Seismic data dimension reduction network training and seismic data processing method, device and system

Abstract

The invention discloses a seismic data dimension reduction network training method, device and system and a seismic data processing method, device and system. The method comprises the following steps: selecting pre-generated multi-channel fusion data, and respectively inputting the pre-generated multi-channel fusion data into at least two generative networks with the same structure to obtain at least two groups of simulated seismic data; and inputting the simulated seismic data and the real seismic data into the discriminant network, training the generative network and the discriminant network through a preferential competition mechanism to obtain a trained discriminant network and a trained network, processing the discriminant network to obtain a dimensionality reduction network, and training the dimensionality reduction network by using the multi-channel fusion data and the simulated seismic data. The trained dimension reduction network is used to carry out dimension reduction processing on real seismic data, features of the seismic data can be learned, real and effective dimension reduction processing can be carried out, the data features cannot be lost, the data after dimension reduction processing cannot be distorted, and the generation network can be used to restore the real seismic data. And dimensionality reduction compression and dimensionality raising decompression of real seismic data are well realized.

Description

Technical Field

[0001] The present invention relates to the field of seismic exploration technology, and in particular to seismic data dimension reduction network training and seismic data processing methods, devices and systems. Background Art

[0002] With the continuous increase in the exploration of oil and gas resources, the number of three-dimensional artificial seismic deployments has gradually increased. The rapid development of seismic exploration in terms of multi-dimensionality, multi-wave and multi-component, and high resolution has led to a sharp increase in the hardware resources consumed by seismic data storage. It is necessary to compress seismic data to reduce hardware resource consumption. On the other hand, with the increase in production demand, some areas with complex surfaces and relatively poor seismic data acquisition effects have gradually entered the exploration stage. For example, in the exploration of the western region with many mountains and deserts, the resolution, fidelity, and signal-to-noise ratio of seismic data are not easy to be effectively guaranteed. Therefore, it is very necessary to more effectively preserve the characteristics of seismic data while compressing the data.

[0003] In recent years, one of the more representative studies on seismic data compression algorithms is wavelet transform. The idea of ​​wavelet transform is to use a wavelet function system to represent or approximate a signal or function. The wavelet function system is constructed by translating or stretching a basic wavelet at different scales. The wavelet transform method is used to decompose seismic data into a series of components, remove the components that contribute less to the original data, and then encode them to achieve the purpose of data compression. Summary of the invention

[0004] In the existing scheme of compressing raw seismic data using wavelet transform, in principle, since some components of the raw data are removed, some features will inevitably be lost in the compression process, causing the compressed data to be distorted.

[0005] In view of the above problems, the present invention is proposed to provide a method and device for dimensionality reduction and compression of seismic data that overcomes the above problems or at least partially solves the above problems.

[0006] The embodiment of the present invention provides a seismic data dimension reduction network training method, comprising:

[0007] Select pre-generated multi-channel fusion data, and input them into at least two generating networks with the same structure respectively, to obtain at least two sets of simulated seismic data; the multi-channel fusion data is obtained by fusing a multi-channel Gaussian signal and a multi-channel random noise signal, the multi-channel Gaussian signal is obtained by performing domain transformation and dimension upgrading on a one-dimensional signal that conforms to the Gaussian distribution; the multi-channel random noise signal is obtained by performing dimension upgrading on random noise;

[0008] Inputting at least two groups of simulated seismic data and the collected real seismic data into the discriminant network for comparison and discrimination respectively, and obtaining a discrimination result characterizing the degree of matching between the simulated seismic data and the real seismic data; updating the network parameters of the generating network and / or updating the network parameters of the discriminant network based on the preferential competition mechanism according to the discrimination result, until a preset number of training times is reached, and a trained generating network and discriminant network are obtained; the discriminant network includes a data analysis and processing layer and a classification layer;

[0009] On the basis of the data analysis and processing layer in the trained discriminant network, a dimensionality reduction processing layer is added to obtain a seismic data dimensionality reduction network; the seismic simulation data generated by the generation network is used as the input of the dimensionality reduction network, and the one-dimensional seismic signal after dimensionality reduction is output. According to the one-dimensional seismic signal and the one-dimensional signal corresponding to the seismic simulation data, the dimensionality reduction network is trained to meet the preset convergence, so as to obtain a trained seismic data dimensionality reduction network.

[0010] In some preferred embodiments, the above-mentioned seismic data dimension reduction network training method further includes:

[0011] The one-dimensional signal that conforms to the Gaussian distribution is processed by changing the domain and increasing the dimension to obtain a multi-channel Gaussian signal;

[0012] The random noise signal is processed by dimension upgrading to obtain a multi-channel random noise signal;

[0013] Multi-channel Gaussian signals and multi-channel random noise signals are fused to obtain multi-channel fused data.

[0014] In some preferred embodiments, the at least two groups of simulated seismic data and the collected real seismic data are respectively input into the discriminant network for comparison and discrimination, and a discrimination result characterizing the degree of matching between the simulated seismic data and the real seismic data is obtained, and the network parameters of the generating network and / or the network parameters of the discriminant network are updated based on the preferential competition mechanism according to the discrimination result, including:

[0015] Input at least two groups of simulated seismic data into the discriminant network and compare them with the acquired real seismic data respectively;

[0016] Obtaining at least two discrimination results of the simulated seismic data according to the matching degree between the simulated seismic data and the real seismic data, wherein the discrimination results include matching or mismatching with the real seismic data;

[0017] If the judgment result of one of the simulated seismic data matches the real seismic data, the network parameters of the generated network corresponding to the simulated seismic data are used to update the network parameters of the other generated networks;

[0018] If the discrimination results of the simulated seismic data are all matched or mismatched with the real seismic data, the network parameters of at least two generating networks and the network parameters of the discriminating network are updated according to a preset parameter updating rule.

[0019] In some preferred embodiments, the above method further comprises:

[0020] A sample database is established based on the simulated seismic data generated by the trained generative network and the corresponding multi-channel fusion data.

[0021] In some preferred embodiments, the seismic simulation data generated by the generation network is used as the input of the dimension reduction network, and the one-dimensional seismic signal after dimension reduction is outputted. According to the one-dimensional seismic signal and the one-dimensional signal corresponding to the seismic simulation data, the dimension reduction network is trained to meet the preset convergence, and the trained seismic data dimension reduction network is obtained, including:

[0022] Obtain earthquake simulation data generated by the generation network from the sample database, input it into the earthquake data dimension reduction network, and output the one-dimensional earthquake signal after dimension reduction;

[0023] According to the one-dimensional seismic signal after dimensionality reduction and the one-dimensional signal in the multi-channel fusion data corresponding to the simulated seismic data, the signal error of the one-dimensional seismic signal and the one-dimensional signal in the multi-channel fusion data is determined. If the signal error does not meet the preset signal error requirement, the step of obtaining the seismic simulation data generated by the generation network from the sample database is continued; until the signal error meets the preset signal error requirement, it is considered that the seismic data dimensionality reduction network converges, and a trained seismic data dimensionality reduction network is obtained.

[0024] An embodiment of the present invention provides a seismic data processing method, comprising:

[0025] Input the three-dimensional seismic data to be processed into the trained seismic data dimensionality reduction network, and output the one-dimensional seismic data after dimensionality reduction and compression;

[0026] The seismic data dimensionality reduction network is trained using the seismic data dimensionality reduction network training method described above.

[0027] In some preferred embodiments, a seismic data processing method further includes:

[0028] The one-dimensional seismic data after dimensionality reduction and compression is input into the trained generative network, and the three-dimensional seismic data after dimensionality increase and decompression is output.

[0029] The embodiment of the present invention provides a seismic data dimension reduction network training device, comprising:

[0030] The data simulation module is used to select pre-generated multi-channel fusion data and input them into at least two generating networks with the same structure respectively to obtain at least two sets of simulated seismic data; the multi-channel fusion data is obtained by fusing a multi-channel Gaussian signal and a multi-channel random noise signal, the multi-channel Gaussian signal is obtained by performing domain transformation and dimension upgrading on a one-dimensional signal that conforms to the Gaussian distribution; the multi-channel random noise signal is obtained by performing random noise dimension upgrading;

[0031] A discrimination module, used to input at least two groups of simulated seismic data and the collected real seismic data into a discrimination network for comparison and discrimination, and obtain a discrimination result representing the degree of matching between the simulated seismic data and the real seismic data;

[0032] A first network training module is used to update the network parameters of the generating network and / or the network parameters of the discriminating network based on the optimal competition mechanism according to the discrimination result, until a preset number of training times is reached to obtain a trained generating network and discriminating network; the discriminating network includes a data analysis and processing layer and a classification layer;

[0033] The second network training module is used to add a dimensionality reduction processing layer on the basis of the data analysis processing layer in the trained discriminant network to obtain a seismic data dimensionality reduction network; use the seismic simulation data generated by the generation network as the input of the dimensionality reduction network, output the reduced one-dimensional seismic signal, and determine the training of the dimensionality reduction network to meet the preset convergence according to the one-dimensional seismic signal and the one-dimensional signal corresponding to the seismic simulation data, so as to obtain a trained seismic data dimensionality reduction network.

[0034] The present invention provides a seismic data processing device, comprising:

[0035] Dimensionality reduction processing module and / or dimensionality increase processing module;

[0036] A dimensionality reduction processing module is used to input the three-dimensional seismic data to be processed into a trained seismic data dimensionality reduction network, and output one-dimensional seismic data after dimensionality reduction and compression; the seismic data dimensionality reduction network is trained using the seismic data dimensionality reduction network training method as described above;

[0037] The dimension-upgrading processing module is used to input the one-dimensional seismic data after dimension reduction and compression into the trained generation network, and output the three-dimensional seismic data after dimension-upgrading and decompression.

[0038] An embodiment of the present invention provides a seismic data processing system, comprising:

[0039] at least two generating networks, for generating simulated seismic data based on multi-channel fusion data;

[0040] A discriminant network is used to discriminate the matching degree between the simulated earthquake data generated by the generator network and the real earthquake data;

[0041] Seismic data dimensionality reduction network, used to perform dimensionality reduction and compression processing on the collected seismic data;

[0042] The seismic data dimensionality reduction network training device as described above;

[0043] A seismic data processing apparatus as described above.

[0044] An embodiment of the present invention provides a computer storage medium, in which computer executable instructions are stored. When the computer executable instructions are executed by a processor, a seismic data dimensionality reduction network training method as described above and / or a seismic data processing method as described above are implemented.

[0045] An embodiment of the present invention provides a data processing device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the above-mentioned seismic data dimensionality reduction network training method and / or the above-mentioned seismic data processing method are implemented.

[0046] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:

[0047] The seismic data dimension reduction network training method provided by the embodiment of the present invention selects pre-generated multi-channel fusion data, and inputs them into at least two generating networks with the same structure respectively, to obtain at least two groups of simulated seismic data; the simulated data and the collected real seismic data are input into the discriminant network, and the generating network and the discriminant network are trained through the optimal competition mechanism to obtain the trained discriminant network and the training network, and the discriminant network is processed to obtain the dimension reduction network, and the multi-channel fusion data and the simulated seismic data are used to train the dimension reduction network. By using the trained dimension reduction network to perform dimension reduction processing on the real seismic data, the characteristics of the seismic data can be learned and the real and effective dimension reduction processing can be performed, and the data characteristics will not be lost, so as to ensure the lossless dimension reduction compression of the seismic data, and the data after the dimension reduction processing will not be distorted.

[0048] The seismic data processing method provided by the embodiment of the present invention utilizes a trained dimensionality reduction network to perform undistorted dimensionality reduction processing on seismic data, and utilizes a trained generative network to perform dimensionality increase and decompression on real seismic dimensionality reduction data, thereby restoring real seismic data, completing the undistorted dimensionality reduction compression and fidelity dimensionality increase and decompression of seismic data, reducing the space occupied by storing a real seismic reagent library, and obtaining seismic data without feature loss.

[0049] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings.

[0050] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0052] Figure 1 Flow chart of the seismic data dimensionality reduction network training method in an embodiment of the present invention;

[0053] Figure 2 A schematic diagram of the network structure of the training generation network and the discrimination network in an embodiment of the present invention;

[0054] Figure 3 It is a schematic diagram of a dimensionality reduction network structure diagram in an embodiment of the present invention;

[0055] Figure 4a This is an example diagram of a set of real seismic data a in an embodiment of the present invention;

[0056] Figure 4b This is an example diagram of a set of real seismic data b in an embodiment of the present invention;

[0057] Figure 5 It is a one-dimensional vector curve diagram corresponding to two sets of real seismic data after dimensionality reduction and compression in an embodiment of the present invention;

[0058] Figure 6a This is one of the display effect diagrams of the gradual change of real earthquake dimensionality reduction data corresponding to dimensionality increase data in an embodiment of the present invention;

[0059] Figure 6b This is the second effect diagram showing the gradual change of real earthquake dimensionality reduction data to corresponding dimensionality increase data in an embodiment of the present invention;

[0060] Figure 6c This is the third effect diagram showing the gradual change of real earthquake dimensionality reduction data to corresponding dimensionality increase data in an embodiment of the present invention;

[0061] Figure 6d This is the fourth effect diagram showing the gradual change of real earthquake dimensionality reduction data to corresponding dimensionality increase data in an embodiment of the present invention;

[0062] Figure 6eThis is the fifth effect diagram showing the gradual change of real earthquake dimensionality reduction data to corresponding dimensionality increase data in an embodiment of the present invention;

[0063] Figure 6f This is the sixth effect diagram showing the gradual change of real earthquake dimensionality reduction data to corresponding dimensionality increase data in an embodiment of the present invention;

[0064] Figure 6g This is the seventh effect diagram showing the gradual change of real earthquake dimensionality reduction data corresponding to dimensionality increase data in an embodiment of the present invention;

[0065] Figure 6h This is the eighth display effect diagram of the gradual change of real earthquake dimensionality reduction data corresponding to dimensionality increase data in an embodiment of the present invention;

[0066] Figure 7 It is a structural schematic diagram of a seismic data dimension reduction network training device in an embodiment of the present invention;

[0067] Figure 8 It is a structural schematic diagram of a seismic data processing device in an embodiment of the present invention;

[0068] Fig. 9 Schematic diagram of the structure of the seismic data processing system in an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0070] In the prior art, the dimensionality reduction and compression method of seismic data will inevitably lose some features during the compression process, causing the compressed data to be distorted. In order to solve this technical problem, an embodiment of the present invention provides a seismic data dimensionality reduction method and device, which adopts a machine learning network method to achieve simulation and dimensionality reduction processing of seismic data through a generation network, a discrimination network and a dimensionality reduction network, and uses intelligent means to achieve dimensionality reduction and compression of seismic data, compress the seismic data to a smaller volume, and meet the requirements for retaining seismic data features, effectively retain the features of the seismic data, avoid the problem that the existing wavelet transform method will lose some features, and make the compressed seismic data have better fidelity.

[0071] In order to solve the problem of data feature loss and data distortion in the prior art of using wavelet transform to compress seismic data, the embodiment of the present invention provides a seismic data dimension reduction network training and seismic data processing method, device and system. The idea of ​​generative adversarial network is used to simulate the signal. The advantage is that the error of the simulated signal of different scale translation or scaling of the wavelet basis is reduced.

[0072] The embodiment of the present invention provides a seismic data dimension reduction network training method, the process of which is as follows: Figure 1 As shown, the following steps are included:

[0073] Step S101: Select pre-generated multi-channel fusion data, and input them into at least two generation networks with the same structure respectively, to obtain at least two sets of simulated seismic data; the multi-channel fusion data is obtained by fusing a multi-channel Gaussian signal and a multi-channel random noise signal, the multi-channel Gaussian signal is obtained by performing domain transformation and dimension upgrading on a one-dimensional signal that conforms to the Gaussian distribution; the multi-channel random noise signal is obtained by performing random noise dimension upgrading.

[0074] An optional implementation manner is to perform domain transformation and dimension upgrading on a one-dimensional signal that conforms to a Gaussian distribution to obtain a multi-channel Gaussian signal; perform dimension upgrading on a random noise signal to obtain a multi-channel random noise signal; and fuse the multi-channel Gaussian signal and the multi-channel random noise signal to obtain multi-channel fused data.

[0075] One-dimensional Gaussian signal and random noise are intelligently upgraded to obtain multi-channel Gaussian signal and multi-channel random noise. Figure 2 , a function that conforms to the Gaussian distribution is used to generate a one-dimensional Gaussian signal with a Gaussian distribution, and the one-dimensional Gaussian signal is transformed into a domain through a neural network to obtain a feature vector; for example, a one-dimensional Gaussian signal has multiple points, specifically 32 points, and the one-dimensional Gaussian signal of 32 points is transformed into another domain, and the output is a 128-point variable domain signal; when the domain is transformed, a neural network model structure is first constructed, and the neural network model structure has a total of three layers, the first layer is the input layer, the input layer has multiple model parameters, and here 32 model parameters are used; the second layer is the hidden layer, the hidden layer has multiple model parameters, and here specifically 1024 model parameters are used; the third layer is the output layer, and the output layer is a variable domain signal of multiple points, and here a 128-point variable domain signal is used; the 32-point one-dimensional Gaussian signal is input into the input layer of the neural network model structure, and a 128-point variable domain signal is obtained in the output layer of the neural network model structure.

[0076] A noise generator is used to generate random noise. The random noise can be one-dimensional Gaussian noise. The random noise can also be transformed into a domain. The transformed random noise and the 128-point transformed domain signal of the one-dimensional Gaussian signal are subjected to dimensionality upgrading to obtain multi-channel Gaussian signals and multi-channel random noise.

[0077] The characteristics of real seismic signals are complex, and there is noise in the seismic phase axis. After the signal is upgraded, random noise is added. In order to improve the authenticity of the generated signal, the multi-channel random noise is fused with the multi-channel Gaussian signal. Specifically, multi-channel random noise is added to the multi-channel Gaussian signal. This can improve the authenticity of the generator and generate multi-channel fused data.

[0078] A plurality of multi-channel fusion data may be generated in advance for subsequent use, or the channel fusion data may be generated instantly each time it is used subsequently.

[0079] Step S102: input at least two groups of simulated seismic data and the collected real seismic data into the discriminant network for comparison and discrimination, and obtain a discrimination result that characterizes the degree of matching between the simulated seismic data and the real seismic data; update the network parameters of the generating network and / or update the network parameters of the discriminant network based on the preferential competition mechanism according to the discrimination result, until a preset number of training times is reached, and a trained generating network and discriminant network are obtained; the discriminant network includes a data analysis and processing layer and a classification layer.

[0080] At least two groups of simulated seismic data are input into a discrimination network and compared with the collected real seismic data respectively; according to the matching degree between the simulated seismic data and the real seismic data, discrimination results of at least two simulated seismic data are obtained, and the discrimination results include matching or not matching with the real seismic data; if the discrimination result of one of the simulated seismic data is matching with the real seismic data, the network parameters of the generating network corresponding to the simulated seismic data are used to update the network parameters of other generating networks; if the discrimination results of the simulated seismic data are all matching or not matching with the real seismic data, the network parameters of at least two generating networks and the network parameters of the discriminating network are updated according to preset parameter updating rules.

[0081] The multi-channel fusion data is used as the input of the generative network, and based on the multi-channel fusion data, simulated seismic data is generated to train the generative network and the discriminative network; see Figure 2 , based on multi-channel fusion data, simulated seismic data is generated, and the generative network and the discriminative network are trained. Using the multi-channel fusion data as the input of the generative network, multiple generative networks with the same structure are established. In the subsequent description, two generative networks 1 and 2 with the same structure are generated. Two simulated seismic data can be generated through the two generative networks; the multi-channel fusion data generated above can be used to input into the two generative networks respectively, and two sets of simulated seismic data are output.

[0082] The two groups of simulated seismic data are respectively combined with the real seismic data as the input of the discriminant network, and the discriminant network discriminates the simulated seismic data. In the discriminant network, the simulated seismic data and the real seismic data are analyzed by the data analysis and processing layer, and the simulated seismic data are classified by the classification layer. If the simulated seismic data is close enough to the real seismic data, for example, the value of the selected difference discrimination parameter is small enough, then the simulated seismic data is considered to be true enough, and the discrimination result of the simulated seismic data is matched with the real seismic data, which can be judged to be true, that is, the output result of the classification layer of the discriminant network is 1, otherwise the output structure is 0; if the output result is 1, it is considered that the generation network that generates the simulated data is better, and the network parameters are copied to another network to continue training the generation network.

[0083] If the discriminant network outputs all 0, it is considered that the generated networks are not good enough. The network parameters of the two generated networks are adjusted according to the preset adjustment rules. The parameters of the discriminant network can also be adjusted. After the adjustment, the generated network and the discriminant network continue to be trained. If the discriminant network outputs all 1, it is considered that the discriminant network parameters are not good enough. The discriminant network parameters are adjusted according to the preset adjustment rules. The network parameters of the generated network can also be appropriately fine-tuned. After the adjustment, the generated network and the discriminant network continue to be trained.

[0084] The generative network and the discriminative network are trained repeatedly in this way to make the two networks compete with each other until a preset number of training times is reached and the winner is retained; for example, the training is performed 1,000 times, and the parameters of the training network and the discriminative network are retained to obtain a trained generative network and a trained discriminative network; the discriminative network is layered to obtain a data analysis and processing layer and a classification layer.

[0085] Step S103: Based on the data analysis and processing layer in the trained discriminant network, a dimensionality reduction processing layer is added to obtain a seismic data dimensionality reduction network; the seismic simulation data generated by the generation network is used as the input of the dimensionality reduction network, and the one-dimensional seismic signal after dimensionality reduction is output; the dimensionality reduction network is trained according to the one-dimensional seismic signal and the one-dimensional signal corresponding to the seismic simulation data until a preset convergence is met, so as to obtain a trained seismic data dimensionality reduction network.

[0086] A sample database is established based on the simulated seismic data generated by the trained generation network and the corresponding multi-channel fusion data; the multi-channel fusion data generated in steps S101-S102 is input into the trained generation network to obtain the simulated seismic data, and the simulated seismic data and the multi-channel fusion data corresponding to the simulated data constitute a sample database.

[0087] Keep the discriminant network parameters, change the discriminant network structure, and establish a dimensionality reduction network; see Figure 3As shown, in steps S101-S102, both the generating network and the discriminant network converge, the discriminant network parameters are retained, the discriminant network structure is changed, and a dimensionality reduction network is established; after the dimensionality reduction network converges, the seismic data is input, and after being processed by the dimensionality reduction network, a one-dimensional vector can be obtained, thus realizing the conversion of three-dimensional seismic data to a one-dimensional vector. To achieve dimensionality reduction, the last classification layer of the discriminant network is removed, and then a dimensionality reduction layer is added to reduce the dimension to a one-dimensional vector. Specifically, the trained discriminant network is layered to obtain a data analysis and processing layer and a classification layer. The classification layer is removed and the data analysis and processing layer is retained. The data analysis and processing layer and the dimensionality reduction layer are combined to obtain a seismic data dimensionality reduction network. As shown Figure 3 As shown, the Gaussian one-dimensional signal in the sample database is used as the output of the seismic data dimensionality reduction network, and the corresponding simulated seismic data in the sample database is used as the input of the seismic data dimensionality reduction network to train the dimensionality reduction network; the generated one-dimensional Gaussian signal and the generated network trained by parameter optimization are used to generate simulated seismic data to train the dimensionality reduction network, which can make the parameters of the trained dimensionality reduction network better and the dimensionality reduction processing effect better.

[0088] Specifically, seismic simulation data generated by a generating network is obtained from a sample database, input into a seismic data dimensionality reduction network, and a one-dimensional seismic signal after dimensionality reduction is output; a signal error between the one-dimensional seismic signal and the one-dimensional signal in the multi-channel fusion data is determined based on the one-dimensional seismic signal after dimensionality reduction and the one-dimensional signal in the multi-channel fusion data corresponding to the simulated seismic data; if the signal error does not meet a preset signal error requirement, the step of obtaining seismic simulation data generated by a generating network from a sample database is continued; until the signal error meets the preset signal error requirement, it is considered that the seismic data dimensionality reduction network has converged, and a trained seismic data dimensionality reduction network is obtained.

[0089] Based on the same inventive concept, an embodiment of the present invention further provides a seismic data processing method, comprising:

[0090] The three-dimensional seismic data to be processed is input into the trained seismic data dimensionality reduction network, and the one-dimensional seismic data after dimensionality reduction and compression is output; the seismic data dimensionality reduction network is trained using the seismic data dimensionality reduction network training method as described above.

[0091] Optionally, the method further includes: inputting the one-dimensional seismic data after dimensionality reduction and compression into a trained generation network, and outputting the three-dimensional seismic data after dimensionality increase and decompression.

[0092] The dimensionality reduction network is used to convert three-dimensional seismic data into one-dimensional vectors, thus completing the dimensionality reduction of real seismic data. The generative network can be used to convert one-dimensional vectors into three-dimensional seismic data, completing the dimensionality increase of seismic data after dimensionality reduction, thus realizing the data compression and decompression process.

[0093] The one-dimensional vector at this time can be converted into three-dimensional seismic data through the generative network trained by S101-S103; thus, the up-down dimensional conversion is completed, thereby effectively realizing the data compression and decompression process; in the embodiment of the present invention, the seismic image is reduced to a one-dimensional vector and visualized. Figure 4a and Figure 4b , Figure 4a and Figure 4b Represent two sets of real seismic data a and b respectively; using the seismic data dimensionality reduction network trained in S101-S103, Figure 4a and Figure 4b The real earthquake data a and b in the dimensionality reduction compression are carried out. For details, see Figure 5 , we get the one-dimensional seismic data after dimensionality reduction and compression corresponding to the real seismic data a and b. Figure 5 The hidden vector 1 (indicated by the solid line) is the one-dimensional seismic data obtained by dimensionality reduction compression of the real seismic data a. Figure 5 The hidden vector 2 (indicated by the dotted line) is the one-dimensional seismic data obtained by dimensionality reduction compression of the real seismic data b. The compression of the hidden vectors 1 and 2 is 3.13% of the real seismic data a and b.

[0094] Figure 6a , 6b , 6c, 6d, 6e, 6f, 6g, and 6h show the process of seismic data a gradually changing into seismic data b. The gradual change of seismic data a into seismic data b is the gradual change of latent vector 1 to latent vector 2. The reverse change of the gradient vector to seismic data is also shown in Figure 6a-6h It is shown in the figure; the latent vector 1 corresponding to the seismic data a is changed into vector data, the changed vector data is decompressed by the trained generative network, and the raw dimension decompressed data is displayed as a recognizable image. It can be seen from the figure that as the latent vector 1 gradually changes to the latent vector 2, the corresponding recognizable image also achieves the effect of gradually changing the seismic data a to the seismic data b. For details, see Figure 6a-6h It shows the subtle change effect of seismic data a gradually changing to seismic data b. Figure 6a-6h The effect of using the seismic data dimensionality reduction network and generation network of the present invention to reduce the dimension of seismic data and to decompress the dimension of seismic data can be demonstrated, which shows that the present invention has excellent feature extraction capability.

[0095] The above method can be applied to the field of seismic exploration. For seismic data, a deep learning method is used to extract waveform features to achieve the purpose of data dimensionality reduction. The method uses a trained dimensionality reduction network to perform dimensionality reduction processing on real seismic data, which can learn the characteristics of seismic data and perform real and effective dimensionality reduction processing without losing data characteristics, ensuring lossless dimensionality reduction compression of seismic data, and the data after dimensionality reduction processing will not be distorted. The trained dimensionality reduction network is used to perform undistorted dimensionality reduction processing on seismic data, and the trained generative network is used to perform dimensionality increase and decompression on real seismic dimensionality reduction data, which can restore the real seismic data, complete the distortion-free dimensionality reduction compression and fidelity dimensionality increase and decompression of seismic data, reduce the space occupied by storing real seismic reagent libraries, and obtain seismic data without feature loss.

[0096] Based on the same inventive concept, the embodiment of the present invention also provides a seismic data dimension reduction network training device, the structure of which is as follows: Figure 7 As shown, it includes: a data simulation module 11, a discrimination module 12, a first network training module 13, and a second network training module 14;

[0097] The data simulation module 11 is used to select pre-generated multi-channel fusion data and input them into at least two generating networks with the same structure respectively to obtain at least two sets of simulated seismic data; the multi-channel fusion data is obtained by fusing a multi-channel Gaussian signal and a multi-channel random noise signal, the multi-channel Gaussian signal is obtained by performing domain transformation and dimension upgrading on a one-dimensional signal that conforms to the Gaussian distribution; the multi-channel random noise signal is obtained by performing random noise dimension upgrading;

[0098] A discrimination module 12, for inputting at least two groups of simulated seismic data and the collected real seismic data into a discrimination network for comparison and discrimination, and obtaining a discrimination result representing the degree of matching between the simulated seismic data and the real seismic data;

[0099] The first network training module 13 is used to update the network parameters of the generating network and / or the network parameters of the discriminating network based on the optimal competition mechanism according to the discrimination result, until a preset number of training times is reached to obtain a trained generating network and discriminating network; the discriminating network includes a data analysis and processing layer and a classification layer;

[0100] The second network training module 14 is used to add a dimensionality reduction processing layer on the basis of the data analysis processing layer in the trained discriminant network to obtain a seismic data dimensionality reduction network; use the seismic simulation data generated by the generation network as the input of the dimensionality reduction network, output the reduced one-dimensional seismic signal, and determine the training of the dimensionality reduction network to meet the preset convergence according to the one-dimensional seismic signal and the one-dimensional signal corresponding to the seismic simulation data, so as to obtain a trained seismic data dimensionality reduction network.

[0101] Based on the same inventive concept, the embodiment of the present invention further provides a seismic data processing device, the structure of which is as follows: Figure 8 As shown, it includes: a dimension reduction processing module 21 and / or a dimension increase processing module 22;

[0102] A dimension reduction processing module 21 is used to input the three-dimensional seismic data to be processed into a trained seismic data dimension reduction network, and output one-dimensional seismic data after dimension reduction and compression; the seismic data dimension reduction network is trained using the seismic data dimension reduction network training method as described above;

[0103] The dimension-increasing processing module 22 is used to input the one-dimensional seismic data after dimension reduction and compression into the trained generation network, and output the three-dimensional seismic data after dimension-increasing and decompression.

[0104] Based on the same inventive concept, the embodiment of the present invention also provides a seismic data processing system, including: at least two generating networks 31, a discriminating network 32, a seismic data dimension reduction network 33, a seismic data dimension reduction network training device 34, and a seismic data processing device 35;

[0105] At least two generating networks 31, used to generate simulated seismic data based on multi-channel fusion data;

[0106] A discriminant network 32, used to discriminate the matching degree between the simulated seismic data generated by the generating network and the real seismic data;

[0107] A seismic data dimensionality reduction network 33 is used to perform dimensionality reduction and compression processing on the collected seismic data;

[0108] The seismic data dimension reduction network training device 34 as described above;

[0109] A seismic data processing apparatus 35 as described above.

[0110] Based on the same inventive concept, an embodiment of the present invention also provides a computer storage medium, in which computer executable instructions are stored. When the computer executable instructions are executed by a processor, a seismic data dimensionality reduction network training method as described above and / or a seismic data processing method as described above are implemented.

[0111] Based on the same inventive concept, an embodiment of the present invention also provides a data processing device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements a seismic data dimensionality reduction network training method as described above and / or a seismic data processing method as described above.

[0112] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0113] Unless otherwise specifically stated, terms such as processing, computing, calculating, determining, displaying, etc. may refer to the actions and / or processes of one or more processing or computing systems, or similar devices, which operate and convert data represented as physical (e.g., electronic) quantities within registers or memories of a processing system into other data similarly represented as physical quantities within memories, registers, or other such information storage, transmission, or display devices of the processing system. Information and signals may be represented using any of a variety of different techniques and methods. For example, data, instructions, commands, information, signals, bits, symbols, and chips mentioned throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, light fields or particles, or any combination thereof.

[0114] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process can be rearranged without departing from the scope of protection of the present disclosure. The attached method claims present the elements of the various steps in an exemplary order and are not intended to be limited to the specific order or hierarchy described.

[0115] In the above detailed description, various features are grouped together in a single embodiment to simplify the disclosure. This method of disclosure should not be interpreted as reflecting an intention that the embodiments of the claimed subject matter require more features than are clearly stated in each claim. On the contrary, as reflected in the appended claims, the invention is in a state of having less than all the features of the disclosed individual embodiments. Therefore, the appended claims are hereby expressly incorporated into the detailed description, with each claim standing on its own as a separate preferred embodiment of the invention.

[0116] Those skilled in the art will also appreciate that the various illustrative logic blocks, modules, circuits, and algorithmic steps described in conjunction with the embodiments herein can all be implemented as electronic hardware, computer software, or a combination thereof. In order to clearly illustrate the interchangeability between hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above around their functions. Whether such functions are implemented as hardware or software depends on specific applications and the design constraints imposed on the entire system. A skilled person can implement the described functions in an alternative manner for each specific application, but such implementation decisions should not be interpreted as departing from the scope of protection of the present disclosure.

[0117] The steps of the method or algorithm described in conjunction with the embodiments herein may be directly embodied as hardware, a software module executed by a processor, or a combination thereof. The software module may be located in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a mobile disk, a CD-ROM, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor so that the processor can read information from the storage medium and can write information to the storage medium. Of course, the storage medium may also be an integral part of the processor. The processor and the storage medium may be located in an ASIC. The ASIC may be located in a user terminal. Of course, the processor and the storage medium may also be present in a user terminal as discrete components.

[0118] For software implementation, the techniques described in this application can be implemented with modules (e.g., procedures, functions, etc.) that perform the functions described in this application. These software codes can be stored in a memory unit and executed by a processor. The memory unit can be implemented within the processor or outside the processor. In the latter case, it is coupled to the processor in a communication manner via various means, which are well known in the art.

[0119] The above description includes examples of one or more embodiments. Of course, it is impossible to describe all possible combinations of components or methods for the purpose of describing the above embodiments, but it should be recognized by those skilled in the art that the various embodiments may be further combined and arranged. Therefore, the embodiments described herein are intended to cover all such changes, modifications and variations that fall within the scope of protection of the appended claims. In addition, with respect to the term "comprising" used in the specification or claims, the word is covered in a manner similar to the term "including", just as "including," is explained as a transitional word in the claims. In addition, any term "or" used in the specification of the claims is intended to mean "non-exclusive or".

Claims

1. A seismic data dimensionality reduction network training method, characterized in that: include: Selecting pre-generated multi-channel fusion data, and inputting them into at least two generation networks with the same structure, respectively, to obtain at least two sets of simulated seismic data; The multi-channel fusion data is obtained by fusing a multi-channel Gaussian signal and a multi-channel random noise signal. The multi-channel Gaussian signal is obtained by performing domain transformation and dimension upgrading on a one-dimensional signal that conforms to a Gaussian distribution; the multi-channel random noise signal is obtained by performing dimension upgrading on random noise. Inputting at least two groups of simulated seismic data and the collected real seismic data into the discrimination network for comparison and discrimination, and obtaining a discrimination result representing the degree of matching between the simulated seismic data and the real seismic data; According to the discrimination result, the network parameters of the generating network and / or the network parameters of the discriminating network are updated based on the optimal competition mechanism until a preset number of training times is reached to obtain a trained generating network and a discriminating network; the discriminating network includes a data analysis and processing layer and a classification layer; On the basis of the data analysis and processing layer in the trained discriminant network, a dimensionality reduction processing layer is added to obtain a seismic data dimensionality reduction network; the seismic simulation data generated by the generation network is used as the input of the dimensionality reduction network, and the one-dimensional seismic signal after dimensionality reduction is output. According to the one-dimensional seismic signal and the one-dimensional signal corresponding to the seismic simulation data, the dimensionality reduction network is trained to meet the preset convergence, so as to obtain a trained seismic data dimensionality reduction network.

2. The method according to claim 1, characterized in that Also includes: The one-dimensional signal that conforms to the Gaussian distribution is processed by changing the domain and increasing the dimension to obtain a multi-channel Gaussian signal; The random noise signal is processed by dimension upgrading to obtain a multi-channel random noise signal; Multi-channel Gaussian signals and multi-channel random noise signals are fused to obtain multi-channel fused data.

3. The method according to claim 1, characterized in that The at least two groups of simulated seismic data and the collected real seismic data are respectively input into the discrimination network for comparison and discrimination, so as to obtain a discrimination result representing the degree of matching between the simulated seismic data and the real seismic data, and the network parameters of the generation network and / or the network parameters of the discrimination network are updated based on the preferential competition mechanism according to the discrimination result, including: Input at least two groups of simulated seismic data into the discriminant network and compare them with the acquired real seismic data respectively; Obtaining at least two discrimination results of the simulated seismic data according to the matching degree between the simulated seismic data and the real seismic data, wherein the discrimination results include matching or mismatching with the real seismic data; If the judgment result of one of the simulated seismic data matches the real seismic data, the network parameters of the generated network corresponding to the simulated seismic data are used to update the network parameters of the other generated networks; If the discrimination results of the simulated seismic data are all matched or mismatched with the real seismic data, the network parameters of at least two generating networks and the network parameters of the discriminating network are updated according to a preset parameter updating rule.

4. The method according to claim 3, characterized in that Also includes: A sample database is established based on the simulated seismic data generated by the trained generative network and the corresponding multi-channel fusion data.

5. The method according to claim 4, characterized in that The seismic simulation data generated by the generation network is used as the input of the dimension reduction network, and the one-dimensional seismic signal after dimension reduction is outputted. According to the one-dimensional seismic signal and the one-dimensional signal corresponding to the seismic simulation data, the dimension reduction network is trained to meet the preset convergence, and a trained seismic data dimension reduction network is obtained, including: Obtain earthquake simulation data generated by the generation network from the sample database, input it into the earthquake data dimension reduction network, and output the one-dimensional earthquake signal after dimension reduction; According to the one-dimensional seismic signal after dimensionality reduction and the one-dimensional signal in the multi-channel fusion data corresponding to the simulated seismic data, the signal error of the one-dimensional seismic signal and the one-dimensional signal in the multi-channel fusion data is determined. If the signal error does not meet the preset signal error requirement, the step of obtaining the seismic simulation data generated by the generation network from the sample database is continued; until the signal error meets the preset signal error requirement, it is considered that the seismic data dimensionality reduction network converges, and a trained seismic data dimensionality reduction network is obtained.

6. A method for processing seismic data, characterized in that: include: Input the three-dimensional seismic data to be processed into the trained seismic data dimensionality reduction network, and output the one-dimensional seismic data after dimensionality reduction and compression; The seismic data dimensionality reduction network is trained using the seismic data dimensionality reduction network training method described in any one of claims 1-5.

7. The method according to claim 6, characterized in that Also includes: The one-dimensional seismic data after dimensionality reduction and compression is input into the trained generative network, and the three-dimensional seismic data after dimensionality increase and decompression is output.

8. A seismic data dimensionality reduction network training device, characterized in that: include: The data simulation module is used to select pre-generated multi-channel fusion data and input them into at least two generating networks with the same structure respectively to obtain at least two sets of simulated seismic data; the multi-channel fusion data is obtained by fusing a multi-channel Gaussian signal and a multi-channel random noise signal, the multi-channel Gaussian signal is obtained by performing domain transformation and dimension upgrading on a one-dimensional signal that conforms to the Gaussian distribution; the multi-channel random noise signal is obtained by performing random noise dimension upgrading; A discrimination module, used to input at least two groups of simulated seismic data and the collected real seismic data into a discrimination network for comparison and discrimination, and obtain a discrimination result representing the degree of matching between the simulated seismic data and the real seismic data; A first network training module is used to update the network parameters of the generating network and / or the network parameters of the discriminating network based on the optimal competition mechanism according to the discrimination result, until a preset number of training times is reached to obtain a trained generating network and discriminating network; the discriminating network includes a data analysis and processing layer and a classification layer; The second network training module is used to add a dimensionality reduction processing layer on the basis of the data analysis processing layer in the trained discriminant network to obtain a seismic data dimensionality reduction network; use the seismic simulation data generated by the generation network as the input of the dimensionality reduction network, output the reduced one-dimensional seismic signal, and determine the training of the dimensionality reduction network to meet the preset convergence according to the one-dimensional seismic signal and the one-dimensional signal corresponding to the seismic simulation data, so as to obtain a trained seismic data dimensionality reduction network.

9. A seismic data processing device, characterized in that: include: Dimensionality reduction processing module and / or dimensionality increase processing module; The dimension reduction processing module is used to input the three-dimensional seismic data to be processed into the trained seismic data dimension reduction network, and output the one-dimensional seismic data after dimension reduction and compression; The seismic data dimensionality reduction network is trained using the seismic data dimensionality reduction network training method described in any one of claims 1 to 5; The dimension-upgrading processing module is used to input the one-dimensional seismic data after dimension reduction and compression into the trained generation network, and output the three-dimensional seismic data after dimension-upgrading and decompression.

10. A seismic data processing system, characterized in that: include: at least two generating networks, for generating simulated seismic data based on multi-channel fusion data; A discriminant network is used to discriminate the matching degree between the simulated earthquake data generated by the generator network and the real earthquake data; Seismic data dimensionality reduction network, used to perform dimensionality reduction and compression processing on the collected seismic data; The seismic data dimensionality reduction network training device according to claim 8; The seismic data processing apparatus as claimed in claim 9.

11. A computer storage medium, characterized in that: The computer storage medium stores computer executable instructions, which, when executed by a processor, implement a seismic data dimensionality reduction network training method as described in any one of claims 1-5 and / or a seismic data processing method as described in any one of claims 6-7.

12. A data processing device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a seismic data dimensionality reduction network training method described in any one of claims 1-5 and / or a seismic data processing method described in any one of claims 6-7 are implemented.

Images