Deep learning-based intelligent quantitative evaluation method and system for amplitude preservation of pre-set gathers

Abstract

The invention discloses a deep learning-based intelligent quantitative evaluation method and system for amplitude preservation of a front trace gather. Constructing an unsupervised deep learning model; inputting a to-be-detected work area seismic data slice seismic image into the unsupervised deep learning model for training to obtain a pre-offset gather amplitude preservation intelligent quantitative evaluation model; and inputting two-dimensional pre-offset gather data to be evaluated into the pre-offset gather amplitude preservation intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data. According to the method, quantitative evaluation can be carried out on the amplitude preserving performance of geological data under the unsupervised condition, a qualified label-free training preamble gather amplitude preserving performance intelligent quantitative evaluation model is established, and the established preamble gather amplitude preserving performance intelligent quantitative evaluation model can be used for evaluating the amplitude preserving performance of the geological data under the condition of not depending on any prior information. Amplitude preservation quantization parameters are intelligently predicted in a data driving mode, and then the precision of lithologic oil and gas reservoir prediction is improved.

Description

Technical Field

[0001] The present invention relates to the field of seismic exploration technology, and in particular to an intelligent quantitative evaluation method and system for amplitude preservation of biased forward gathers based on deep learning. Background Art

[0002] With the continuous development of oil and gas exploration technology and the demand for deeper oil and gas resources, the exploration of concealed oil and gas reservoirs has gradually become an important direction of research in the field of oil and gas exploration. Lithologic oil and gas reservoirs are an important category of concealed oil and gas reservoirs, but the exploration and prediction of lithologic oil and gas reservoirs faces many challenges.

[0003] Amplitude-preserving processing refers to the process of improving the quality of seismic data by using reasonable and effective processing methods while maintaining the effective reflection seismic information of the original seismic data without relative distortion. The result of amplitude-preserving seismic data processing is reflected in the relative preservation of the waveform characteristics of the reflected wavelet and the reflection coefficient characteristics.

[0004] There are many challenges in the research work on lithologic oil and gas reservoirs, from qualitative to quantitative, from reservoir lithology to physical properties and oil and gas property discrimination. The lithologic trap identification method requires high-quality seismic data, and at the same time requires that the processed seismic data have high amplitude preservation. If the relative amplitude preservation of the processed seismic data is very low, the application of fine reservoir prediction technology usually does not achieve good results. Therefore, how to make the seismic data still maintain a high relative amplitude preservation after processing, so as to improve the accuracy of lithologic oil and gas reservoir prediction, is an urgent problem that needs to be solved.

[0005] The existing methods for distinguishing and evaluating amplitude preservation processing of seismic data have evaluated the amplitude preservation of various seismic data processing technologies from different perspectives, taking the seismic wave amplitude characteristics, frequency characteristics, seismic attribute slices, and the consistency of wave impedance and logging results as the evaluation basis. However, these methods have a common defect that they all rely on the assumptions or prior knowledge of the model, which makes the application of amplitude preservation evaluation limited. When these assumptions and priors are violated, the performance of amplitude preservation evaluation will be greatly reduced. Summary of the invention

[0006] In order to solve the above problems, the inventors have made the present invention, and through specific implementation methods, provide a method and system for intelligent quantitative evaluation of amplitude preservation of biased forward gathers based on deep learning.

[0007] In a first aspect, an embodiment of the present invention provides an intelligent quantitative evaluation method for amplitude preservation of bias-forward gathers based on deep learning, comprising the following steps:

[0008] Build unsupervised deep learning models;

[0009] The seismic data slices of the work area to be tested are input into the unsupervised deep learning model for training, and an intelligent quantitative evaluation model for amplitude preservation of the biased forward gathers is obtained;

[0010] The two-dimensional pre-bias gather data to be evaluated is input into the pre-bias gather amplitude-preserving intelligent quantitative evaluation model to obtain the corresponding quantitative evaluation data.

[0011] Specifically, building an unsupervised deep learning model includes the following steps:

[0012] Constructing a biased forward gather feature extraction network, including sequentially setting 10 convolutional layers with channel numbers of 32, 32, 64, 64, 128, 128, 256, 256, 512, and 512, respectively, the first convolutional layer is the input end of the biased forward gather feature extraction network, the first pooling layer is set between the second and third convolutional layers, the second pooling layer is set between the fourth and fifth convolutional layers, the third pooling layer is set between the sixth and seventh convolutional layers, the fourth pooling layer is set between the eighth and ninth convolutional layers, the fifth pooling layer is set after the tenth convolutional layer, and the fifth pooling layer is the output end of the biased forward gather feature extraction network;

[0013] A seismic image quality quantitative assessment network is constructed. After the output end of the biased forward gather feature extraction network, a first fully connected layer with a channel number of 512 and a second fully connected layer with a channel number of 1 are sequentially set. The second fully connected layer is the output end of the seismic image quality quantitative assessment network.

[0014] Specifically, the seismic data slices of the work area to be tested are input into the unsupervised deep learning model for training to obtain an intelligent quantitative evaluation model for amplitude preservation of the biased forward gathers, including the following steps:

[0015] Input the seismic image of the seismic data slices in the work area to be tested into the biased forward gather feature extraction network to obtain the target seismic image features;

[0016] The seismic image quality quantitative assessment network is used to learn the characteristics of the target seismic image, and an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers is obtained.

[0017] Specifically, the seismic data slices of the work area to be tested are input into the seismic image to construct a biased forward gather feature extraction network to obtain the target seismic image features, including the following steps:

[0018] Slice the seismic image of the seismic data of the work area to be measured into multiple regions, input the regions into the input end of the biased forward gather feature extraction network, perform feature extraction through the first and second convolutional layers, and obtain the first seismic image feature;

[0019] Inputting the first seismic image feature into the activation function for processing to obtain a second seismic image feature;

[0020] Inputting the second seismic image feature into the first pooling layer for feature selection and information filtering to obtain a third seismic image feature;

[0021] After the third and fourth convolutional layers, feature extraction is performed on the third seismic image feature to obtain a fourth seismic image feature;

[0022] Inputting the fourth seismic image feature into the second pooling layer for feature selection and information filtering to obtain a fifth seismic image feature;

[0023] After the fifth and sixth convolutional layers, the fifth seismic image feature is extracted to obtain the sixth seismic image feature;

[0024] Inputting the sixth seismic image feature into the third pooling layer for feature selection and information filtering to obtain the seventh seismic image feature;

[0025] After the seventh and eighth convolutional layers, feature extraction is performed on the seventh seismic image feature to obtain the eighth seismic image feature;

[0026] Inputting the eighth seismic image feature into the fourth pooling layer for feature selection and information filtering to obtain a ninth seismic image feature;

[0027] After the ninth and tenth convolutional layers, the ninth seismic image feature is extracted to obtain the tenth seismic image feature;

[0028] The tenth seismic image feature is input into the fifth pooling layer for feature selection and information filtering to obtain the target seismic image feature.

[0029] Specifically, the seismic image quality quantitative assessment network is used to learn the target seismic image features to obtain an intelligent quantitative assessment model for amplitude preservation of biased forward gathers, which also includes the following steps:

[0030] After any of the third to tenth convolutional layers, before inputting the next convolutional layer or pooling layer, the seismic image features output by the corresponding convolutional layer are input into the activation function for processing. The activation function is

[0031]

[0032] Among them, g and w i 、a i They represent output, weight and input respectively, and i represents the sequence number.

[0033] Specifically, the area size of the seismic data slice seismic image input biased gather feature extraction network in the work area to be tested is 32×32, and end-to-end training is performed. The feature extraction function is:

[0034] y R =F(x, {W i})+x1

[0035] The function of each convolutional layer is F(x), and the feature vector from the first convolutional layer is x 1 , Wi is the weight of the i-th convolution layer, y R Represents the feature vector output after unsupervised learning;

[0036] The convolution kernel size used in all convolutional layers is 3×3, and the convolution kernel adopts a zero-padding strategy to make the input and output sizes of each convolutional layer consistent, and all pooling layers are set to 2×2.

[0037] Specifically, a seismic image quality quantitative assessment network is used to learn the target seismic image features to obtain an intelligent quantitative assessment model for amplitude preservation of biased forward gathers, including the following steps:

[0038] The probability of temporarily discarding the neural network unit of the seismic image quality quantitative assessment network from the network is set to 0.5. Each area of ​​the target seismic image feature is regressed and scored through the first fully connected layer. The weights are learned through the second fully connected layer, and the learned weights are assigned to the scoring results of each corresponding area of ​​the target seismic image feature. An intelligent quantitative evaluation model for amplitude preservation of biased forward gathers is obtained.

[0039] In a second aspect, an embodiment of the present invention provides an intelligent quantitative evaluation system for amplitude preservation of biased forward gathers based on deep learning, comprising:

[0040] The quantitative evaluation model determination module is used to construct an unsupervised deep learning model; the seismic data slices of the work area to be tested are input into the unsupervised deep learning model for training to obtain an intelligent quantitative evaluation model for amplitude preservation of the biased forward gather;

[0041] The quantitative evaluation module is used to input the two-dimensional pre-biased gather data to be evaluated into the pre-biased gather amplitude-preserving intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data.

[0042] In a third aspect, an embodiment of the present invention provides a method for constructing an intelligent quantitative evaluation model for amplitude preservation of bias-forward gathers, comprising the following steps:

[0043] Constructing a biased forward gather feature extraction network, including sequentially setting 10 convolutional layers with channel numbers of 32, 32, 64, 64, 128, 128, 256, 256, 512, and 512, respectively, the first convolutional layer is the input end of the biased forward gather feature extraction network, the first pooling layer is set between the second and third convolutional layers, the second pooling layer is set between the fourth and fifth convolutional layers, the third pooling layer is set between the sixth and seventh convolutional layers, the fourth pooling layer is set between the eighth and ninth convolutional layers, the fifth pooling layer is set after the tenth convolutional layer, and the fifth pooling layer is the output end of the biased forward gather feature extraction network;

[0044] A seismic image quality quantitative assessment network is constructed, and a first fully connected layer with a channel number of 512 and a second fully connected layer with a channel number of 1 are sequentially set after the output end of the biased forward gather feature extraction network, and the second fully connected layer is the output end of the seismic image quality quantitative assessment network;

[0045] Input the seismic image of the seismic data slices in the work area to be tested into the biased forward gather feature extraction network to obtain the target seismic image features;

[0046] The seismic image quality quantitative assessment network is used to learn the characteristics of the target seismic image, and an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers is obtained.

[0047] In a fourth aspect, an embodiment of the present invention provides a system for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers, comprising:

[0048] A network construction module is used to construct a biased forward gather feature extraction network, including sequentially setting 10 convolutional layers with channel numbers of 32, 32, 64, 64, 128, 128, 256, 256, 512, and 512, respectively, the first convolutional layer is the input end of the biased forward gather feature extraction network, the first pooling layer is set between the second and third convolutional layers, the second pooling layer is set between the fourth and fifth convolutional layers, the third pooling layer is set between the sixth and seventh convolutional layers, the fourth pooling layer is set between the eighth and ninth convolutional layers, and the fifth pooling layer is set after the tenth convolutional layer, and the fifth pooling layer is the output end of the biased forward gather feature extraction network; constructing a seismic image quality quantitative assessment network, sequentially setting a first fully connected layer with a channel number of 512 and a second fully connected layer with a channel number of 1 after the output end of the biased forward gather feature extraction network, and the second fully connected layer is the output end of the seismic image quality quantitative assessment network;

[0049] The feature extraction module is used to construct a biased forward gather feature extraction network by inputting the seismic data slice seismic image of the work area to be tested, and obtain the target seismic image features;

[0050] The quantitative evaluation model determination module is used to learn the target seismic image characteristics through the seismic image quality quantitative evaluation network to obtain the amplitude-preserving intelligent quantitative evaluation model of the biased forward gather.

[0051] In a fifth aspect, an embodiment of the present invention provides an intelligent quantitative evaluation method for amplitude preservation of biased forward gathers based on deep learning, comprising the following steps:

[0052] The two-dimensional pre-biased gather data to be evaluated is input into a pre-biased gather amplitude preservation intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data; the pre-biased gather amplitude preservation intelligent quantitative evaluation model is used to extract and learn target seismic image features.

[0053] In a sixth aspect, an embodiment of the present invention provides an intelligent quantitative evaluation system for amplitude preservation of biased forward gathers based on deep learning, comprising:

[0054] The quantitative evaluation module is used to input the two-dimensional pre-biased gather data to be evaluated into the pre-biased gather amplitude preservation intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data; the pre-biased gather amplitude preservation intelligent quantitative evaluation model is used to extract and learn the target seismic image features.

[0055] Based on the same inventive concept, an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the aforementioned deep learning-based intelligent quantitative evaluation method for amplitude preservation of biased forward trace gathers or the method for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward trace gathers.

[0056] Based on the same inventive concept, 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, the aforementioned deep learning-based intelligent quantitative evaluation method for amplitude preservation of biased forward trace gathers or the method for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward trace gathers is implemented.

[0057] The beneficial effects of the above technical solution provided by the embodiment of the present invention include at least:

[0058] It is able to realize quantitative evaluation of the amplitude preservation of geological data under unsupervised conditions, and a qualified unlabeled trained biased forward gather amplitude preservation intelligent quantitative evaluation model is established. Using the established biased forward gather amplitude preservation intelligent quantitative evaluation model, it is possible to intelligently predict the amplitude preservation quantitative parameters in a data-driven manner without relying on any prior information, thereby improving the accuracy of lithologic oil and gas reservoir prediction.

[0059] 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.

[0060] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] 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:

[0062] Figure 1It is a flow chart of an intelligent quantitative evaluation method for amplitude preservation of biased forward gathers based on deep learning in an embodiment of the present invention;

[0063] Figure 2 A seismic image input to the network in an embodiment of the present invention;

[0064] Figure 3 A seismic image output by the network in an embodiment of the present invention;

[0065] Figure 4 This is a specific flow chart of unsupervised quantitative evaluation prediction in an embodiment of the present invention;

[0066] Figure 5 This is a schematic diagram of the structure of a deep learning network for quantitative evaluation in an embodiment of the present invention;

[0067] Figure 6 FIG. 1 is a VSP data slice diagram in an embodiment of the present invention;

[0068] Figure 7 This is a result diagram obtained by quantitative evaluation of VSP data slice 1 in an embodiment of the present invention;

[0069] Figure 8 FIG. 2 is a VSP data slice diagram in an embodiment of the present invention;

[0070] Fig. 9 This is a result diagram obtained by quantitative evaluation of VSP data slice 2 in an embodiment of the present invention;

[0071] Fig.10 This is a slice diagram of F3 data in an embodiment of the present invention;

[0072] Fig.11 This is a result diagram obtained by performing quantitative evaluation on F3 data slices in an embodiment of the present invention;

[0073] Fig.12 A three-dimensional VSP data diagram in an embodiment of the present invention;

[0074] Fig.13 This is a result diagram obtained by quantitatively evaluating three-dimensional VSP data in an embodiment of the present invention;

[0075] Fig.14 It is a three-dimensional F3 data diagram in an embodiment of the present invention;

[0076] Fig.15 This is a result diagram obtained by quantitatively evaluating the three-dimensional F3 data in an embodiment of the present invention;

[0077] Fig.16 The figure is a schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0078] 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.

[0079] In order to solve the problems existing in the prior art, an embodiment of the present invention provides a method and system for intelligent quantitative evaluation of amplitude preservation of biased forward gathers based on deep learning.

[0080] The embodiment of the present invention provides an intelligent quantitative evaluation method for amplitude preservation of biased forward gathers based on deep learning, and the process is as follows: Figure 1 As shown, the following steps are included:

[0081] Step S1: Build an unsupervised deep learning model.

[0082] In some specific embodiments, constructing an unsupervised deep learning model includes the following steps:

[0083] A biased-forward gather feature extraction network is constructed, including sequentially setting 10 convolutional layers with channel numbers of 32, 32, 64, 64, 128, 128, 256, 256, 512, and 512, respectively, the first convolutional layer is the input end of the biased-forward gather feature extraction network, the first pooling layer is set between the second and third convolutional layers, the second pooling layer is set between the fourth and fifth convolutional layers, the third pooling layer is set between the sixth and seventh convolutional layers, the fourth pooling layer is set between the eighth and ninth convolutional layers, and the fifth pooling layer is set after the tenth convolutional layer, and the fifth pooling layer is the output end of the biased-forward gather feature extraction network; a seismic image quality quantitative assessment network is constructed, and a first fully connected layer with a channel number of 512 and a second fully connected layer with a channel number of 1 are sequentially set after the output end of the biased-forward gather feature extraction network, and the second fully connected layer is the output end of the seismic image quality quantitative assessment network.

[0084] Step S2: Input the seismic data slices of the work area to be tested into the unsupervised deep learning model for training to obtain an intelligent quantitative evaluation model for amplitude preservation of the biased forward gather.

[0085] In some specific embodiments, the seismic images of seismic data slices in the work area to be tested are input into an unsupervised deep learning model for training to obtain an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers, comprising the following steps: the seismic images of seismic data slices in the work area to be tested are input into a biased forward gather feature extraction network to construct target seismic image features; the target seismic image features are learned using a seismic image quality quantitative assessment network to obtain an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers.

[0086] In some specific embodiments, the seismic image of the seismic data slice of the work area to be measured is input into a biased forward gather feature extraction network to obtain the target seismic image features, including the following steps: the seismic image of the seismic data slice of the work area to be measured is divided into multiple regions, and the regions are input into the input end of the biased forward gather feature extraction network, and the features are extracted through the first and second convolutional layers to obtain the first seismic image features; the first seismic image features are input into the activation function for processing to obtain the second seismic image features; the second seismic image features are input into the first pooling layer for feature selection and information filtering to obtain the third seismic image features; the third seismic image features are extracted through the third and fourth convolutional layers to obtain the fourth seismic image features; the fourth seismic image features are input into the second pooling layer for feature extraction The fifth seismic image feature is obtained by performing feature selection and information filtering on the fifth and sixth convolutional layers; the sixth seismic image feature is extracted from the fifth seismic image feature to obtain the sixth seismic image feature; the sixth seismic image feature is input into the third pooling layer for feature selection and information filtering to obtain the seventh seismic image feature; the seventh and eighth convolutional layers are used to extract the seventh seismic image feature to obtain the eighth seismic image feature; the eighth seismic image feature is input into the fourth pooling layer for feature selection and information filtering to obtain the ninth seismic image feature; the ninth and tenth convolutional layers are used to extract the ninth seismic image feature to obtain the tenth seismic image feature; the tenth seismic image feature is input into the fifth pooling layer for feature selection and information filtering to obtain the target seismic image feature.

[0087] In some specific embodiments, a seismic image quality quantification evaluation network is used to learn the target seismic image features to obtain an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers, which further includes the following steps:

[0088] After any of the third to tenth convolutional layers, before inputting the next convolutional layer or pooling layer, the seismic image features output by the corresponding convolutional layer are input into the activation function for processing. The activation function is

[0089]

[0090] Among them, g and w i 、a i They represent output, weight and input respectively, and i represents the sequence number.

[0091] In some specific embodiments, the seismic data slice seismic image of the work area to be tested is divided into different small areas, each area is called a patch, and the area patch size of the seismic data slice seismic image of the work area to be tested input into the biased forward gather feature extraction network is 32×32, and end-to-end training is performed, and the feature extraction function is:

[0092] y R =F(x, {W i})+x1

[0093] The function of each convolutional layer is F(x), and the feature vector from the first convolutional layer is x 1 , Wi is the weight of the i-th convolution layer, y R Represents the feature vector output after unsupervised learning;

[0094] The convolution kernel size used in all convolutional layers is 3×3, and the convolution kernel adopts a zero-padding strategy to make the input and output sizes of each convolutional layer consistent, and all pooling layers are set to 2×2.

[0095] In some specific embodiments, a seismic image quality quantification evaluation network is used to learn the target seismic image features to obtain an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers, including the following steps: setting the probability of temporarily discarding the neural network unit of the seismic image quality quantification evaluation network from the network to 0.5, regressing and scoring each area of ​​the target seismic image features through the first fully connected layer, learning weights through the second fully connected layer, and assigning the learned weights to the scoring results of each corresponding area of ​​the target seismic image features, thereby obtaining an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers. Setting the probability of temporarily discarding the neural network unit of the seismic image quality quantification evaluation network from the network to 0.5 can prevent overfitting.

[0096] Two fully connected layers (FC) are used for regression, and the obtained features are input into two FC sub-networks. Both FCs are fully connected layers of 512 and 1. One sub-network is used to regress and score the patch, and the other sub-network is used to learn weights. The learned weights are then assigned to the patch. By regressing the quality of each patch, the quantitative evaluation result of the entire image is finally obtained.

[0097] Step S3: input the two-dimensional pre-biased gather data to be evaluated into the pre-biased gather amplitude-preserving intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data.

[0098] In some specific embodiments, the deep learning-based intelligent quantitative evaluation method for amplitude preservation of biased forward gathers includes the following steps:

[0099] Inputting the acquired seismic image into the convolution layer in the convolutional neural network model to obtain the first seismic image feature;

[0100] Inputting the first seismic image feature into the pooling layer of the convolutional neural network to obtain the target seismic image feature, wherein the number of the convolutional layers may be twice the number of the pooling layers, and the seismic image is input into a pooling layer for feature extraction and screening after being extracted by two convolutional layers, wherein the convolution kernel size of the convolutional layer is 3×3 pixels, and the size of the pooling layer is 2×2 pixels;

[0101] The quantitative evaluation output includes:

[0102] Inputting the target seismic image features obtained by the convolution layer and the pooling layer into the scoring fully connected layer in the convolutional neural network to perform regression scoring on the seismic image features;

[0103] Inputting the target seismic image features obtained by the convolution layer and the pooling layer into the weight learning fully connected layer in the convolutional neural network to obtain the feature weights of the target seismic image features;

[0104] Based on the regression score and feature weight of the target seismic image, a quantitative amplitude preservation evaluation result of the seismic image is obtained, and a trained quantitative amplitude preservation evaluation model is obtained;

[0105] After obtaining the trained amplitude preservation quantitative evaluation model, the amplitude preservation quantitative evaluation prediction can be performed based on the trained amplitude preservation quantitative evaluation model, including:

[0106] The two-dimensional biased forward gather data to be evaluated is input into the trained amplitude-preserving quantitative evaluation model to obtain the amplitude-preserving quantitative evaluation numerical output corresponding to the two-dimensional biased forward gather data to be evaluated.

[0107] Figure 2 represents the seismic data image to be evaluated, which is the input part of the network. Figure 3 It represents the results of quantitative evaluation and is the output of the network. Figure 4 This is a specific flow chart of unsupervised quantitative evaluation and prediction. The seismic data image to be evaluated is first subjected to feature extraction. Then the extracted features are passed through two modules consisting of two fully connected networks, one module for evaluation and the other module for weight analysis. The results of the two modules are then fused through spatial pooling to obtain the evaluation score. Figure 5The schematic diagram of the structure of the deep learning network for quantitative evaluation is shown in Figure 1. The network structure consists of two parts: one is the part that extracts information from seismic images, and the other is the part that quantifies the quality of this information. In the first part, 10 convolutional layers are used, with the number of channels being 32, 32, 64, 64, 128, 128, 256, 256, 512, 512, and 5 pooling layers. All convolutional layers use 3×3 convolution kernels, and ReLU is used for nonlinear transformation to achieve optimal performance. The convolution kernel uses a zero-filling strategy, and the pooling layers are all set to 2×2. In order to prevent overfitting, the dropout in the fully connected layer is set to 0.5. Dropout refers to the probability of temporarily discarding a neural network unit from the network during the training process of the deep learning network. The network input is a 32*32 patch, which belongs to an end-to-end training network. After that, two fully connected layers are used for regression, and the obtained features are input into two FC sub-networks, both of which are fully connected layers with FC of 512 and 1. Then, the image quality assessment score is obtained through spatial pooling.

[0108] Specifically, the model training process includes:

[0109] Obtain seismic images from seismic data;

[0110] Then the seismic image is first Figure 5 The first two convolutional layers (Conv3-32) in the first part of the image are used to extract features and obtain the first seismic image features;

[0111] The extracted first seismic image feature is processed by an activation function, and after the processing, a second seismic image feature is obtained;

[0112] The obtained second seismic image features can then be input into the above Figure 5 In the first pooling layer (Conv3-64) of the second part, feature selection and information filtering are performed to obtain the third seismic image features;

[0113] The obtained third seismic image feature is input into the above Figure 5 The second part of the image is extracted from two convolutional layers (Conv3-64) adjacent to the pooling layer to obtain the fourth seismic image feature;

[0114] Input the fourth seismic image feature into the pooling layer of the Conv3-128 part for feature selection and information filtering to obtain the fifth seismic image feature;

[0115] Input the fifth seismic image feature into the two convolutional layers of the Conv3-128 part for feature extraction to obtain the sixth image feature;

[0116] Input the sixth seismic image feature into the pooling layer of the Conv3-256 part for feature selection and information filtering to obtain the seventh seismic image feature;

[0117] Input the seventh seismic image feature into the two convolutional layers of the Conv3-256 part for feature extraction to obtain the eighth image feature;

[0118] Input the eighth seismic image feature into the pooling layer of the Conv3-512 part for feature selection and information filtering to obtain the ninth seismic image feature;

[0119] Input the ninth seismic image feature into the two convolutional layers of the Conv3-512 part for feature extraction to obtain the tenth seismic image feature;

[0120] The tenth seismic image feature is input into the last pooling layer for feature selection and information filtering to obtain the target seismic image feature.

[0121] It should be noted that after each convolutional layer processing, it will be processed by the activation function, and the network input of the seismic image is a 32*32 patch, which belongs to the end-to-end training network, that is, it is carried out in different regions.

[0122] Finally, the obtained target seismic image features are input into the last two fully connected layers (FC-512, FC-1), one of which is responsible for scoring and the other is responsible for learning weights. Finally, the scoring of the target seismic image features by the two fully connected layers and the learned weights are combined to obtain the final result. The learned weights are assigned to the regression scoring results of each area respectively.

[0123] The model usage process includes:

[0124] The corresponding quantitative evaluation value output can be obtained by inputting the two-dimensional bias-forward gather data to be evaluated into the trained amplitude-preserving quantitative evaluation model.

[0125] Finally, the experimental data are combined to illustrate:

[0126] Two different datasets and their parts in different working areas were used in the experiment. The results and their analysis were obtained by experimenting on the synthetic vertical seismic profile (VSP) dataset and the more representative F3 Netherlands dataset. The obtained two-dimensional and three-dimensional results are plotted respectively: Figure 6 and Figure 8 are slices of two different work areas on the synthetic vertical seismic profile (VSP) dataset, are seismic images to be evaluated, Figure 7 and Fig. 9 It is the result output after evaluating the two slice images. Fig.10is a slice image from F3 data, Fig.11 Yes Fig.10 quality assessment results.

[0127] Figure 7 , 9 , 11 shows the two-dimensional results of the application of the present invention, Figures 12 to 15 The performance on three-dimensional data is shown. Fig.12 and 14 The three-dimensional data volumes of VSP and F3 are respectively represented and used as network input to obtain Fig.13 and 15 The results shown indicate that the present invention has the same superior performance on both two-dimensional data and three-dimensional data.

[0128] There are three main challenges currently faced by seismic image quality assessment: ① It is difficult or even impossible to obtain reference images in most scenarios; ② Seismic data noise is diverse, and the generalization ability of traditional quality assessment algorithms is insufficient to match this diversity; ③ Since it is time-consuming and labor-intensive to manually identify data sets, there is currently no labeled data set that can be used for neural network training. In order to meet the above challenges, the present invention proposes a seismic image quality assessment method based on unsupervised training, which uses a convolutional neural network to perform quality assessment on the amplitude preservation of biased forward gather seismic data. The experimental results on the synthetic vertical seismic profile data set and the F3 data set show that our algorithm can achieve quantitative evaluation of the amplitude preservation of geological data under unsupervised conditions, establish a qualified quality assessment model, and obtain quality assessment reference images in any scenario. The quality assessment results can reflect a variety of seismic data noises, and there is no need to manually identify data or use labeled data sets.

[0129] In the above method of this embodiment, it is possible to implement quantitative evaluation of the amplitude preservation of geological data under unsupervised conditions, and a qualified unlabeled trained biased forward gather amplitude preservation intelligent quantitative evaluation model is established. Using the established biased forward gather amplitude preservation intelligent quantitative evaluation model, it is possible to intelligently predict the amplitude preservation quantitative parameters in a data-driven manner without relying on any prior information, thereby improving the accuracy of lithologic oil and gas reservoir prediction.

[0130] Those skilled in the art can change the above sequence without departing from the protection scope of the present disclosure.

[0131] Another embodiment of the present invention provides an intelligent quantitative evaluation system for amplitude preservation of biased forward gathers based on deep learning, comprising:

[0132] The quantitative evaluation model determination module is used to construct an unsupervised deep learning model; the seismic data slices of the work area to be tested are input into the unsupervised deep learning model for training to obtain an intelligent quantitative evaluation model for amplitude preservation of the biased forward gather;

[0133] The quantitative evaluation module is used to input the two-dimensional pre-biased gather data to be evaluated into the pre-biased gather amplitude-preserving intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data.

[0134] Regarding the system 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.

[0135] In this embodiment, the amplitude preservation of geological data can be quantitatively evaluated under unsupervised conditions, and a qualified unlabeled trained biased forward gather amplitude preservation intelligent quantitative evaluation model is established. Using the established biased forward gather amplitude preservation intelligent quantitative evaluation model, the amplitude preservation quantitative parameters can be intelligently predicted in a data-driven manner without relying on any prior information, thereby improving the accuracy of lithologic oil and gas reservoir prediction.

[0136] Another embodiment of the present invention provides a method for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers, comprising the following steps:

[0137] Constructing a biased forward gather feature extraction network, including sequentially setting 10 convolutional layers with channel numbers of 32, 32, 64, 64, 128, 128, 256, 256, 512, and 512, respectively, the first convolutional layer is the input end of the biased forward gather feature extraction network, the first pooling layer is set between the second and third convolutional layers, the second pooling layer is set between the fourth and fifth convolutional layers, the third pooling layer is set between the sixth and seventh convolutional layers, the fourth pooling layer is set between the eighth and ninth convolutional layers, the fifth pooling layer is set after the tenth convolutional layer, and the fifth pooling layer is the output end of the biased forward gather feature extraction network;

[0138] A seismic image quality quantitative assessment network is constructed, and a first fully connected layer with a channel number of 512 and a second fully connected layer with a channel number of 1 are sequentially set after the output end of the biased forward gather feature extraction network, and the second fully connected layer is the output end of the seismic image quality quantitative assessment network;

[0139] Input the seismic image of the seismic data slices in the work area to be tested into the biased forward gather feature extraction network to obtain the target seismic image features;

[0140] The seismic image quality quantitative assessment network is used to learn the characteristics of the target seismic image, and an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers is obtained.

[0141] In this embodiment, a qualified unlabeled trained biased forward gather amplitude preservation intelligent quantitative evaluation model can be established. Using the established biased forward gather amplitude preservation intelligent quantitative evaluation model, the amplitude preservation of geological data can be quantitatively evaluated under unsupervised conditions without relying on any prior information, and the amplitude preservation quantitative parameters can be intelligently predicted in a data-driven manner, thereby improving the accuracy of lithologic oil and gas reservoir prediction.

[0142] Another embodiment of the present invention provides a system for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers, comprising:

[0143] A network construction module is used to construct a biased forward gather feature extraction network, including sequentially setting 10 convolutional layers with channel numbers of 32, 32, 64, 64, 128, 128, 256, 256, 512, and 512, respectively, the first convolutional layer is the input end of the biased forward gather feature extraction network, the first pooling layer is set between the second and third convolutional layers, the second pooling layer is set between the fourth and fifth convolutional layers, the third pooling layer is set between the sixth and seventh convolutional layers, the fourth pooling layer is set between the eighth and ninth convolutional layers, and the fifth pooling layer is set after the tenth convolutional layer, and the fifth pooling layer is the output end of the biased forward gather feature extraction network; constructing a seismic image quality quantitative assessment network, sequentially setting a first fully connected layer with a channel number of 512 and a second fully connected layer with a channel number of 1 after the output end of the biased forward gather feature extraction network, and the second fully connected layer is the output end of the seismic image quality quantitative assessment network;

[0144] The feature extraction module is used to construct a biased forward gather feature extraction network by inputting the seismic data slice seismic image of the work area to be tested, and obtain the target seismic image features;

[0145] The quantitative evaluation model determination module is used to learn the target seismic image characteristics through the seismic image quality quantitative evaluation network to obtain the amplitude-preserving intelligent quantitative evaluation model of the biased forward gather.

[0146] In this embodiment, a qualified unlabeled trained biased forward gather amplitude preservation intelligent quantitative evaluation model can be established. Using the established biased forward gather amplitude preservation intelligent quantitative evaluation model, the amplitude preservation of geological data can be quantitatively evaluated under unsupervised conditions without relying on any prior information, and the amplitude preservation quantitative parameters can be intelligently predicted in a data-driven manner, thereby improving the accuracy of lithologic oil and gas reservoir prediction.

[0147] Another embodiment of the present invention provides an intelligent quantitative evaluation method for amplitude preservation of biased forward gathers based on deep learning, comprising the following steps:

[0148] The two-dimensional pre-biased gather data to be evaluated is input into a pre-biased gather amplitude preservation intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data; the pre-biased gather amplitude preservation intelligent quantitative evaluation model is used to extract and learn target seismic image features.

[0149] In this embodiment, the amplitude preservation of geological data can be quantitatively evaluated under unsupervised conditions, and a qualified unlabeled trained biased forward gather amplitude preservation intelligent quantitative evaluation model is established. Using the established biased forward gather amplitude preservation intelligent quantitative evaluation model, the amplitude preservation quantitative parameters can be intelligently predicted in a data-driven manner without relying on any prior information, thereby improving the accuracy of lithologic oil and gas reservoir prediction.

[0150] Another embodiment of the present invention provides an intelligent quantitative evaluation system for amplitude preservation of biased forward gathers based on deep learning, comprising:

[0151] The quantitative evaluation module is used to input the two-dimensional pre-biased gather data to be evaluated into the pre-biased gather amplitude preservation intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data; the pre-biased gather amplitude preservation intelligent quantitative evaluation model is used to extract and learn the target seismic image features.

[0152] In this embodiment, the amplitude preservation of geological data can be quantitatively evaluated under unsupervised conditions, and a qualified unlabeled trained biased forward gather amplitude preservation intelligent quantitative evaluation model is established. Using the established biased forward gather amplitude preservation intelligent quantitative evaluation model, the amplitude preservation quantitative parameters can be intelligently predicted in a data-driven manner without relying on any prior information, thereby improving the accuracy of lithologic oil and gas reservoir prediction.

[0153] Based on the same inventive concept, an embodiment of the present invention provides an electronic device, whose structure is as follows Fig.16 As shown, it includes: a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the aforementioned intelligent quantitative evaluation method for amplitude preservation of biased forward trace gathers based on deep learning or the method for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward trace gathers is implemented.

[0154] Based on the same inventive concept, 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, the aforementioned deep learning-based intelligent quantitative evaluation method for amplitude preservation of biased forward trace gathers or the method for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward trace gathers is implemented.

[0155] Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention shall still fall within the patent scope of the present invention.

Claims

1. An intelligent quantitative evaluation method for amplitude preservation of biased forward gathers based on deep learning, characterized in that: The following steps are involved: Build unsupervised deep learning models; The seismic data slices of the work area to be tested are input into the unsupervised deep learning model for training, and an intelligent quantitative evaluation model for amplitude preservation of the biased forward gathers is obtained; The two-dimensional pre-bias gather data to be evaluated is input into the pre-bias gather amplitude-preserving intelligent quantitative evaluation model to obtain the corresponding quantitative evaluation data.

2. The method according to claim 1, characterized in that Building an unsupervised deep learning model involves the following steps: Constructing a biased forward gather feature extraction network, including sequentially setting 10 convolutional layers with channel numbers of 32, 32, 64, 64, 128, 128, 256, 256, 512, and 512, respectively, the first convolutional layer is the input end of the biased forward gather feature extraction network, the first pooling layer is set between the second and third convolutional layers, the second pooling layer is set between the fourth and fifth convolutional layers, the third pooling layer is set between the sixth and seventh convolutional layers, the fourth pooling layer is set between the eighth and ninth convolutional layers, the fifth pooling layer is set after the tenth convolutional layer, and the fifth pooling layer is the output end of the biased forward gather feature extraction network; A seismic image quality quantitative assessment network is constructed. After the output end of the biased forward gather feature extraction network, a first fully connected layer with a channel number of 512 and a second fully connected layer with a channel number of 1 are sequentially set. The second fully connected layer is the output end of the seismic image quality quantitative assessment network.

3. The method according to claim 2, characterized in that The seismic data slices of the work area to be tested are input into the unsupervised deep learning model for training to obtain the intelligent quantitative evaluation model of amplitude preservation of the biased forward gather, including the following steps: Input the seismic image of the seismic data slices in the work area to be tested into the biased forward gather feature extraction network to obtain the target seismic image features; The seismic image quality quantitative assessment network is used to learn the characteristics of the target seismic image, and an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers is obtained.

4. The method according to claim 3, characterized in that The seismic data slices of the work area to be tested are input into the construction of a biased forward gather feature extraction network to obtain the target seismic image features, including the following steps: Slice the seismic image of the seismic data of the work area to be measured into multiple regions, input the regions into the input end of the biased forward gather feature extraction network, perform feature extraction through the first and second convolutional layers, and obtain the first seismic image feature; Inputting the first seismic image feature into the activation function for processing to obtain a second seismic image feature; Inputting the second seismic image feature into the first pooling layer for feature selection and information filtering to obtain a third seismic image feature; After the third and fourth convolutional layers, feature extraction is performed on the third seismic image feature to obtain a fourth seismic image feature; Inputting the fourth seismic image feature into the second pooling layer for feature selection and information filtering to obtain a fifth seismic image feature; After the fifth and sixth convolutional layers, the fifth seismic image feature is extracted to obtain the sixth seismic image feature; Inputting the sixth seismic image feature into the third pooling layer for feature selection and information filtering to obtain the seventh seismic image feature; After the seventh and eighth convolutional layers, feature extraction is performed on the seventh seismic image feature to obtain the eighth seismic image feature; Inputting the eighth seismic image feature into the fourth pooling layer for feature selection and information filtering to obtain a ninth seismic image feature; After the ninth and tenth convolutional layers, the ninth seismic image feature is extracted to obtain the tenth seismic image feature; The tenth seismic image feature is input into the fifth pooling layer for feature selection and information filtering to obtain the target seismic image feature.

5. The method according to claim 3, characterized in that The target seismic image features are learned by using a seismic image quality quantitative assessment network to obtain an intelligent quantitative assessment model for amplitude preservation of biased forward gathers, which also includes the following steps: After any of the third to tenth convolutional layers, before inputting the next convolutional layer or pooling layer, the seismic image features output by the corresponding convolutional layer are input into the activation function for processing. The activation function is Among them, g and w i 、a i They represent output, weight and input respectively, and i represents the sequence number.

6. The method according to claim 4, characterized in that The area size of the seismic data slice of the work area to be tested is 32×32 for the input of the biased forward gather feature extraction network. End-to-end training is performed and the feature extraction function is: y R =F(x,{W i })+x1 The function of each convolutional layer is F(x), the feature vector from the first convolutional layer is x1, Wi is the weight of the i-th convolution layer, y R Represents the feature vector output after unsupervised learning; The convolution kernel size used in all convolutional layers is 3×3, and the convolution kernel adopts a zero-padding strategy to make the input and output sizes of each convolutional layer consistent, and all pooling layers are set to 2×2.

7. The method according to claim 3, characterized in that The seismic image quality quantitative assessment network is used to learn the target seismic image features and obtain an intelligent quantitative assessment model for amplitude preservation of biased forward gathers, which includes the following steps: The probability of temporarily discarding the neural network unit of the seismic image quality quantitative assessment network from the network is set to 0.

5. Each area of ​​the target seismic image feature is regressed and scored through the first fully connected layer. The weights are learned through the second fully connected layer, and the learned weights are assigned to the scoring results of each corresponding area of ​​the target seismic image feature. An intelligent quantitative evaluation model for amplitude preservation of biased forward gathers is obtained.

8. An intelligent quantitative evaluation system for amplitude preservation of biased forward gathers based on deep learning, characterized in that: include: Quantitative evaluation model determination module, used to build unsupervised deep learning models; The seismic data slices of the work area to be tested are input into the unsupervised deep learning model for training, and an intelligent quantitative evaluation model for amplitude preservation of the biased forward gathers is obtained; The quantitative evaluation module is used to input the two-dimensional pre-biased gather data to be evaluated into the pre-biased gather amplitude-preserving intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data.

9. A method for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers, characterized in that: The following steps are involved: Constructing a biased forward gather feature extraction network, including sequentially setting 10 convolutional layers with channel numbers of 32, 32, 64, 64, 128, 128, 256, 256, 512, and 512, respectively, the first convolutional layer is the input end of the biased forward gather feature extraction network, the first pooling layer is set between the second and third convolutional layers, the second pooling layer is set between the fourth and fifth convolutional layers, the third pooling layer is set between the sixth and seventh convolutional layers, the fourth pooling layer is set between the eighth and ninth convolutional layers, the fifth pooling layer is set after the tenth convolutional layer, and the fifth pooling layer is the output end of the biased forward gather feature extraction network; A seismic image quality quantitative assessment network is constructed, and a first fully connected layer with a channel number of 512 and a second fully connected layer with a channel number of 1 are sequentially set after the output end of the biased forward gather feature extraction network, and the second fully connected layer is the output end of the seismic image quality quantitative assessment network; Input the seismic image of the seismic data slices in the work area to be tested into the biased forward gather feature extraction network to obtain the target seismic image features; The seismic image quality quantitative assessment network is used to learn the characteristics of the target seismic image, and an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers is obtained.

10. A system for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward gathers, characterized in that: include: A network construction module is used to construct a biased forward gather feature extraction network, including sequentially setting 10 convolutional layers with channel numbers of 32, 32, 64, 64, 128, 128, 256, 256, 512, and 512, respectively, the first convolutional layer is the input end of the biased forward gather feature extraction network, the first pooling layer is set between the second and third convolutional layers, the second pooling layer is set between the fourth and fifth convolutional layers, the third pooling layer is set between the sixth and seventh convolutional layers, the fourth pooling layer is set between the eighth and ninth convolutional layers, and the fifth pooling layer is set after the tenth convolutional layer, and the fifth pooling layer is the output end of the biased forward gather feature extraction network; constructing a seismic image quality quantitative assessment network, sequentially setting a first fully connected layer with a channel number of 512 and a second fully connected layer with a channel number of 1 after the output end of the biased forward gather feature extraction network, and the second fully connected layer is the output end of the seismic image quality quantitative assessment network; The feature extraction module is used to construct a biased forward gather feature extraction network by inputting the seismic data slice seismic image of the work area to be tested, and obtain the target seismic image features; The quantitative evaluation model determination module is used to learn the target seismic image characteristics through the seismic image quality quantitative evaluation network to obtain the amplitude-preserving intelligent quantitative evaluation model of the biased forward gather.

11. An intelligent quantitative evaluation method for amplitude preservation of biased forward gathers based on deep learning, characterized in that: The following steps are involved: The two-dimensional pre-biased gather data to be evaluated is input into a pre-biased gather amplitude preservation intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data; the pre-biased gather amplitude preservation intelligent quantitative evaluation model is used to extract and learn target seismic image features.

12. An intelligent quantitative evaluation system for amplitude preservation of biased forward gathers based on deep learning, characterized in that: include: A quantitative evaluation module is used to input the two-dimensional pre-biased gather data to be evaluated into the pre-biased gather amplitude-preserving intelligent quantitative evaluation model to obtain corresponding quantitative evaluation data; The biased forward gather amplitude preservation intelligent quantitative evaluation model is used to extract and learn target seismic image features.

13. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the computer program, it implements the deep learning-based intelligent quantitative evaluation method for amplitude preservation of biased forward trace gathers as described in any one of claims 1 to 7 and 11, or the method for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward trace gathers as described in claim 9.

14. A computer storage medium, characterized in that: The computer storage medium stores computer executable instructions, which, when executed, implement the deep learning-based intelligent quantitative evaluation method for amplitude preservation of biased forward trace gathers as described in any one of claims 1 to 7 and 11, or the method for constructing an intelligent quantitative evaluation model for amplitude preservation of biased forward trace gathers as described in claim 9.

Images