Three-dimensional gravity inversion method based on closed-loop network
By adopting a closed-loop network self-supervised learning method in gravity inversion, combining universal gravitational formulas and prior knowledge, the traditional method solves the problems of speed and accuracy and multi-solvency, and achieves efficient and accurate three-dimensional density data inversion. Overcome the limitation of insufficient training data.
Patent Information
- Application Number
- CN202210250081.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-14
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2042-03-14
AI Technical Summary
The existing gravity inversion technology has challenges in taking into account both speed and accuracy. The traditional method solves fast but has low accuracy, while the machine learning algorithm has high accuracy but slow solution speed. Multi-solution problems and lack of training data limit deep learning on Earth. Applications in the field of physics.
Using a three-dimensional gravity inversion method based on a closed-loop network, a closed-loop structure is formed by building a forward network and an inversion network, and a closed-loop self-supervision learning is carried out, and a universal gravitational formula and prior knowledge constrain the inversion process is used to achieve self-supervision that does not rely on training data. study.
High-precision three-dimensional density data inversion is achieved, which avoids the risk of overfitting and does not require additional training data, improving the efficiency and accuracy of gravity inversion.
Smart Images

Figure CN114662302B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of geophysical inversion imaging technology, and in particular to a three-dimensional gravity inversion method based on a closed-loop network. Background Art
[0002] Gravity inversion is an important, non-destructive geophysical exploration method. Three-dimensional gravity inversion is the solution from two-dimensional gravity field observation data to three-dimensional density data. It is an ill-posed problem with a non-unique solution space. Currently, the more mature gravity inversion algorithms include traditional methods such as Newton's method, conjugate gradient method, least squares method, and machine learning algorithms such as simulated annealing algorithm, particle swarm algorithm, ant colony algorithm, and genetic algorithm. Traditional algorithms have fast solution speed and low accuracy, while machine learning algorithms have high accuracy but slow solution speed. Balancing speed and accuracy is an important problem facing gravity inversion. Another problem that gravity inversion needs to solve is the problem of multiple solutions, which requires full use of prior knowledge from domain experts or related data.
[0003] In recent years, deep learning methods have developed rapidly, and a large number of deep learning algorithms have successfully promoted the solution of image-related pathological problems such as super-resolution and deblurring. Treating the inverted matrix data as an image can improve the quality of the inversion results from the perspective of image super-resolution. A more direct method is to directly use a deep learning algorithm to implement gravity inversion, and the solution can be quickly obtained during the reasoning process. However, this method relies on high-quality training data and cannot adjust the inversion results based on prior knowledge.
[0004] The lack of training data is the main obstacle to the application of deep learning technology in fields such as geophysics and medical imaging. The current main solution is to synthesize training data based on expert knowledge and apply it to actual data after training. This method can alleviate the data problem to a certain extent, but it is affected by the difference between synthetic data and actual data, and its effect is limited. Summary of the invention
[0005] The present application provides a three-dimensional gravity inversion method based on a closed-loop network to solve the problems existing in the prior art.
[0006] In a first aspect, the present application provides a three-dimensional gravity inversion method based on a closed-loop network, comprising:
[0007] Build a forward modeling network and an inversion network. The inversion network and the forward modeling network form a closed-loop structure.
[0008] The gravity inversion results are obtained by performing closed-loop self-supervised learning through the forward network and the inversion network.
[0009] In some embodiments, the process of building a forward modeling network includes:
[0010] According to the universal gravitation formula, determine the gravitational force exerted on the surface observation point by the underground area;
[0011] Determine the component of gravity acting on the observation point on the surface in the direction perpendicular to the surface;
[0012] According to the component forces corresponding to multiple surface observation points, the forward solution is obtained.
[0013] In some embodiments, according to the universal gravitation formula, the gravitational force exerted on the surface observation point by the underground area is calculated by the following formula:
[0014]
[0015] Among them, G is the gravitational coefficient, m1 is the mass of the observation point on the surface, and m i is the mass of the ith underground region, r is the distance from the surface observation point to the ith underground region, F is the gravitational force between the surface observation point and the ith underground region, and n is the number of underground regions;
[0016] The component of gravity acting on the surface observation point in the direction perpendicular to the surface is calculated using the following formula:
[0017]
[0018] Among them, F y is the component of gravity acting on the observation point on the surface in the direction perpendicular to the surface, h i is the distance from the ith underground area to the surface;
[0019] According to the component forces corresponding to multiple surface observation points, the forward solution is calculated by the following formula:
[0020]
[0021] Among them, d is the one-dimensional vector pulled out by the three-dimensional density model according to the position relationship, A is the forward modeling matrix, and g is the forward modeling solution result.
[0022] In some embodiments, the process of building an inversion network includes:
[0023] An inversion network is constructed using an encoder, an expander, a decoder and a three-dimensional refiner, and an inversion solution result is obtained through the inversion network;
[0024] Wherein, the encoder is used to obtain a first characteristic map according to the gravity field data;
[0025] The expander is used to obtain a second characteristic map according to the energy distribution curve;
[0026] The decoder is used to perform feature fusion on the first feature map and the second feature map to obtain a fused feature map;
[0027] The three-dimensional refiner is used to obtain three-dimensional density data according to the fused feature map, and the inversion solution result includes the three-dimensional density data.
[0028] In some embodiments, the encoder is specifically used to extract features from gravity field data to obtain a plurality of first feature maps of different sizes.
[0029] In some embodiments, the expander is specifically used to expand the energy distribution curve into a plurality of second characteristic graphs, the number of the second characteristic graphs is the same as the number of the first characteristic graphs, and the sizes thereof are correspondingly the same.
[0030] In some embodiments, the decoder is specifically used to perform feature fusion on a first feature map and a second feature map of the same size to obtain feature fusion results of multiple different sizes, and perform feature fusion on multiple feature fusion results of different sizes to obtain a fused feature map.
[0031] In some embodiments, the three-dimensional refiner is specifically used to convert the fused feature map into a three-dimensional feature map, and obtain three-dimensional density data based on the three-dimensional feature map.
[0032] In some embodiments, closed-loop self-supervised learning is performed through a forward network and an inversion network to obtain a gravity inversion result, including:
[0033] According to the original gravity field data and energy distribution curve, the three-dimensional density data is obtained by inversion through the inversion network;
[0034] According to the three-dimensional density data, forward modeling is performed through the forward modeling network to obtain the reconstructed gravity field data;
[0035] Network parameter optimization training is performed based on the original gravity field data and the reconstructed gravity field data.
[0036] In some embodiments, the inversion and forward modeling are solved by the following formula:
[0037]
[0038]
[0039] Among them, g is the original gravity field data, is the inversion solution process, d′ is the three-dimensional density data obtained by inversion solution, is the forward solution process, g′ is the reconstructed gravity field data obtained by the forward solution, ‖d′‖1 and TV(d′) are the regularization constraints of the inversion solution results, φ is the mapping function from the one-dimensional vector d pulled from the position relationship of the three-dimensional density model to the energy distribution curve l, and λ1, λ2, λ3, and λ4 are preset parameters.
[0040] The present application provides a three-dimensional gravity inversion method based on a closed-loop network, the method comprising: building a forward network and an inversion network, the inversion network and the forward network forming a closed-loop structure; performing closed-loop self-supervised learning through the forward network and the inversion network to obtain a gravity inversion result. The method adopts a closed-loop learning method, and realizes self-supervised learning of the inversion process with the help of a forward process with known parameters and a unique solution space. At the same time, the inversion process is constrained by prior knowledge from expert knowledge or observation data. Compared with the existing gravity inversion algorithm based on deep learning, the method of this embodiment has the beneficial effect of not relying on training data and without the risk of overfitting. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0042] Figure 1 A schematic diagram of a three-dimensional gravity inversion method based on a closed-loop network provided in an embodiment of the present application;
[0043] Figure 2 A schematic diagram of closed-loop self-supervised learning in an embodiment of the present application;
[0044] Figure 3 Schematic diagram of the inversion solution process in the embodiment of the present application;
[0045] Figure 4 Schematic diagram of an encoder in an embodiment of the present application;
[0046] Figure 5 A schematic diagram of a decoder in an embodiment of the present application;
[0047] Figure 6 Schematic diagram of an attention fusion module in an embodiment of the present application;
[0048] Figure 7 Schematic diagram of a cross-scale feature fusion module in an embodiment of the present application;
[0049] Figure 8 is a schematic diagram of the original gravity field data in this embodiment;
[0050] Fig. 9 Schematic diagram of three-dimensional density data (inversion solution result) obtained by the inversion network in this embodiment;
[0051] Fig.10 Schematic diagram of reconstructed gravity field data (forward modeling result) obtained by processing three-dimensional density data (inversion solution result) through a forward modeling network in this embodiment;
[0052] Fig.11is a schematic diagram of a longitudinal section of the inversion solution result in this embodiment;
[0053] Fig.12 A schematic diagram of the structure of a terminal device provided in an embodiment of the present application.
[0054] The above drawings show clear embodiments of the present disclosure, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present disclosure in any way, but to illustrate the concepts of the present disclosure to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0056] Gravity inversion is an important, non-destructive geophysical exploration method. Due to the gravitational force, objects on the surface are subject to the gravitational force of the earth (i.e., gravity). The uneven distribution of materials with different densities underground will cause anomalies in the surface gravity field. Gravity inversion estimates the density distribution of underground materials through the gravity field information detected on the surface. Gravity inversion is widely used in many research fields such as geological structure research, mineral exploration, and engineering environment research.
[0057] Three-dimensional gravity inversion is the solution from two-dimensional gravity field observation data to three-dimensional density data. It is an ill-posed problem with a non-unique solution space. Currently, the more mature gravity inversion algorithms include traditional methods such as Newton's method, conjugate gradient method, least squares method, and machine learning algorithms such as simulated annealing algorithm, particle swarm algorithm, ant colony algorithm, and genetic algorithm. Traditional algorithms have fast solution speed and low accuracy, while machine learning algorithms have high accuracy but slow solution speed. Balancing speed and accuracy is an important problem facing gravity inversion. Another problem that gravity inversion needs to solve is the problem of multiple solutions, which requires full use of prior knowledge from domain experts or related data.
[0058] In recent years, deep learning methods have developed rapidly, and a large number of deep learning algorithms have successfully promoted the solution of image-related pathological problems such as super-resolution and deblurring. Treating the inverted matrix data as an image can improve the quality of the inversion results from the perspective of image super-resolution. A more direct method is to directly use a deep learning algorithm to implement gravity inversion, and the solution can be quickly obtained during the reasoning process. However, this method relies on high-quality training data and cannot adjust the inversion results based on prior knowledge.
[0059] The lack of training data is the main obstacle to the application of deep learning technology in fields such as geophysics and medical imaging. The current main solution is to synthesize training data based on expert knowledge and apply it to actual data after training. This method can alleviate the data problem to a certain extent, but it is affected by the difference between synthetic data and actual data, and its effect is limited.
[0060] The three-dimensional gravity inversion method based on a closed-loop network provided in this application is intended to solve the above technical problems of the prior art.
[0061] The present application uses a self-supervised algorithm based on a closed-loop network to realize three-dimensional gravity inversion. The closed-loop network includes a forward modeling network and an inversion network. The forward modeling problem solution space is unique. The present application constrains the inversion process through a known forward modeling process. The present application builds a neural network to realize the inversion process and the forward modeling process respectively, and constrains the inversion process through an inversion-forward closed-loop process. In addition, the closed-loop network proposed in the present application is a self-supervised algorithm and does not require additional training data.
[0062] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.
[0063] In some embodiments, a three-dimensional gravity inversion method based on a closed-loop network is provided.
[0064] Figure 1 A schematic diagram of a three-dimensional gravity inversion method based on a closed-loop network provided in an embodiment of the present application, such as Figure 1 As shown, the method mainly includes the following steps:
[0065] S100, build a forward modeling network and an inversion network, and the inversion network and the forward modeling network form a closed-loop structure;
[0066] This step mainly includes: (1) building a neural network to realize the forward calculation process, and calculating the parameters of each neuron in the network through the universal gravitation formula; (2) building a neural network to realize the inverse solution process. The inverse neural network includes an encoder, an expander and a decoder. Under the constraints of prior knowledge, it realizes the mapping from gravity field data (two-dimensional) to density data (three-dimensional).
[0067] Among them, the inversion network is used to perform an inversion solution process based on the original gravity field data to obtain the inversion solution result, that is, three-dimensional density data; the forward network is used to perform a forward solution based on the three-dimensional density data obtained by inversion to obtain the reconstructed gravity field data.
[0068] S200, performing closed-loop self-supervised learning through the forward network and the inversion network to obtain the gravity inversion result.
[0069] Figure 2 Schematic diagram of closed-loop self-supervised learning in the embodiment of the present application, such as Figure 2 As shown in the figure, after the forward network and the inversion network are built, the inversion and forward solution are iteratively performed on the given data to be solved. The mean absolute error (MAE) between the reconstructed gravity field data and the original gravity field data obtained by closed-loop solution, the deviation between the inverted three-dimensional density data and the prior knowledge, and the regularization constraint of the density data are used as the loss function. The Adam optimizer is used to train the parameters of the inversion model, and a decaying learning rate is used during the training process.
[0070] The present embodiment provides a three-dimensional gravity inversion method based on a closed-loop network. The method adopts a closed-loop learning method and realizes self-supervised learning of the inversion process by means of a forward process with known parameters and a unique solution space. At the same time, the inversion process is constrained by using prior knowledge from expert knowledge or observation data. Compared with the existing gravity inversion algorithm based on deep learning, the method of the present embodiment has the beneficial effect of not relying on training data and having no risk of overfitting.
[0071] In some embodiments, the process of building a forward modeling network includes: determining the gravitational force exerted on the surface observation point by the underground area according to the universal gravitation formula; determining the component of the gravitational force exerted on the surface observation point in the direction perpendicular to the surface; and obtaining the forward modeling solution based on the component forces corresponding to multiple surface observation points.
[0072] Specifically, the forward neural network implements matrix multiplication through neurons to achieve the forward process. The basis of the forward process is the universal gravitational formula (1):
[0073]
[0074] Where G is the gravitational coefficient, m1 and m2 are the masses of the two objects, and r and F are the physical distance and gravity of the two objects respectively.
[0075] Objects on the surface are affected by the gravity of the earth, which is the combined force of the earth's gravity and the inertial centrifugal force caused by the earth's rotation. Due to the uneven density distribution of underground materials, the gravitational effects on the surface are different. The underground three-dimensional space is rasterized into several regions, and the gravitational force on each surface observation point is the combined force of these n regions. Therefore, according to the universal gravitation formula, the gravitational force of the underground region on the surface observation point is calculated by the following formula (2):
[0076]
[0077] Among them, G is the gravitational coefficient, m1 is the mass of the surface observation point (which can be understood as the mass of the observer), and m i is the mass of the ith underground region (determined by the density when the volume is known), r is the distance from the surface observation point to the ith underground region, F is the gravitational force between the surface observation point and the ith underground region, and n is the number of underground regions.
[0078] After obtaining the gravitational force F exerted by the underground area on the surface observation point, the component of the gravitational force exerted on the surface observation point in the direction perpendicular to the surface is calculated by the following formula (3):
[0079]
[0080] Among them, F y is the component of gravity acting on the observation point on the surface in the direction perpendicular to the surface, h i is the distance from the ith area underground to the surface; multiple observation points are collected on the surface to form an observation point matrix, and gravity field data can be obtained.
[0081] The gravity forward modeling process is the process of solving the underground density information shown in formula (3) to obtain gravity field data, which can be generalized into the following formula (4):
[0082]
[0083] Among them, d is a one-dimensional vector (underground density information) pulled out according to the position relationship of the three-dimensional density model, including the terms related to i in formula (3); A is the forward modeling matrix, including the remaining terms in formula (3); g is the forward modeling solution, that is, the gravity field observed on the surface. The forward modeling processing based on the neural network can be realized by using a neural network composed of neurons to implement the matrix multiplication operation shown in formula (4).
[0084] In some embodiments, the process of building an inversion network includes: using an encoder, an expander, a decoder and a three-dimensional refiner to construct an inversion network, and obtaining an inversion solution result through the inversion network; wherein the encoder is used to obtain a first feature map based on gravity field data; the expander is used to obtain a second feature map based on an energy distribution curve; the decoder is used to perform feature fusion on the first feature map and the second feature map to obtain a fused feature map; the three-dimensional refiner is used to obtain three-dimensional density data based on the fused feature map, and the inversion solution result includes three-dimensional density data.
[0085] The inversion process is the process of solving the underground density information d from the gravity field data g observed on the surface, that is, The inversion problem is an ill-posed problem with a non-unique solution space, so a deep learning network is used to fit the process and the inversion results are constrained with the help of prior knowledge (energy distribution curve) from expert knowledge or observation data.
[0086] Figure 3 Schematic diagram of the inversion solution process in the embodiment of the present application, such as Figure 3 As shown, the encoder and expander are used to process the two-dimensional gravity field data and energy distribution curve respectively, and the processing results are input into the decoder for solution, and then converted into the three-dimensional refiner for post-processing to obtain the three-dimensional density data.
[0087] Specifically, the encoder is used to extract features from gravity field data to obtain a plurality of first feature maps of different sizes. Figure 4 is a schematic diagram of an encoder in an embodiment of the present application, such as Figure 4 As shown, an encoder is explained by taking a cascade of four jump blocks (ResBlock) as an example, wherein the input of the latter jump block is the output of the previous jump block, the input of the first jump block is the original gravity field data, and the four jump blocks output the first feature map of four sizes respectively, the number of feature map channels is C, the size ratio is 8:4:2:1, and the largest feature map size is consistent with the input gravity field size. Three downsampling operations are performed in the encoder, and the downsampling method is a convolutional layer with a step size of two. In some embodiments, the expander is specifically used to expand the energy distribution curve into multiple second feature maps, the number of second feature maps is the same as the number of first feature maps, and the corresponding sizes are the same.
[0088] Specifically, the expander resamples the energy distribution curve into a vector of length C and then expands it into multiple three-dimensional second feature maps. The size of the second feature map is similar to the encoder output feature map. Figure 1 That is to say, when the encoder outputs four first feature maps with a size ratio of 8:4:2:1, the expander also outputs four second feature maps with a size ratio of 8:4:2:1.
[0089] In some embodiments, the decoder is specifically used to perform feature fusion on a first feature map and a second feature map of the same size to obtain feature fusion results of multiple different sizes, and perform feature fusion on multiple feature fusion results of different sizes to obtain a fused feature map.
[0090] Figure 5 Schematic diagram of a decoder in an embodiment of the present application, such as Figure 5 As shown in Figure 2, the decoder receives feature maps of four sizes from the encoder and expander, and uses the jump block as the basic unit for feature extraction. Figure 6 The attention fusion module shown fuses the feature maps from the encoder, the dilator, and the upsampled feature maps. The decoder contains four feature propagation channels, processing feature maps of four sizes, using Figure 7 The cross-scale feature fusion module shown implements cross-scale feature fusion.
[0091] In some embodiments, the three-dimensional refiner is specifically used to convert the fused feature map into a three-dimensional feature map, and obtain three-dimensional density data based on the three-dimensional feature map.
[0092] Specifically, the 3D refiner transforms the 2D feature map of size [B, C, H, W] (B and C are batch parameters and number of channels, respectively, and H and W are two dimensions of the feature map) output by the decoder into a 3D feature map of size [B, C, D, H, W], and obtains the final 3D density data after processing using a 3D jump block.
[0093] In some embodiments, closed-loop self-supervised learning is performed through a forward modeling network and an inversion network to obtain gravity inversion results, including: performing inversion solutions through an inversion network based on original gravity field data and energy distribution curves to obtain three-dimensional density data; performing forward solutions through a forward modeling network based on the three-dimensional density data to obtain reconstructed gravity field data; and performing network parameter optimization training based on the original gravity field data and the reconstructed gravity field data.
[0094] Specifically, based on the given density data to be solved, the inversion and forward modeling can be performed using the following formula (5):
[0095]
[0096]
[0097] Among them, g is the original gravity field data, is the inversion solution process, d′ is the three-dimensional density data obtained by inversion solution, is the forward solution process, g′ is the reconstructed gravity field data obtained by the forward solution, ‖d′‖ and TV(d′) are the regularization constraints of the inversion solution results, and λ1, λ2, λ3, and λ4 are preset parameters. φ is the mapping function from the one-dimensional vector d pulled from the three-dimensional density model according to the position relationship to the energy distribution curve l, and its operation is to average the three-dimensional density data of size [D, H, W] in the two dimensions of H and W to obtain a vector of length D, and then perform the following normalization, as shown in the following formula (6):
[0098]
[0099] The vector obtained by normalization is the energy distribution curve. In addition, the Adam optimizer is used to train the parameters of the inversion model, and the decay learning rate is used during the training process.
[0100] In some embodiments, in order to verify the effectiveness and superiority of the present application, the method proposed in the present application is applied to a certain observed gravity field to demonstrate the processing effect of the closed-loop self-supervised network model in the present application.
[0101] In this embodiment, the training data and test data are sampled at intervals of 100m in the horizontal direction and 50m in the depth direction. In the loss function (Formula 5), the parameters λ1=λ3=λ4=1, λ2=0.1, the initial learning rate of the network is 0.0001, the target data is iterated 3000 times, and the learning rate is decayed to 50% of the current learning rate at the 1000th, 2000th, and 2500th iterations, respectively.
[0102] Figures 8 to 11 The super-resolution prediction results of this application are shown, where: Figure 8 is a schematic diagram of the original gravity field data in this embodiment, Fig. 9 is the three-dimensional density data (inversion solution result) obtained by the inversion network in this embodiment, Fig.10 is the reconstructed gravity field data (forward modeling result) obtained by processing the three-dimensional density data (inversion solution result) through the forward modeling network in this embodiment, Fig.11 This is a longitudinal section of the inversion solution result in this embodiment, refer to Figures 8 to 11 ,It can be seen that the closed-loop processing results are consistent with the original gravity field data in ,visual perception.
[0103] In addition, the algorithm implemented in the present application and the existing deep learning-based algorithm are quantified and compared from the dimensions of MAE, Peak Signal to Noise Ratio (PSNR), Structural Similarity (SSIM), the number of inversion network parameters, and whether additional synthetic data is needed for training. The results shown in Table 1 are obtained. It can be seen that the present application achieves better results using fewer parameters without the need for additional training data.
[0104]
[0105] It should be understood that, although the various steps in the flowcharts in the above-described embodiments are sequentially displayed according to the indications of the arrows, these steps are not necessarily executed sequentially in the order indicated by the arrows. Unless there is a clear description in this article, the execution of these steps is not strictly limited in order, and they can be executed in other orders. Moreover, at least a portion of the steps in the figure may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily to be carried out sequentially, but can be executed in turn or alternately with other steps or at least a portion of the sub-steps or stages of other steps.
[0106] In some embodiments, a terminal device is provided.
[0107] Fig.12A schematic diagram of the structure of a terminal device provided in an embodiment of the present application, such as Fig.12 As shown, the terminal device includes: a processor 11 and a memory 12.
[0108] The memory 12 is used to store programs and data. The processor 11 calls the program stored in the memory to execute the following technical solutions:
[0109] Build a forward modeling network and an inversion network. The inversion network and the forward modeling network form a closed-loop structure.
[0110] The gravity inversion results are obtained by performing closed-loop self-supervised learning through the forward network and the inversion network.
[0111] In some embodiments, the process of building a forward modeling network includes:
[0112] According to the universal gravitation formula, determine the gravitational force exerted on the surface observation point by the underground area;
[0113] Determine the component of gravity acting on the observation point on the surface in the direction perpendicular to the surface;
[0114] According to the component forces corresponding to multiple surface observation points, the forward solution is obtained.
[0115] In some embodiments, according to the universal gravitation formula, the gravitational force exerted on the surface observation point by the underground area is calculated by the following formula:
[0116]
[0117] Among them, G is the gravitational coefficient, m1 is the mass of the observation point on the surface, and m i is the mass of the ith underground region, r is the distance from the surface observation point to the ith underground region, F is the gravitational force between the surface observation point and the ith underground region, and n is the number of underground regions;
[0118] The component of gravity acting on the surface observation point in the direction perpendicular to the surface is calculated using the following formula:
[0119]
[0120] Among them, F y is the component of gravity acting on the observation point on the surface in the direction perpendicular to the surface, h i is the distance from the ith underground area to the surface;
[0121] According to the component forces corresponding to multiple surface observation points, the forward solution is calculated by the following formula:
[0122]
[0123] Among them, d is the one-dimensional vector pulled out by the three-dimensional density model according to the position relationship, A is the forward modeling matrix, and g is the forward modeling solution result.
[0124] In some embodiments, the process of building an inversion network includes:
[0125] An inversion network is constructed using an encoder, an expander, a decoder and a three-dimensional refiner, and an inversion solution result is obtained through the inversion network;
[0126] Wherein, the encoder is used to obtain a first characteristic map according to the gravity field data;
[0127] The expander is used to obtain a second characteristic map according to the energy distribution curve;
[0128] The decoder is used to perform feature fusion on the first feature map and the second feature map to obtain a fused feature map;
[0129] The three-dimensional refiner is used to obtain three-dimensional density data according to the fused feature map, and the inversion solution result includes the three-dimensional density data.
[0130] In some embodiments, the encoder is specifically used to extract features from gravity field data to obtain a plurality of first feature maps of different sizes.
[0131] In some embodiments, the expander is specifically used to expand the energy distribution curve into a plurality of second characteristic graphs, the number of the second characteristic graphs is the same as the number of the first characteristic graphs, and the sizes thereof are correspondingly the same.
[0132] In some embodiments, the decoder is specifically used to perform feature fusion on a first feature map and a second feature map of the same size to obtain feature fusion results of multiple different sizes, and perform feature fusion on multiple feature fusion results of different sizes to obtain a fused feature map.
[0133] In some embodiments, the three-dimensional refiner is specifically used to convert the fused feature map into a three-dimensional feature map, and obtain three-dimensional density data based on the three-dimensional feature map.
[0134] In some embodiments, closed-loop self-supervised learning is performed through a forward network and an inversion network to obtain a gravity inversion result, including:
[0135] According to the original gravity field data and energy distribution curve, the three-dimensional density data is obtained by inversion through the inversion network;
[0136] According to the three-dimensional density data, forward modeling is performed through the forward modeling network to obtain the reconstructed gravity field data;
[0137] Network parameter optimization training is performed based on the original gravity field data and the reconstructed gravity field data.
[0138] In some embodiments, the inversion and forward modeling are solved by the following formula:
[0139]
[0140]
[0141] Among them, g is the original gravity field data, is the inversion solution process, d′ is the three-dimensional density data obtained by inversion solution, is the forward solution process, g′ is the reconstructed gravity field data obtained by the forward solution, ‖d′‖1 and TV(d′) are the regularization constraints of the inversion solution results, φ is the mapping function from the one-dimensional vector d pulled from the position relationship of the three-dimensional density model to the energy distribution curve l, and λ1, λ2, λ3, and λ4 are preset parameters.
[0142] In the above terminal device, the memory and the processor are electrically connected directly or indirectly to realize data transmission or interaction. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines, such as through a bus connection. The memory stores computer-executable instructions for implementing the data access control method, including at least one software function module that can be stored in the memory in the form of software or firmware. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory.
[0143] The memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. The memory is used to store programs, and the processor executes the programs after receiving the execution instructions. Furthermore, the software programs and modules in the above-mentioned memory may also include an operating system, which may include various software components and / or drivers for managing system tasks (such as memory management, storage device control, power management, etc.), and may communicate with various hardware or software components to provide an operating environment for other software components.
[0144] The processor may be an integrated circuit chip having the ability to process signals. The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc. The disclosed methods, steps and logic block diagrams in the embodiments of the present application may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.
[0145] In some embodiments, a computer-readable storage medium is provided, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the steps of the various method embodiments of the present application.
[0146] In some embodiments, a computer program product is provided, including a computer program, which implements the steps of the method embodiments of the present application when executed by a processor.
[0147] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0148] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the application disclosed herein. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0149] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. A three-dimensional gravity inversion method based on a closed-loop network, characterized in that: include: Building a forward modeling network and an inversion network, wherein the inversion network and the forward modeling network form a closed-loop structure; Performing closed-loop self-supervised learning through the forward modeling network and the inversion network to obtain a gravity inversion result; The process of building the forward modeling network includes: According to the universal gravitation formula, determine the gravitational force exerted on the surface observation point by the underground area; Determining the component of the gravitational force acting on the surface observation point in a direction perpendicular to the surface; According to the component forces corresponding to multiple surface observation points, the forward solution results are obtained; According to the universal gravitation formula, the gravitational force exerted on the surface observation point by the underground area is calculated by the following formula: in, is the gravitational coefficient, is the mass of the surface observation point, For the underground The quality of the region, From the surface observation point to the underground The distance between the regions, The surface observation point and the underground The gravitational force of a region, is the number of underground areas; The component of gravity acting on the surface observation point in the direction perpendicular to the surface is calculated using the following formula: in, is the component of gravity acting on the observation point on the surface in the direction perpendicular to the surface, For the underground The distance from the region to the surface; According to the component forces corresponding to multiple surface observation points, the forward solution is calculated by the following formula: in, is a one-dimensional vector drawn according to the position relationship of the three-dimensional density model. is the forward modeling matrix, The forward modeling result is obtained; The process of building the inversion network includes: An inversion network is constructed using an encoder, an expander, a decoder, and a three-dimensional refiner, and an inversion solution result is obtained through the inversion network; Wherein, the encoder is used to obtain a first characteristic map according to gravity field data; The expander is used to obtain a second characteristic diagram according to the energy distribution curve; The decoder is used to perform feature fusion on the first feature map and the second feature map to obtain a fused feature map; The three-dimensional refiner is used to obtain three-dimensional density data according to the fusion feature map, and the inversion solution result includes the three-dimensional density data; The closed-loop self-supervised learning is performed through the forward modeling network and the inversion network to obtain a gravity inversion result, including: According to the original gravity field data and energy distribution curve, the three-dimensional density data is obtained by inversion through the inversion network; According to the three-dimensional density data, forward modeling is performed through a forward modeling network to obtain reconstructed gravity field data; Network parameter optimization training is performed according to the original gravity field data and the reconstructed gravity field data.
2. The method according to claim 1, characterized in that: The encoder is specifically used to extract features from the gravity field data to obtain a plurality of first feature maps of different sizes.
3. The method according to claim 2, characterized in that The expander is specifically used to expand the energy distribution curve into a plurality of second characteristic graphs, wherein the number of the second characteristic graphs is the same as the number of the first characteristic graphs and the sizes thereof are correspondingly the same.
4. The method according to claim 3, characterized in that The decoder is specifically used to perform feature fusion on the first feature map and the second feature map of the same size to obtain a plurality of feature fusion results of different sizes, and perform feature fusion on the plurality of feature fusion results of different sizes to obtain a fused feature map.
5. The method according to claim 1, characterized in that The three-dimensional refiner is specifically used to convert the fused feature map into a three-dimensional feature map, and obtain three-dimensional density data according to the three-dimensional feature map.
6. The method according to claim 1, characterized in that The inversion and forward modeling are solved by the following formula: in, is the original gravity field data, For the inversion solution process, is the three-dimensional density data obtained by inversion. is the forward solution process, is the reconstructed gravity field data obtained by forward modeling, and is the regularization constraint of the inversion solution, It is a one-dimensional vector drawn according to the position relationship of the three-dimensional density model Energy distribution curve The mapping function of are preset parameters.
