original shot gather data is subtracted by the dispersion surface wave model data of the current order obtained by prediction to obtain the best suppression processing result of the dispersion surface wave of the current order; Wherein, the best suppression processing result of the current order surface wave is taken as the input of the next loop to perform the prediction and suppression processing operation of the higher order dispersion surface wave, until the prediction and suppression processing of the highest order dispersion surface wave is completed, and the suppression processing result of all orders is obtained.

6. An electronic device, comprising: The processor, the communication interface, the memory and the communication bus are used to complete the communication among each other; The memory is used to store the computer program; The processor is used to execute the program stored on the memory to realize the method of any one of claims 1-5.

7. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the method of any one of claims 1-5.

Images