Deep learning-based two-dimensional geological profile map generation method and system
By constructing a deep learning network model, using Transformer's attention mechanism and two-dimensional CNN coding layer, combined with random noise, the problem of difficulty in taking into account the spatial correlation in the entire modeling area is solved, and the two-dimensional geological profile diagram is improved Generation quality and degree of refinement.
Patent Information
- Application Number
- CN202411072226.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-06
- Publication Date
- 2024-11-08
AI Technical Summary
The existing geological profile generation methods are difficult to take into account the spatial correlation of all prior data in the entire modeling area.
Using a deep learning method, a deep learning network model including a two-dimensional CNN encoding layer, a Transformer decoding layer and an inverse mapping layer is constructed. Through formation information encoding and spatial position information encoding, and combined with random noise, a two-dimensional geological profile map is generated.
The generation quality and refinement of the two-dimensional geological profile map are improved, and its application value is enhanced. It can automatically generate two-dimensional geological profile maps that fit known drills.
Smart Images

Figure CN118918208A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of geological modeling, and in particular to a method and system for generating a two-dimensional geological profile based on deep learning. Background Art
[0002] Geological modeling is a method of establishing the two-dimensional or three-dimensional spatial distribution of target geological bodies by means of digitalization and visualization, using geological expertise and geological survey data. In recent years, following the digital earth and smart cities, the introduction of digital twins has further promoted the digitization, informatization and intelligence of various industries. Geological modeling technologies such as geological profile generation can clearly represent the distribution of geological bodies in underground space, which is conducive to increasing people's understanding of underground space and providing important data support for the quantitative evaluation of underground space required for the exploitation of mines and oil fields. It is the basis and key technology of digital twin mines and digital twins of oil, and is also an important part of digital earth and smart cities.
[0003] A geological profile is a geological model that describes the distribution of stratigraphic structures and rock layer properties in a certain section of underground three-dimensional geological space in two dimensions. It has always been an important map for people to understand and analyze underground space. The cooperation between two-dimensional geological modeling and three-dimensional geological modeling technology not only plays a major role in engineering exploration and mineral mining, but also develops rapidly in more and more aspects such as urban planning and construction, water resource management and geological disaster analysis. It is the key technical support for promoting the three-dimensional reconstruction of above-ground and underground entities and the development of urban digital twin technology.
[0004] With the rapid development of artificial intelligence, artificial intelligence technology has injected fresh blood into the traditional field of geology and brought more new methods to geological modeling. For example, machine learning methods such as BP neural network, support vector machine, random forest and deep learning methods such as convolutional neural network, generative adversarial network, variational autoencoder have made certain progress in reservoir modeling, hydrogeological modeling, geological structure modeling and other fields, promoting the further development of geological modeling technology.
[0005] Most existing geological modeling methods can consider local spatial correlation or spatial correlation in certain directions, but it is difficult to take into account the spatial correlation of all prior data in the entire modeling area. In recent years, Transformer, as an emerging deep learning technology, has achieved outstanding results in natural language processing, computer vision and other fields. Its attention mechanism can better establish the correlation between known data and unrequired data. However, there is still a lack of research on the application of attention mechanism in the field of geological profile generation. Summary of the invention
[0006] The present application provides a two-dimensional geological profile generation method and system based on deep learning to solve the problem that the existing geological profile generation method is difficult to take into account the spatial correlation of all prior data in the entire modeling area.
[0007] According to the first aspect, an embodiment provides a method for generating a two-dimensional geological profile based on deep learning, the method comprising:
[0008] Constructing a sample set of two-dimensional geological profiles, randomly extracting local images from the two-dimensional geological profiles to obtain virtual borehole images, and constructing a training set and a test set based on the extracted virtual borehole images;
[0009] Constructing a deep learning network model, the deep learning network model includes a coding layer based on two-dimensional CNN, a decoding layer based on Transformer, and an inverse mapping layer, inputting a virtual borehole image into the coding layer to encode stratum information and spatial position information to obtain a stratum information coding sequence and a spatial position information coding sequence, adding the stratum information coding sequence, the spatial position information coding sequence, and random noise, and then inputting them into the decoding layer for decoding, and inputting the decoded information into the inverse mapping layer to restore it to the pixel layer to obtain a one-dimensional geological profile pixel value result;
[0010] Using the training set and the test set to train and test the deep learning network model;
[0011] The collected measured borehole image data is processed and input into the trained deep learning network model, and the pixel value results of the one-dimensional geological profile output by the model are restored to obtain a two-dimensional geological profile.
[0012] Furthermore, a local image is randomly extracted from the two-dimensional geological profile to obtain a virtual borehole image, specifically including:
[0013] According to all the stratum types appearing in the two-dimensional geological profile sample set, a stratum coding comparison table is prepared, and label coding is performed on each stratum;
[0014] According to the prepared stratigraphic code comparison table, the attribute value represented by a certain point in the two-dimensional geological section map is set according to the stratigraphic code comparison table;
[0015] If the original two-dimensional geological profile is in vector form, regular sampling is performed to determine which stratum the sampling point is located above, and the attribute value of the sampling point is set according to the stratum coding comparison table, thereby converting the two-dimensional geological profile in vector form into a raster image.
[0016] Furthermore, a local image is randomly extracted from the two-dimensional geological profile to obtain a virtual borehole image, specifically including:
[0017] The encoded two-dimensional geological profile is randomly masked, a part of the image with a preset size is retained as the virtual borehole image, and the rest is assigned to 0; the maximum and minimum value normalization method is used to normalize the obtained virtual borehole image.
[0018] Furthermore, the virtual borehole image is input into the coding layer to perform stratigraphic information coding and spatial position information coding to obtain a stratigraphic information coding sequence and a spatial position information coding sequence, specifically including:
[0019] The input image is divided into blocks by convolution. The size of the two-dimensional convolution kernel is the size of each two-dimensional pixel block. The step size of the two-dimensional convolution is the same as the size of the convolution kernel. Each divided pixel block is mapped to a word vector separately.
[0020] The feature vectors of the input image after two-dimensional convolution are grouped into a sequence. The length of the sequence is the number of blocks the input image is divided into. The dimension of the word vector in the sequence is the number of pixels in the pixel block after the block division, thus obtaining the stratigraphic information coding sequence.
[0021] Furthermore, the virtual borehole image is input into the coding layer to perform stratigraphic information coding and spatial position information coding to obtain a stratigraphic information coding sequence and a spatial position information coding sequence, specifically including:
[0022] The [0,1] range coding combined with compression coding is used to position code the two-dimensional geological drilling images after segmentation, and the relative coordinates of each pixel in the two-dimensional geological drilling images are coded. Then the coordinates of each pixel are normalized to obtain the [0,1] range coding of the spatial position information of each pixel.
[0023] The two-dimensional position information encoded in the [0,1] range is subjected to two-dimensional convolution. The size and step size of the two-dimensional convolution kernel are both the size of the pixel block. The spatial position feature vectors of the two-dimensional geological borehole image after the two-dimensional convolution are combined into a sequence. The length of the sequence is the number of blocks the two-dimensional borehole image is divided into. The dimension of the word vector in the sequence is the number of pixels in the pixel block after the division, thus obtaining the spatial position information encoding sequence.
[0024] Furthermore, the stratigraphic information coding sequence, the spatial position information coding sequence and the random noise are added and input into the decoding layer for decoding, which specifically includes:
[0025] The random noise is obtained by linear mapping a random vector. The length of the random vector is the length of the word vector sequence input to the decoding layer. The number of channels is 1 and conforms to the normal distribution. The random vector is input to the linear mapping layer, and the number of channels of the random vector is mapped from 1 to the number of channels of the input decoding layer.
[0026] Furthermore, the deep learning network model also includes an inverse mapping layer;
[0027] The inverse mapping layer is used to restore the decoded information to the pixel layer to obtain the pixel value of the one-dimensional geological profile; and a two-dimensional geological profile can be obtained by restoring the pixel value of the one-dimensional geological profile.
[0028] Furthermore, the deep learning network model is trained and tested using the training set and the test set, specifically including:
[0029] A multi-weighted cross entropy loss function is constructed, and the loss between the model prediction value and the true value is calculated based on the loss function. The network is trained using backpropagation and stochastic gradient descent methods.
[0030] Furthermore, a multi-weighted cross entropy loss function is constructed, and the loss between the model prediction value and the true value is calculated based on the loss function, specifically including:
[0031] The loss function weight l0 of the drilling area is set to w0, and the loss function l0 of the first-level neighborhood of the drilling area is set to w0. n1 The weight is set to w1, and the loss function of the secondary neighborhood is l n2 The weight is set to w2, and the loss function of the remaining regions is l n3 The weight is set to w3, that is, the loss function is loss = w0l0 + w1l n1 +w2l n2 +w3l n3 ;
[0032] Among them, the first-level neighborhood of the borehole is the set of eight-plane neighborhoods of each pixel on the borehole; the second-level neighborhood of the borehole is the set of twenty-five-plane neighborhoods of each pixel on the borehole; the first-level and second-level neighborhoods of the borehole are set according to a preset ratio.
[0033] According to the second aspect, an embodiment provides a system for generating a two-dimensional geological profile based on deep learning, the system comprising:
[0034] A data set construction module is used to construct a sample set of two-dimensional geological profiles, randomly extract local images from the two-dimensional geological profiles to obtain virtual borehole images, and construct a training set and a test set based on the extracted virtual borehole images;
[0035] A model building module is used to build a deep learning network model, wherein the deep learning network model includes a coding layer based on a two-dimensional CNN, a decoding layer based on a Transformer, and an inverse mapping layer, wherein a virtual borehole image is input into the coding layer for stratigraphic information coding and spatial position information coding to obtain a stratigraphic information coding sequence and a spatial position information coding sequence, wherein the stratigraphic information coding sequence, the spatial position information coding sequence, and random noise are added and then input into the decoding layer for decoding, and the decoded information is input into the inverse mapping layer for restoration to a pixel layer to obtain a one-dimensional geological profile pixel value result;
[0036] A model training module, used to train and test the deep learning network model using the training set and the test set;
[0037] The two-dimensional geological profile generation module is used to process the collected measured drilling image data and input it into the trained deep learning network model, and restore the pixel value results of the one-dimensional geological profile output by the model to obtain a two-dimensional geological profile.
[0038] According to a third aspect, an embodiment provides an electronic device, the device comprising: a processor and a memory;
[0039] The memory is used to store one or more program instructions;
[0040] The processor is used to run one or more program instructions to execute the steps of a method for generating a two-dimensional geological profile based on deep learning as described in any of the above items.
[0041] According to the fourth aspect, an embodiment provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for generating a two-dimensional geological profile based on deep learning as described in any one of the above items are implemented.
[0042] The present application provides a method and system for generating a two-dimensional geological profile based on deep learning, constructing a two-dimensional geological profile sample set, randomly extracting local images from the two-dimensional geological profile to obtain virtual borehole images, and constructing a training set and a test set based on the extracted virtual borehole images; constructing a deep learning network model, the deep learning network model includes a coding layer based on a two-dimensional CNN, a decoding layer based on a Transformer, and an inverse mapping layer, inputting the virtual borehole image into the coding layer for stratigraphic information coding and spatial position information coding to obtain a stratigraphic information coding sequence and a spatial position information coding sequence, adding the stratigraphic information coding sequence, the spatial position information coding sequence and random noise and inputting them into the decoding layer for decoding, inputting the decoded information into the inverse mapping layer for restoration to the pixel layer to obtain a one-dimensional geological profile pixel value result; using the training set and the test set to train and test the deep learning network model; processing the collected measured borehole image data and inputting it into the trained deep learning network model, and restoring the one-dimensional geological profile pixel value result output by the model to obtain a two-dimensional geological profile. The present invention constructs a coding layer based on a two-dimensional CNN, making it possible to apply the Transformer, which is good at establishing global correlation, in the field of two-dimensional geological profile generation. It also extracts stratigraphic information and location information, constructs a multi-weighted cross entropy loss function to strengthen the training of borehole image data and its neighborhood, and constructs random noise to assist training and denoising, and automatically generates a two-dimensional geological profile that fits a known borehole, thereby improving the quality and refinement of the generated two-dimensional geological profile, thereby enhancing its application value. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 A flowchart of a method for generating a two-dimensional geological profile based on deep learning provided by one embodiment of the present invention;
[0044] Figure 2 A schematic diagram of the logical structure of a two-dimensional geological profile generation system based on deep learning provided for an embodiment of the present invention. DETAILED DESCRIPTION
[0045] The present invention is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are for making the present application better understood. However, those skilled in the art can easily recognize that some features can be omitted in different situations, or can be replaced by other elements, materials, methods. In some cases, some operations related to the present application are not shown or described in the specification, this is to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations according to the description in the specification and the general technical knowledge in the art.
[0046] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.
[0047] A method for generating a two-dimensional geological profile based on deep learning is provided in the first embodiment of the present invention. Figure 1 Provide detailed explanation.
[0048] like Figure 1 As shown, in step S100, a two-dimensional geological profile sample set is constructed, local images are randomly extracted from the two-dimensional geological profile to obtain virtual borehole images, and a training set and a test set are constructed based on the extracted virtual borehole images.
[0049] A sample set of two-dimensional geological profiles is constructed, and virtual boreholes are randomly selected from the two-dimensional geological profiles.
[0050] In this embodiment, the two-dimensional geological profile refers to the original geological profile. The measured borehole image data is the borehole image data in the actual exploration work, and the virtual borehole image data is the simulated borehole image data obtained from the two-dimensional geological profile through masking.
[0051] The specific steps for constructing a two-dimensional geological profile sample set and virtual borehole image data are as follows:
[0052] S110, preparing a stratum coding comparison table according to all stratum types appearing in the two-dimensional geological profile sample set, and labeling each stratum;
[0053] S120, according to the stratum code comparison table in S110, the attribute value represented by a certain point in the two-dimensional geological profile is set according to the stratum code comparison table;
[0054] S130, organizing the geological model in S120 into a raster image format to facilitate the construction of a deep learning data set. If the original two-dimensional geological profile is vector data, regular sampling is performed to determine which stratum the sampling point is located above, and then the attribute value of the sampling point is set in the manner of S120, thereby converting the vector data into raster data;
[0055] S140. According to the preset size of the geological profile, randomly extract images of a specified size from the two-dimensional geological profile constructed in S130 as the minimum unit for deep learning to generate the geological profile, and use 80% of the extracted two-dimensional geological profiles as training sets and 20% as test sets to construct a two-dimensional geological profile dataset for deep learning.
[0056] like Figure 1 As shown, in step S200, a deep learning network model is constructed, and the deep learning network model includes a coding layer based on two-dimensional CNN, a decoding layer based on Transformer, and an inverse mapping layer. The virtual drilling image is input into the coding layer to encode the stratigraphic information and the spatial position information to obtain a stratigraphic information coding sequence and a spatial position information coding sequence. The stratigraphic information coding sequence, the spatial position information coding sequence, and random noise are added and input into the decoding layer for decoding. The decoded information is input into the inverse mapping layer to be restored to the pixel layer to obtain a one-dimensional geological profile pixel value result.
[0057] This embodiment constructs a Transformer-based decoding layer, which uses the constructed two-dimensional geological profile sample set to learn geological modeling rules and obtain global relevant features in the geological profile according to the measured borehole image data.
[0058] Before using the decoding layer for decoding, the processing includes: randomly masking the two-dimensional geological profile raster image sorted in step S100, retaining part of the image as a virtual borehole, and assigning the rest to 0; normalizing the obtained virtual borehole image using the maximum and minimum normalization method. Set the number of blocks and the number of multi-head attention modules of the Transformer decoding layer, pre-set the number of blocks of the two-dimensional geological profile, calculate the dimension of the word vector sequence required to be input to the Transformer decoding layer, and encode the stratigraphic information and spatial position information of the input virtual borehole data.
[0059] This embodiment constructs a two-dimensional CNN-based encoding layer before the Transformer-based decoding layer to encode the stratigraphic information and spatial position information of the input two-dimensional geological profile data. Based on the encoding layer, the stratigraphic encoding information and spatial position encoding information of the two-dimensional geological profile extracted from the virtual borehole are mapped to a one-dimensional word vector with the same shape. The details are as follows:
[0060] A two-dimensional CNN is used to convolve the measured borehole image data, and the two-dimensional geological borehole image is divided into blocks by convolution. The size of the two-dimensional convolution kernel is the size of each two-dimensional pixel block, and the step size of the two-dimensional convolution is the same as the size of the convolution kernel, so that each borehole pixel block after blocking can be mapped to a word vector separately; the feature vectors of the two-dimensional geological borehole image after the two-dimensional convolution are combined into a sequence, the length of the sequence is the number of blocks the two-dimensional borehole image is divided into, and the dimension of the word vector in the sequence is the number of pixels in the pixel block after blocking, so that the stratigraphic information encoding sequence of the two-dimensional geological borehole image can be obtained; when encoding the stratigraphic information in the two-dimensional geological borehole image, it is necessary to split the two-dimensional geological borehole image into more word vectors as much as possible according to the hardware equipment, that is, the pixel block represented by a word vector is smaller, so that the self-attention module in the decoding layer of the Transformer can extract more features in the geological profile.
[0061] The position encoding of the two-dimensional geological borehole image after the block is performed by combining the [0,1] range encoding with the compression encoding method: the relative coordinate encoding of each pixel in the two-dimensional geological borehole image is performed, and then the coordinates of each pixel are normalized to obtain the [0,1] range encoding of the spatial position information of each pixel. Then, the two-dimensional position information after the [0,1] range encoding is subjected to two-dimensional convolution in the same way as the stratigraphic information encoding. The size and step length of the convolution kernel are both the size of the pixel block. The spatial position feature vectors of the two-dimensional geological borehole image after the two-dimensional convolution are combined into a sequence. The length of the sequence is the number of blocks into which the two-dimensional borehole image is divided. The dimension of the word vector in the sequence is the number of pixels in the pixel block after the block division. In this way, the spatial position information encoding sequence of the two-dimensional geological borehole image can be obtained.
[0062] Then a random vector is mapped into random noise with the same length and dimension as the stratigraphic information coding sequence and the position information coding sequence, and the random noise is added to the obtained stratigraphic information coding and spatial position information coding sequences and then input into the Transformer-based decoding layer for decoding.
[0063] In this embodiment, random noise is obtained by linear mapping of a random vector, and inputting it together with the position information coding sequence and the stratigraphic information coding sequence into the Transformer decoding layer can improve the noise resistance of the two-dimensional geological profile generation network and reduce noise. The length of the random vector is the length of the word vector sequence input to the Transformer-based decoding layer, the number of channels is 1 and conforms to the normal distribution; the random vector is input into the fully connected linear mapping layer, and its number of channels is mapped from 1 to the number of channels input into the Transformer-based decoding layer; the stratigraphic information coding sequence, the position information coding sequence and the random noise are added and input into the Transformer decoding layer to fully extract the features in the two-dimensional geological profile.
[0064] The deep learning network model constructed in this embodiment also includes an inverse mapping layer, which is constructed using a fully connected layer and is used to restore the two-dimensional geological features decoded by the Transformer decoding layer to a geological profile; the number of input channels of the fully connected layer is the number of channels of the Transformer decoding layer, and the number of output channels is the product of the total number of pixels in the geological profile and the number of stratum categories divided by the length of the word vector sequence input by the Transformer decoding layer; the result of the fully connected layer is then input into the argmax function to obtain a one-dimensional pixel value of the geological profile; the one-dimensional pixel value of the geological profile is restored to two dimensions to obtain a two-dimensional geological profile.
[0065] like Figure 1 As shown, in step S300, the deep learning network model is trained and tested using the training set and the test set.
[0066] In this embodiment, a multi-weighted cross entropy loss function is constructed, the loss between the model prediction value and the true value is calculated based on the loss function, and the network is trained using the back propagation and stochastic gradient descent methods.
[0067] Specifically, the loss function weight l0 of the drilling area is set to w0, and the loss function l0 of the first-level neighborhood of the drilling area is set to w0. n1 The weight is set to w1, and the loss function of the secondary neighborhood is l n2 The weight is set to w2, and the loss function of the remaining regions is l n3 The weight is set to w3, that is, the loss function is loss = w0l0 + w1l n1 +w2l n2 +w3l n3 ;
[0068] The first-level neighborhood of the borehole is the set of eight-plane neighborhoods of each pixel on the borehole; the second-level neighborhood of the borehole is the set of twenty-five-plane neighborhoods of each pixel on the borehole; the first-level and second-level neighborhoods of the borehole are set according to a preset ratio. In this embodiment, w0 is 0.5, w1 is 0.25, w2 is 0.15, and w3 is 0.1.
[0069] By calculating the loss function between the prediction results and the original two-dimensional geological model, and using the back propagation and stochastic gradient descent methods to train the network, we finally get a deep learning network model for automatic generation of two-dimensional geological profiles.
[0070] like Figure 1 As shown, in step S400, the collected measured borehole image data is processed and input into the trained deep learning network model, and the pixel value results of the one-dimensional geological profile output by the model are restored to obtain a two-dimensional geological profile.
[0071] By inputting the actual collected borehole image data of a specified size into the trained deep learning network model for generation, a two-dimensional geological profile that takes into account global correlation can be obtained.
[0072] The embodiment of the present invention provides a method for generating a two-dimensional geological profile based on deep learning, which constructs a two-dimensional geological profile sample set, randomly extracts local images from the two-dimensional geological profile to obtain virtual borehole images, and constructs a training set and a test set based on the extracted virtual borehole images; constructs a deep learning network model, wherein the deep learning network model includes a coding layer based on a two-dimensional CNN, a decoding layer based on a Transformer, and an inverse mapping layer, and the virtual borehole image is input into the coding layer for stratigraphic information encoding and spatial position information encoding to obtain a stratigraphic information encoding sequence and a spatial position information encoding sequence, and the stratigraphic information encoding sequence, the spatial position information encoding sequence, and random noise are added and input into the decoding layer for decoding; the deep learning network model is trained and tested using the training set and the test set; the collected measured borehole image data is processed and input into the trained deep learning network model, and the pixel value result of the one-dimensional geological profile output by the model is restored to obtain a two-dimensional geological profile. The present invention constructs a coding layer based on a two-dimensional CNN, making it possible to apply the Transformer, which is good at establishing global correlation, in the field of two-dimensional geological profile generation. It also extracts stratigraphic information and location information, constructs a multi-weighted cross entropy loss function to strengthen the training of borehole image data and its neighborhood, and constructs random noise to assist training and denoising, and automatically generates a two-dimensional geological profile that fits a known borehole, thereby improving the quality and refinement of the generated two-dimensional geological profile, thereby enhancing its application value.
[0073] Corresponding to the above-disclosed method for generating a two-dimensional geological profile based on deep learning, the embodiment of the present invention further discloses a system for generating a two-dimensional geological profile based on deep learning, such as Figure 2 As shown, it specifically includes:
[0074] A data set construction module is used to construct a sample set of two-dimensional geological profiles, randomly extract local images from the two-dimensional geological profiles to obtain virtual borehole images, and construct a training set and a test set based on the extracted virtual borehole images;
[0075] A model construction module is used to construct a deep learning network model, wherein the deep learning network model includes a coding layer based on a two-dimensional CNN, a decoding layer based on a Transformer, and an inverse mapping layer, wherein a virtual borehole image is input into the coding layer for stratigraphic information coding and spatial position information coding to obtain a stratigraphic information coding sequence and a spatial position information coding sequence, and the stratigraphic information coding sequence, the spatial position information coding sequence, and random noise are added and input into the decoding layer for decoding;
[0076] A model training module, used to train and test the deep learning network model using the training set and the test set;
[0077] The two-dimensional geological profile generation module is used to process the collected measured drilling image data and input it into the trained deep learning network model, and restore the pixel value results of the one-dimensional geological profile output by the model to obtain a two-dimensional geological profile.
[0078] It should be noted that for the detailed description of a two-dimensional geological profile generation system based on deep learning provided in an embodiment of the present invention, reference can be made to the relevant description of a two-dimensional geological profile generation method based on deep learning provided in an embodiment of the present application, which will not be repeated here.
[0079] In addition, an embodiment of the present invention also provides an electronic device, comprising: a processor and a memory; the memory is used to store one or more program instructions; the processor is used to run one or more program instructions to execute the steps of a two-dimensional geological profile generation method based on deep learning as described in any of the above items.
[0080] It should be noted that, for the detailed description of an electronic device provided in an embodiment of the present invention, reference can be made to the relevant description of a two-dimensional geological profile generation method based on deep learning provided in an embodiment of the present application, which will not be repeated here.
[0081] In addition, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for generating a two-dimensional geological profile based on deep learning as described in any of the above items are implemented.
[0082] It should be noted that, for a detailed description of a computer-readable storage medium provided in an embodiment of the present invention, reference can be made to the relevant description of a two-dimensional geological profile generation method based on deep learning provided in an embodiment of the present application, which will not be repeated here.
[0083] Those skilled in the art will appreciate that all or part of the functions of the various methods in the above-mentioned embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to implement the above-mentioned functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the above-mentioned functions can be implemented. In addition, when all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can also be stored in a storage medium such as a server, another computer, disk, optical disk, flash disk or mobile hard disk, and can be downloaded or copied and saved in the memory of the local device, or the system of the local device is updated, and when the program in the memory is executed by the processor, all or part of the functions in the above-mentioned embodiments can be implemented.
[0084] The above specific examples are used to illustrate the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the art, according to the concept of the present invention, some simple deductions, modifications or substitutions can be made.
Claims
1. A method for generating a two-dimensional geological profile based on deep learning, characterized in that: The method comprises: Constructing a sample set of two-dimensional geological profiles, randomly extracting local images from the two-dimensional geological profiles to obtain virtual borehole images, and constructing a training set and a test set based on the extracted virtual borehole images; Constructing a deep learning network model, the deep learning network model includes a coding layer based on two-dimensional CNN, a decoding layer based on Transformer, and an inverse mapping layer, inputting a virtual borehole image into the coding layer to encode stratum information and spatial position information to obtain a stratum information coding sequence and a spatial position information coding sequence, adding the stratum information coding sequence, the spatial position information coding sequence, and random noise, and then inputting them into the decoding layer for decoding, and inputting the decoded information into the inverse mapping layer to restore it to the pixel layer to obtain a one-dimensional geological profile pixel value result; Using the training set and the test set to train and test the deep learning network model; The collected measured borehole image data is processed and input into the trained deep learning network model, and the pixel value results of the one-dimensional geological profile output by the model are restored to obtain a two-dimensional geological profile.
2. The method for generating a two-dimensional geological profile based on deep learning according to claim 1, characterized in that: A virtual borehole image is obtained by randomly extracting a local image from a two-dimensional geological profile, including: According to all the stratum types appearing in the two-dimensional geological profile sample set, a stratum coding comparison table is prepared, and label coding is performed on each stratum; According to the prepared stratigraphic code comparison table, the attribute value represented by a certain point in the two-dimensional geological section map is set according to the stratigraphic code comparison table; If the original two-dimensional geological profile is in vector form, regular sampling is performed to determine which stratum the sampling point is located above, and the attribute value of the sampling point is set according to the stratum coding comparison table, thereby converting the two-dimensional geological profile in vector form into a raster image.
3. The method for generating a two-dimensional geological profile based on deep learning according to claim 2, characterized in that: A virtual borehole image is obtained by randomly extracting a local image from a two-dimensional geological profile, including: The encoded two-dimensional geological profile is randomly masked, a portion of the image with a preset size is retained as a virtual drilling image, and the rest is assigned a value of 0; The maximum and minimum normalization method is used to normalize the obtained virtual drilling image.
4. The method for generating a two-dimensional geological profile based on deep learning according to claim 1, characterized in that: Inputting the virtual borehole image into the coding layer to perform stratigraphic information coding and spatial position information coding to obtain a stratigraphic information coding sequence and a spatial position information coding sequence, specifically comprising: The input image is divided into blocks by convolution. The size of the two-dimensional convolution kernel is the size of each two-dimensional pixel block. The step size of the two-dimensional convolution is the same as the size of the convolution kernel. Each divided pixel block is mapped to a word vector separately. The feature vectors of the input image after two-dimensional convolution are grouped into a sequence. The length of the sequence is the number of blocks the input image is divided into. The dimension of the word vector in the sequence is the number of pixels in the pixel block after the block division, thus obtaining the stratigraphic information coding sequence.
5. The method for generating a two-dimensional geological profile based on deep learning according to claim 4, characterized in that: Inputting the virtual borehole image into the coding layer to perform stratigraphic information coding and spatial position information coding to obtain a stratigraphic information coding sequence and a spatial position information coding sequence, specifically comprising: The [0,1] range coding combined with compression coding is used to position code the two-dimensional geological drilling images after segmentation, and the relative coordinates of each pixel in the two-dimensional geological drilling images are coded. Then the coordinates of each pixel are normalized to obtain the [0,1] range coding of the spatial position information of each pixel. The two-dimensional position information encoded in the [0,1] range is subjected to two-dimensional convolution. The size and step size of the two-dimensional convolution kernel are both the size of the pixel block. The spatial position feature vectors of the two-dimensional geological borehole image after the two-dimensional convolution are combined into a sequence. The length of the sequence is the number of blocks the two-dimensional borehole image is divided into. The dimension of the word vector in the sequence is the number of pixels in the pixel block after the division, thus obtaining the spatial position information encoding sequence.
6. The method for generating a two-dimensional geological profile based on deep learning according to claim 1, characterized in that: The stratigraphic information coding sequence, the spatial position information coding sequence and the random noise are added and input into the decoding layer for decoding, which specifically includes: The random noise is obtained by linear mapping a random vector. The length of the random vector is the length of the word vector sequence input to the decoding layer. The number of channels is 1 and conforms to the normal distribution. The random vector is input to the linear mapping layer, and the number of channels of the random vector is mapped from 1 to the number of channels of the input decoding layer.
7. The method for generating a two-dimensional geological profile based on deep learning according to claim 1, characterized in that: The deep learning network model is trained and tested using the training set and the test set, specifically including: A multi-weighted cross entropy loss function is constructed, and the loss between the model prediction value and the true value is calculated based on the loss function. The network is trained using backpropagation and stochastic gradient descent methods.
8. The method for generating a two-dimensional geological profile based on deep learning according to claim 7, characterized in that: Construct a multi-weighted cross entropy loss function, and calculate the loss between the model prediction value and the true value based on the loss function, specifically including: The loss function weight l0 of the drilling area is set to w0, and the loss function l0 of the first-level neighborhood of the drilling area is set to w0. n1 The weight is set to w1, and the loss function of the secondary neighborhood is l n2 The weight is set to w2, and the loss function of the remaining regions is l n3 The weight is set to w3, that is, the loss function is loss = w0l0 + w1l n1 +w2l n2 +w3l n3 ; Among them, the first-level neighborhood of the borehole is the set of eight-plane neighborhoods of each pixel on the borehole; the second-level neighborhood of the borehole is the set of twenty-five-plane neighborhoods of each pixel on the borehole; the first-level and second-level neighborhoods of the borehole are set according to a preset ratio.
9. A two-dimensional geological profile generation system based on deep learning, characterized in that: The system comprises: A data set construction module is used to construct a sample set of two-dimensional geological profiles, randomly extract local images from the two-dimensional geological profiles to obtain virtual borehole images, and construct a training set and a test set based on the extracted virtual borehole images; A model building module is used to build a deep learning network model, wherein the deep learning network model includes a coding layer based on a two-dimensional CNN, a decoding layer based on a Transformer, and an inverse mapping layer, wherein a virtual borehole image is input into the coding layer for stratigraphic information coding and spatial position information coding to obtain a stratigraphic information coding sequence and a spatial position information coding sequence, wherein the stratigraphic information coding sequence, the spatial position information coding sequence, and random noise are added and then input into the decoding layer for decoding, and the decoded information is input into the inverse mapping layer for restoration to a pixel layer to obtain a one-dimensional geological profile pixel value result; A model training module, used to train and test the deep learning network model using the training set and the test set; The two-dimensional geological profile generation module is used to process the collected measured drilling image data and input it into the trained deep learning network model, and restore the pixel value results of the one-dimensional geological profile output by the model to obtain a two-dimensional geological profile.
10. An electronic device, characterized in that: The device comprises: a processor and a memory; The memory is used to store one or more program instructions; The processor is used to run one or more program instructions to execute the steps of the method for generating a two-dimensional geological profile based on deep learning as described in any one of claims 1 to 8.
Citation Information
Cited By
Gravity dam intelligent design method based on physical mechanics information generative adversarial network
CN120068242A
Intelligent Design Method of Gravity Dam Based on Physical Mechanics Information Generative Adversarial Network
CN120068242B
