Well logging constrained gravity inversion method and system and storage medium

Abstract

The invention provides a logging constrained gravity inversion method and system and a storage medium, and belongs to the technical field of geophysical exploration. The method comprises the following steps: constructing a multi-modal input layer fusing two-dimensional gravity anomaly and three-dimensional mapping logging data; the method comprises the following steps: designing a UCA-W network architecture comprising a channel-space double attention mechanism (CBAM); establishing a robustness training mechanism based on dynamic well position deviation; a hybrid Dice-MSE loss function optimization strategy is put forward; and developing a multi-density geologic body reconstruction algorithm. Millimeter-level positioning (MAE < = 0.0094 g / cm) of an underground anomalous body boundary is achieved by fusing logging information three-dimensional space coding and gravity data deep features, compared with a traditional UNet inversion method, the structural similarity (SSIM) is improved by 13.6%, the intersection-to-union ratio (IOU) is improved by 11.9%, and the ore body positioning error is smaller than 4.7 m in actual measurement of the Sannicous sulfide ore deposit.

Description

Technical Field

[0001] This invention belongs to the field of geophysical exploration technology, specifically relating to a gravity inversion method, system, and storage medium for well logging constraints. Background Technology

[0002] Gravity exploration, as a fundamental method in geophysical exploration, plays an irreplaceable role in crustal structure analysis, oil and gas resource exploration, and deep mineral exploration. Traditional inversion techniques establish mathematical models based on the physical response relationship between the observation field and subsurface physical parameters, mainly employing two types of methods: linear inversion (such as Tikhonov regularization) and nonlinear inversion (such as the improved conjugate gradient method). Although these algorithms have low dependence on the initial model, they face a dual technical bottleneck when implementing 3D gravity data inversion: on the one hand, they require a large amount of computational resources (a single inversion typically takes several hours to several days); on the other hand, due to the solution characteristics of ill-conditioned equations, the inversion results exhibit excessive smoothing characteristics, making it difficult to accurately restore the boundary morphology of geological bodies.

[0003] In recent years, deep learning technology has provided a new paradigm for overcoming the limitations of traditional inversion methods. Gravity inversion methods based on convolutional neural networks (CNNs) model the mapping from two-dimensional surface gravity anomaly data to three-dimensional subsurface density structure as an end-to-end learning problem. Existing literature reports the use of the UNet architecture to construct the mapping relationship from the two-dimensional gravity field to the depth direction; the implementation of gravity gradient data inversion through improved UNet; and the innovative introduction of random walk algorithms to enhance the generalization ability of the training set. Although these methods have achieved significant improvements in computational efficiency (inference time reduced to the second level), their data-driven nature leads to two key technical defects: ① They are highly dependent on the quality of training data; when there is a deviation between the training set and the actual geological conditions of the work area, the inversion results will produce systematic errors; ② They lack effective physical constraint mechanisms, which easily lead to multiple solutions in deep and complex structural areas.

[0004] Well logging data, as primary data directly characterizing the physical properties of subsurface media, is of great value in improving inversion accuracy. In existing technologies, some researchers have attempted to inject well logging information into the network training process as a one-dimensional vector. However, this approach suffers from dimensionality mismatch: three-dimensional density field reconstruction requires spatial correlation constraints, while one-dimensional well logging data fails to effectively preserve its relative position information in three-dimensional space. This dimensional compression prevents the network from establishing spatial topological relationships between well logging points and surrounding geological bodies, significantly weakening the effectiveness of prior constraints. Summary of the Invention

[0005] The purpose of this invention is to address the aforementioned shortcomings of existing technologies by providing a well-logging-constrained gravity inversion method, system, and storage medium. This method transforms discrete well-logging data into a three-dimensional constraint tensor through spatial location encoding, forming a multimodal input with gravity anomaly data. This processing approach not only strictly preserves the spatial attributes of the well-logging data but, more importantly, achieves a deep fusion of physical constraints and data-driven processes within a deep learning framework. When the well-logging constraint dimension is increased from one-dimensional to three-dimensional, the solution space search range can be significantly reduced, effectively improving the ill-posedness of the inversion problem. Experiments have demonstrated that this technical solution improves the accuracy of the model's representation of deep geological structures, correspondingly reduces the ambiguity of the inversion, and further enhances the reliability and accuracy of the interpretation results.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The primary objective of this invention is to provide a gravity inversion method constrained by well logging, comprising the following specific steps: The simulation dataset is divided into training samples, test samples, and validation samples. The simulation dataset includes surface gravity anomalies. and well logging information ; Construct a UCA-W network model, input training samples for training, and fuse Dice loss. With MSE loss The network parameters are optimized by minimizing the loss function through a weighted combination of parameters, and the performance is evaluated using validation samples. The parameter set with the best performance is selected as the final model parameters. The UCA-W network model consists of a multi-layer downsampling encoder and a transposed convolutional decoder, with CBAM attention units embedded between the transposed convolutional decoders. A random perturbation algorithm is applied to the training samples to generate enhanced training samples, wherein the random perturbation algorithm includes perturbation of surface gravity anomalies. Apply offset to the corresponding well location coordinates And the offset follows a uniform distribution. ,in The standard deviation; The UCA-W network model is trained using the enhanced training samples to obtain a robustly optimized network model. The test samples are input into the robustly optimized network model, which outputs a prediction model. The prediction model is then used for gravity inversion under well logging constraints in the mining area.

[0007] Furthermore, the method for constructing the simulation dataset is as follows: N underground models are first generated using a random walk algorithm. Then calculate the corresponding underground model for each Theoretical surface gravity anomaly Next, surface gravity anomalies were selected. The coordinates of the location corresponding to the maximum gravity anomaly are used as the logging location, and the density information in the depth direction at that location is used as the logging constraint. And map it to three-dimensional space; the underground model As a label, surface gravity anomaly and well logging information As network input.

[0008] Furthermore, the encoder in the UCA-W network model By 3D convolution operator Batch normalization layer and activation function The structure satisfies the mathematical expression:

[0009] This formula describes the encoder's first... The specific calculation process of the layer on the input data involves three operations performed sequentially from the inside out: 3D convolution → batch normalization → activation function; For encoder number Layer input feature tensor, Batch size, representing the number of data samples input into the network at one time; The size of a single sample in three-dimensional space. For the three-dimensional convolution kernel space dimension, These represent the number of input and output channels, Represents 3D convolution. Representing the encoder's first layer.

[0010] Furthermore, the decoder in the UCA-W network model By 3D transpose convolution operator Batch normalization layer and activation function The structure involves fusing the 3D transposed convolution output with the corresponding layer encoder output, followed by batch normalization and activation, satisfying the mathematical expression:

[0011] In the formula For decoder number Layer input feature tensor, Represents 3D convolution. This indicates a channel splicing operation.

[0012] Furthermore, the first The input feature tensor of the layer decoder is derived from the first layer. The output feature vector of the layer decoder With encoder number Layer output feature tensor Assemble by splicing: .

[0013] Furthermore, the CBAM attention module, its channel attention Spatial attention The calculation process is as follows:

[0014]

[0015] In the formula, GAP / GMP represent global average / max pooling, respectively. , The number of channels in the input feature map. This refers to the channel compression ratio. It is 3D spatial convolution kernel, This indicates element-wise multiplication.

[0016] Furthermore, the loss function is:

[0017]

[0018]

[0019] in, and Represent the first and second elements in the prediction model vector and the theoretical model vector, respectively. One element, and Represent the first and second digits of the surface gravity anomaly values ​​obtained from the forward modeling and the actual surface gravity anomaly values, respectively. One element, This is the total loss function.

[0020] Furthermore, a gradient descent-like algorithm is used to minimize the loss function to train the parameters of the UCA-W network model. The trained parameters are as follows: .

[0021] The second objective of this invention is to provide a measurement system for well-logging-constrained gravity inversion, the system being based on the aforementioned method and comprising at least the following components built within the system: The data acquisition module is configured to acquire a simulation dataset, which is divided into training samples, test samples, and validation samples. The simulation dataset includes surface gravity anomalies. and well logging information ; The UCA-W network model building module is configured to take the training samples as input for training and incorporate Dice loss. With MSE loss The network parameters are optimized by minimizing the loss function through a weighted combination of parameters, and the performance is evaluated using validation samples. The parameter set with the best performance is selected as the final model parameters. The UCA-W network model consists of a multi-layer downsampling encoder and a transposed convolutional decoder, with CBAM attention units embedded between the transposed convolutional decoders. The data augmentation module is configured to apply a random perturbation algorithm to the training samples to generate augmented training samples, wherein the random perturbation algorithm includes applying an offset to the well location coordinates. And the offset is evenly distributed. ,in The standard deviation; The training optimization module is configured to train the UCA-W network model using the enhanced training samples to obtain a robustly optimized network model. The prediction module is configured to input the test samples into the robustly optimized network model and output a prediction model, which is used for gravity inversion of well logging constraints in the mining area.

[0022] A third objective of this invention is to provide a computer-readable storage medium storing a program that can be executed by one or more processors to implement the aforementioned gravity inversion method for well logging constraints.

[0023] The English abbreviations used in this invention are explained as follows: UCA-W: Multimodal Feature Fusion Network Model; CBAM: Plug and Play Attention Module; ACC: Average Accuracy; MAE: Mean Absolute Error; IOU: Intersection over Union; SSIM: Structural Similarity.

[0024] Compared with the prior art, the beneficial effects of the technical solution provided by the present invention are as follows: (1) This invention provides a gravity inversion method constrained by well logging. This method constructs a multimodal input layer that integrates two-dimensional gravity anomaly and three-dimensional mapped well logging data; designs a UCA-W network architecture that includes a channel-space dual attention mechanism (CBAM); establishes a robust training mechanism based on dynamic well location offset; proposes a hybrid Dice-MSE loss function optimization strategy; and develops a multi-density resolution reconstruction algorithm for geological bodies. By encoding well logging information in three-dimensional space and fusing it with deep features of gravity data, millimeter-level positioning of the boundary of underground anomalies (MAE≤0.0094g / cm³) is achieved. Compared with the traditional UNet inversion method, the structural similarity (SSIM) is improved by 13.6%, and the intersection-over-union (IOU) ratio is improved by 11.9%. In the actual measurement of the San Nicolás sulfide deposit, the ore body positioning error is less than 4.7 meters.

[0025] (2) The method provided by the present invention makes the inversion results closer to the real situation in detail, and can be implemented based on the PyTorch or Tensorflow framework, which is practical and convenient. Attached Figure Description

[0026] Figure 1 Construct a flowchart for the simulation dataset; Figure 2 This is a schematic diagram of the UCA-W structure constructed according to the present invention; Figure 3 For CBAM attention module; Figure 4 The figures show the inversion results of the present invention on simulation data. In the figures: (a) actual underground density distribution; (b) underground density distribution predicted by the UCA-W network; (c) underground density distribution predicted by the UNet-W network; (d) underground density distribution predicted by the SSGI-W network; (e) underground density distribution predicted by the UCA network. Figure 5 The robust inversion results of well logging position fluctuations in this embodiment of the invention are shown in the figure: (a) actual subsurface density distribution; (b1) predicted subsurface density distribution after offsetting the well logging position; (b2) well logging information after offsetting the well logging position; (c1) predicted subsurface density distribution without offsetting the well logging position; (c2) well logging information without offsetting the well logging position. Figure 6 The figure shows the results of the multi-density modality adaptive inversion in an embodiment of the present invention. In the figure: (a) actual underground density distribution; (b) predicted underground density distribution. Figure 7 The results of the application in the real-world mining area of ​​St. Nicholas are shown in the figure: (a1) The UCA-W network in... The vertical profile of the (a2) UCA-W network in (b1) Vertical profile of UNet-W network in The vertical profile of the UNet-W network, (b2) The vertical cross-section shows the actual outline of the ore body, with the black line representing the ore body. Figure 8 A schematic diagram of the system structure provided by the present invention is shown. Figure 9 A block diagram of an electronic device suitable for implementing an information acquisition method according to an embodiment of the present invention is shown schematically. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the specific embodiments and accompanying drawings are described in further detail below. Where specific techniques or conditions are not specified in the embodiments, they are performed in accordance with the techniques or conditions described in the literature in this field or according to the product manual.

[0028] This invention provides a gravity inversion method with well logging constraints, comprising the following steps: Step 1: Construct 22,600 simulation datasets The training set consists of 20,000 samples, the test set consists of 2,000 samples, and the validation set consists of 600 samples. The specific process is as follows: 22,600 underground models were first generated using a random walk algorithm. Each size is Then calculate the corresponding underground model for each Theoretical surface gravity anomaly Size is Next, surface gravity anomalies were selected. The coordinates of the location corresponding to the maximum gravity anomaly are used as the logging location, and the density information in the depth direction at that location is used as the logging constraint. And map it to three-dimensional space, size Underground model As a label, surface gravity anomaly and well logging information As network input.

[0029] Step 2: Construct a multimodal feature fusion network model UCA-W, which consists of a 4-layer downsampling encoder and a 4-layer transposed convolutional decoder, and embeds CBAM attention units in the downsampling; The UCA-W network model is constructed as follows: No. l Layer encoder By 3D convolution operator Batch normalization layer and activation function The structure satisfies the mathematical expression:

[0030] This formula describes the encoder's first... The specific calculation process of the layer on the input data involves three operations performed sequentially from the inside out: 3D convolution → batch normalization → activation function. For encoder number Layer input feature tensor, The batch size represents the number of data samples input into the network at one time. The size of a single sample in three-dimensional space. For the three-dimensional convolution kernel space dimension, These represent the number of input and output channels, Represents 3D convolution. Representing the encoder's first layer.

[0031] No. l Layer decoder By 3D transpose convolution operator Batch normalization layer and activation function The structure involves fusing the 3D transposed convolution output with the corresponding layer encoder output, followed by batch normalization and activation, satisfying the mathematical expression:

[0032] In the formula For decoder number Layer input feature tensor, This indicates a channel splicing operation.

[0033] During the upsampling process of the UCA-W network, each decoder layer... Not only does it use the transposed convolution operator for feature reconstruction, but it also utilizes the corresponding downsampling layer information in the encoder to achieve multi-scale feature fusion through skip connections. Specifically, the... The input feature tensor of the layer decoder is derived from the first layer. The output feature vector of the layer decoder With encoder number Layer output feature tensor Assemble by splicing: .

[0034] Furthermore, a CBAM attention module is introduced in layers 1-4, with its channel attention... Spatial attention The calculation process is as follows:

[0035]

[0036] In the formula, GAP / GMP represent global average / max pooling, respectively. , The number of channels in the input feature map. This represents the channel compression ratio. It is 3D spatial convolution kernel, This indicates element-wise multiplication.

[0037] Step 3: Construct a hybrid loss optimizer that incorporates Dice loss. With MSE loss weighted combination ; Step 4: Input the training samples from Step 1 into the UCA-W network constructed in Step 2 for training, and calculate the loss function from Step 3. Train the network model parameters by minimizing the loss function, and then verify the performance of the trained network parameters on the validation dataset. Select the parameters with the highest performance as the optimized UCA-W parameters. loss function for:

[0038]

[0039]

[0040] in, and Represent the first and second elements in the prediction model vector and the theoretical model vector, respectively. One element, and Represent the first and second digits of the surface gravity anomaly values ​​obtained from the forward modeling and the actual surface gravity anomaly values, respectively. Each element.

[0041] Batch training is used when training UCA-W, meaning that a batch of sample pairs is input into UCA-W in each iteration, and a gradient descent-like algorithm is used to minimize the loss function to train the UCA-W parameters. The trained parameters are: .

[0042] In some implementations, gradient descent algorithms include, but are not limited to, batch gradient descent, stochastic gradient descent, and the Adam algorithm.

[0043] On the validation set, mean accuracy (ACC), mean absolute error (MAE), intersection-over-union ratio (IOU), and structural similarity (SSIM) were used for evaluation.

[0044]

[0045]

[0046]

[0047]

[0048] in, and These represent the prediction model vector and the theoretical model vector, respectively. and Represent the first and second elements in the prediction model vector and the theoretical model vector, respectively. One element, Representing vectors Dimensions. Indicates the threshold. The value is selected as 0.01, when and The absolute value of the error is less than the threshold. If so, the prediction is considered correct and used. The accuracy value is calculated by dividing the number of correct predictions by the total number of anomaly meshes. MAE (Mean Absolute Error) reflects the average error between the reconstructed model and the theoretical model. IOU (Intersection over Union) indicates the overlap between the predicted and actual boundaries of the subsurface region; a higher IOU value indicates better alignment between the predicted and actual regions, reflecting a more accurate model. SSIM (Structural Similarity Score) represents the number of correctly predicted predictions. and respectively and The mean, and respectively and standard deviation for and covariance, and SSIM is a constant. The value range of SSIM is [-1,1], which can effectively evaluate the similarity between the reconstructed subsurface model and the real subsurface model, especially for the details and edge features of geological structures.

[0049] Step 5: Generate well location coordinate offset using a random perturbation algorithm. To enhance the robustness of the network to well location fluctuations, a random perturbation algorithm is used to generate well location coordinate offsets in practical applications of well logging. This is because the complex terrain and construction conditions can lead to offset errors in well logging coordinates. Let the original well logging location coordinates be... Its corresponding gravity anomaly value is ; Generate offsets that follow a uniform distribution. ,in The standard deviation is the preset value. The offset logging position The corresponding density information serves as a new logging constraint. Input the data into the UCA-W network for training; The network is adapted through iterative training. The coordinate disturbance within the range enhances the fault tolerance capability for the uncertainty of the actual well logging position.

[0050] Step 6: Input the test samples into the robustly optimized network model and output the prediction model; this also verifies the adaptability of the UCA-W network in handling multi-density geological structures.

[0051] The process of applying the predictive model to a real mining area is as follows: Data preprocessing: Kriging interpolation was performed on the Bouguer gravity anomaly data of the mining area to generate 32×32 gridded data. ); Log logging constraints were generated based on the density-depth curve of well SAL-25. ), and mapped to a 16×32×32 three-dimensional tensor; Model fine-tuning: Based on the geological characteristics of the mining area, a dedicated dataset (including the main ore body with a residual density of 1.1 g / cm³ and the edge gradient zone with a density of 0.8~1.1 g / cm³) was constructed, and the UCA-W network was trained by transfer learning. Inversion execution: will ( Input the finely tuned UCA-W network and output a three-dimensional density model. ); Results verification: Compare the predicted ore body outline with the actual location of the sulfide ore body revealed by the borehole (e.g., Northing=-400m profile).

[0052] Example 1 This invention provides a gravity inversion method with well logging constraints, comprising the following steps: Step 1: Construct the simulation dataset: like Figure 1 As shown, a three-dimensional underground density model is first generated using a random walk algorithm. ; The theoretical surface gravity anomaly corresponding to the calculation model ; Locating the coordinates of the maximum gravity anomaly As a virtual logging location; Extract the vertical density at this location and map it to a 3D logging constraint. ; Building simulation sets .

[0053] Step 2: Construct the UCA-W network architecture: like Figure 2 As shown, the overall architecture of the UCA-W network is U-shaped and consists of three main parts: Encoder: Gradually extracts input data (surface gravity anomaly) through a series of convolution and downsampling operations. and well logging information It can improve the high-dimensional features of ) while reducing spatial resolution.

[0054] Decoder: Through a series of upsampling and convolution operations, the high-dimensional feature map is gradually restored to its original resolution to generate the final output.

[0055] Skip Connections: The diagram shows connections from the encoder ( , , , Laterally connected to the decoder ( , , , The arrow () represents the low-level spatial features from different levels in the encoder, which directly pass them to the decoder, helping the decoder to reconstruct the output more accurately and solving the problem of spatial information loss in deep networks.

[0056] Before entering the backbone network, the raw data (surface gravity anomaly) and well logging information First, the embedding layer transforms the heterogeneous input data into a unified, high-dimensional feature representation. This allows convolutional networks to process data effectively.

[0057] An encoder-decoder structure is adopted, as follows: The encoder consists of four downsampling modules, each containing two consecutive Conv 3×3 + BN + ELU layers. The Conv 3×3 is a 3D convolutional kernel used to extract spatial features; BN (Batch Normalization) is used to accelerate training convergence and improve model stability; and ELU (Exponential Linear Unit) is the activation function to alleviate the vanishing gradient problem. After each convolutional stage, an attention mechanism module CBAM (Convolutional Block Attention Model) is inserted. At the end of each stage, downsampling reduces the spatial resolution of the feature map while increasing the number of feature channels, preparing for the next stage.

[0058] The last layer of the encoder ( After that, the bottom of the "U"-shaped connection network is connected to two Conv 3×3+BN+ELU convolutional layers to process the highest level semantic features at the lowest spatial resolution, and then upsampling begins.

[0059] The decoder consists of four upsampling stages, symmetrical to the encoder. Each stage first increases the resolution of the feature maps through upsampling. The upsampled feature maps are then skipped between the upsampled feature maps and the corresponding feature maps from the encoder. Example: In the second stage of the decoder, the upsampling input has two different "information sources," one of which is... The dimensions are 4×4×512, one is The feature map has dimensions of 4×4×512. Both are transposed convolutions to restore the spatial dimensions of the feature map to 8×8×256. After being concatenated by channels, they are then compared with the encoder. (8×8×256) skip connections are made. This allows for the utilization of abstract semantic information from deeper layers ( , ), and can also utilize fine spatial information from shallow layers ( This allows for accurate localization and reconstruction; the stitched feature maps will then pass through two Conv 3×3+BN+ELU convolutional layers, and after further fusion, they will serve as the input for the next decoder step.

[0060] The last layer ( ) after the last upsampling ( ) and jump connection (with After concatenation, the data is processed through two Conv 3×3 + BN + ELU convolutional layers and one Conv 1×1 + Sigmoid layer to obtain a high-resolution feature map. Then, the 3D Refinder module outputs a three-dimensional density volume.

[0061] Step 3: Design the hybrid loss function: (Structural similarity loss) (Gravity data fitting loss) Total loss function:

[0062] Step 4: Network Training and Validation The Adam optimizer was used with an initial learning rate of 0.0001 and a batch size of 32. The training process includes an early stopping strategy to prevent overfitting. Monitor four metrics on the validation set:

[0063]

[0064]

[0065]

[0066] Figure 4 This presentation compares the inversion results of four different networks on two density volumes using stochastic models. The density volume on the left is a shallow layer, and the density volume on the right is a deep layer. For the shallow density volume on the left, the UCA-W network, UNet-W network, and UCA network can all invert the density value and edge information of the density volume relatively well. However, for the deep density volume on the right, it can be seen that the UCA-W network can invert the density volume in the depth direction better than the UNet-W network and UCA network, without too many false anomalies.

[0067] Step 5: Well Logging Location Robustness Training: For the original well logging coordinates Apply random perturbation: ,in The standard deviation is the preset value. Using the perturbated coordinates Extracting new logging constraints ; Figure 5 Displays the inversion results after well logging location shift. Even if the well logging location changes, it can still be used. It can guide gravity inversion and effectively invert underground density distribution.

[0068] Step 6: Multi-density model fitness verification: First, a new simulation dataset is constructed based on step 1, generating a composite model composed of volumes with different residual densities (1.0 g / cm³ and 0.6 g / cm³). ), calculate its theoretical gravity anomaly ( ) and logging constraints ( ); Will( Input the UCA-W network and output the prediction model. ; Figure 6 The results show the inversion of multiple densities, which can not only distinguish blocks of different densities, but also invert density shape and density value in a good way.

[0069] Step 7, Practical Application Process: (1) Input processing: processing measured gravity data Perform Kriging meshing (32×32); generate based on well logging curves (16×32×32); (2) Network fine-tuning: Construct a target mining area-specific dataset (including typical density structure); the UCA-W network structure remains unchanged, and the network parameters are fine-tuned using the specific dataset to adapt to the real mining area; (3) Inversion execution: Input Obtain a 3D density model ( (4) Result verification: Figure 7 The application effect in the San Nicholas mining area is shown: It can be seen that the UCA-W network proposed in this invention is more focused in its predicted distribution compared with the traditional Unet network. The location information of the predicted results is basically consistent with the drilling information. The shape of the anomaly source and the physical boundary inverted by the prediction model are also in good agreement with the actual situation.

[0070] Example 2 This embodiment, based on the design foundation of Embodiment 1, discloses a measurement system for well-logging constrained gravity inversion, such as... Figure 8 As shown, this system, based on the aforementioned detection method, includes at least the following components built into the system: The data acquisition module is configured to acquire a simulation dataset, which is divided into training samples, test samples, and validation samples. The simulation dataset includes surface gravity anomaly d and well logging information L. The UCA-W network model building module is configured to input the training samples for training, optimize the network parameters by minimizing the loss function, evaluate the performance using validation samples, and select the parameter set with the best performance as the final model parameters; the UCA-W network model consists of a multi-layer downsampling encoder and a transposed convolutional decoder, and CBAM attention units are embedded between the transposed convolutional decoders. The data augmentation module is configured to apply a random perturbation algorithm to the training samples to generate augmented training samples, wherein the random perturbation algorithm includes applying offsets Δx and Δy to the well location coordinates, and the offsets are uniformly distributed U(-3σ, 3σ), where σ is the standard deviation. The training optimization module is configured to train the UCA-W network model using the enhanced training samples to obtain a robustly optimized network model. The prediction module is configured to input the test samples into the robustly optimized network model and output a prediction model, which is used for gravity inversion of well logging constraints in the mining area.

[0071] Example 3 Based on the design of embodiments 1 and 2, this embodiment discloses a computer-readable storage medium storing a program that can be executed by one or more processors to implement the above-mentioned well-logging constrained gravity inversion method.

[0072] Figure 9 A block diagram schematically illustrates an electronic device suitable for implementing an information acquisition method according to an embodiment of the present disclosure.

[0073] like Figure 9 As shown, an electronic device 900 according to an embodiment of the present disclosure includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage portion 908 into a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0074] RAM 903 stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via bus 904. Processor 901 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 902 and / or RAM 903. It should be noted that programs may also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in one or more memories.

[0075] According to embodiments of this disclosure, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to a bus 904. The electronic device 900 may also include one or more of the following components connected to the I / O interface 905: an input section 906 including a keyboard, mouse, etc.; an output section 907 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 908 including a hard disk, etc.; and a communication section 909 including a network interface card such as a LAN card, modem, etc. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as needed. A removable medium 911, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 910 as needed so that computer programs read from it can be installed into the storage section 908 as needed.

[0076] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0077] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 902 and / or RAM 903 and / or one or more memories other than ROM 902 and RAM 903 described above.

[0078] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code enables the computer system to implement the information acquisition method provided in the embodiments of this disclosure.

[0079] When the computer program is executed by the processor 901, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0080] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 909, and / or installed from a removable medium 911. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0081] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0082] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on a user's computing device, partially on a user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0083] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0084] Where there is no conflict, the above embodiments and features described herein can be combined with each other.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A gravity inversion method constrained by well logging, characterized in that, The specific steps include the following: S1. Obtain the simulation dataset, which is divided into training samples, test samples, and validation samples. The simulation dataset includes surface gravity anomalies. and well logging information ; S2. Construct a UCA-W network model, input the training samples for training, and fuse the Dice loss. With MSE loss The weighted combination of parameters optimizes the network parameters by minimizing the loss function, and uses validation samples to evaluate the reconstruction performance. The parameter set with the best performance is selected as the final model parameters. The UCA-W network model consists of a multi-layer downsampling encoder and a transposed convolutional decoder, with CBAM attention units embedded between the downsampling encoders; S3. Apply a random perturbation algorithm to the training samples to generate enhanced training samples, wherein the random perturbation algorithm includes perturbation of surface gravity anomalies. Apply offset to the corresponding well location coordinates And the offset follows a uniform distribution. ,in The standard deviation; S4. Train the UCA-W network model using the enhanced training samples to obtain a robustly optimized network model; S5. Input the test sample into the robustly optimized network model and output the prediction model. The prediction model is used for gravity inversion of well logging constraints in the mining area.

2. The well-logging-constrained gravity inversion method according to claim 1, characterized in that, The method for constructing the simulation dataset is as follows: N underground models are first generated using a random walk algorithm. ; Then calculate the corresponding underground model for each Theoretical surface gravity anomaly Next, surface gravity anomalies were selected. The coordinates of the location corresponding to the maximum gravity anomaly are used as the logging location, and the density information in the depth direction at that location is used as the logging constraint. And map it to three-dimensional space; the underground model As a label, surface gravity anomaly and well logging information As network input.

3. The well-logging-constrained gravity inversion method according to claim 1, characterized in that, The encoder in the UCA-W network model By 3D convolution operator Batch normalization layer and activation function The structure satisfies the mathematical expression: In the formula For encoder number Layer input feature tensor, Batch size represents the number of data samples input into the network at one time. The size of a single sample in three-dimensional space. For the three-dimensional convolution kernel space dimension, These represent the number of input and output channels, Represents 3D convolution. Representing the encoder's first layer.

4. The logging-constrained gravity inversion method according to claim 3, characterized in that, Decoder in the UCA-W network model By 3D transpose convolution operator Batch normalization layer and activation function The structure involves fusing the 3D transposed convolution output with the corresponding layer encoder output, followed by batch normalization and activation, satisfying the mathematical expression: In the formula For the decoder Layer input feature tensor, Represents 3D convolution. This indicates a channel splicing operation.

5. The well-logging-constrained gravity inversion method according to claim 4, characterized in that, No. The input feature tensor of the layer decoder is derived from the first layer. The output feature vector of the layer decoder With encoder number Layer output feature tensor Assemble by splicing: 。 6. The logging-constrained gravity inversion method according to claim 5, characterized in that, CBAM attention module, its channel attention Spatial attention The calculation process is as follows: In the formula, GAP / GMP represent global average / max pooling, respectively. , The number of channels in the input feature map. This refers to the channel compression ratio. It is 3D spatial convolution kernel, This indicates element-wise multiplication.

7. The gravity inversion method with logging constraints according to claim 1, characterized in that, The loss function is: in, and Represent the first and second elements in the prediction model vector and the theoretical model vector, respectively. One element, and Represent the first and second digits of the surface gravity anomaly values ​​obtained from the forward modeling and the actual surface gravity anomaly values, respectively. One element, This is the total loss function.

8. The logging-constrained gravity inversion method according to claim 7, characterized in that, The loss function is minimized using a gradient descent-like algorithm to train the parameters of the UCA-W network model.

9. A detection system for a well-logging-constrained gravity inversion method, characterized in that, The system, based on the method of any one of claims 1-8, includes at least the following components built within the system: The data acquisition module is configured to acquire a simulation dataset, which is divided into training samples, test samples, and validation samples. The simulation dataset includes surface gravity anomalies. and well logging information ; The UCA-W network model building module is configured to take the training samples as input for training and incorporate Dice loss. With MSE loss The weighted combination of parameters optimizes the network parameters by minimizing the loss function, and the performance is evaluated using validation samples. The parameter set with the best performance is selected as the final model parameters. The UCA-W network model consists of a multi-layer downsampling encoder and a transposed convolutional decoder, with CBAM attention units embedded between the transposed convolutional decoders; A data augmentation module is configured to apply a random perturbation algorithm to the training samples to generate augmented training samples, wherein the random perturbation algorithm includes perturbation of surface gravity anomalies. Apply offset to the corresponding well location coordinates And the offset is evenly distributed. ,in The standard deviation; The training optimization module is configured to train the UCA-W network model using the enhanced training samples to obtain a robustly optimized network model. The prediction module is configured to input the test samples into the robustly optimized network model and output a prediction model, which is used for gravity inversion of well logging constraints in the mining area.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that can be executed by one or more processors to implement the gravity inversion method for well logging constraints as described in any one of claims 1-8.

Images