Potential field data downward continuation method, device, equipment and medium

Abstract

The invention discloses a potential field data downward continuation method and device, equipment and a medium, and relates to the technical field of geophysical geological resource exploration, and the method comprises the steps: learning a rule in downward continuation through employing a physical information neural network containing a total loss function, and training a better downward continuation model; wherein the total loss function comprises partial differential structure loss, boundary value loss and true value loss; the partial differential structure loss in the gravity field is an error of a product of a Laplacian operator of prediction data and gravity; the partial differential structure loss in the magnetic field is an error of a product of a Laplacian operator of prediction data and magnetic force; the boundary value loss is obtained by predicting data calculation after downward continuation; the true value loss is the error of the data after the downward continuation and the data after the downward continuation of the label, and the trained physical information neural network is utilized to carry out the downward continuation prediction on the actually measured potential field data, so that the high-precision downward continuation of the potential field data is realized.

Description

Technical Field

[0001] The present application relates to the technical field of geophysical geological resource exploration, and in particular to a method, device, equipment and medium for downward extension of potential field data. Background Art

[0002] Due to the limitations of the downward continuation theory, its inherent unstable characteristics will quickly amplify noise interference during downward continuation. A large number of studies have shown that when using traditional methods, the noise amplification will be particularly strong after only three to five point extensions, and even cover up the information in the original data. Even the most mainstream iterative methods cannot fundamentally change the instability of downward continuation.

[0003] The essence of downward continuation is to find an optimal downward continuation function by analyzing and processing data, but downward continuation is a more complex and nonlinear problem. Artificially finding expressions for the relationship before and after continuation generally relies on personal assumptions and experience. Therefore, the resulting equations have certain deviations and cannot capture the changing rules and details of the downward continuation. Therefore, a method for downward continuation of potential field data that improves the accuracy of downward continuation is urgently needed. Summary of the invention

[0004] The purpose of this application is to provide a method, device, equipment and medium for downward extension of potential field data, which can improve the accuracy of downward extension.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for downward extension of potential field data, comprising:

[0007] Acquiring measured potential field data; the measured potential field data is gravity field data or magnetic field data;

[0008] The measured potential field data are input into a trained physical information neural network to obtain predicted downward extended data; the trained physical information neural network is trained by importing sample potential field data into a physical information neural network containing a total loss function; the total loss function includes partial differential structure loss, boundary value loss and true value loss; when the predicted downward extended data are gravity field data, the partial differential structure loss is the error between the Laplace operator and the gravity product of the predicted downward extended data; the gravity product is the product of the gravity field and the density field; when the predicted downward extended data are magnetic field data, the partial differential structure loss is the error between the Laplace operator and the magnetic product of the predicted downward extended data; the magnetic product is the product of the magnetic field and the magnetization intensity; the boundary value loss is calculated by predicting the downward extended data; the true value loss is the error between the predicted downward extended data and the label downward extended data corresponding to the predicted downward extended data.

[0009] In a second aspect, the present application provides a device for downward extension of potential field data, comprising:

[0010] The measured potential field data acquisition module is used to: acquire the measured potential field data; the measured potential field data is gravity field data or magnetic field data;

[0011] The downward extended data prediction module is used for: inputting the measured potential field data into the trained physical information neural network to obtain the predicted downward extended data; the trained physical information neural network is trained by importing the sample potential field data into the physical information neural network containing the total loss function; the total loss function includes partial differential structure loss, boundary value loss and true value loss; when the predicted downward extended data is gravity field data, the partial differential structure loss is the error between the Laplace operator and the gravity product of the predicted downward extended data; the gravity product is the product of the gravity field and the density field; when the predicted downward extended data is magnetic field data, the partial differential structure loss is the error between the Laplace operator and the magnetic product of the predicted downward extended data; the magnetic product is the product of the magnetic field and the magnetization intensity; the boundary value loss is calculated by predicting the downward extended data; the true value loss is the error between the predicted downward extended data and the label downward extended data corresponding to the predicted downward extended data.

[0012] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned method for downward extension of bit field data.

[0013] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for downward extension of bit field data.

[0014] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0015] The present application provides a method, device, equipment and medium for downward extension of potential field data, wherein a trained physical information neural network is obtained through total loss function training; wherein the total loss function includes partial differential structure loss, boundary value loss and true value loss; the partial differential structure loss of the gravity field is the error between the Laplace operator of the predicted data after downward extension and the product of the gravity field and the density field; the partial differential structure loss in the magnetic field is the error between the Laplace operator of the predicted data after downward extension and the product of the magnetic field and the magnetization intensity; the boundary value loss is calculated by predicting the data after downward extension; the true value loss is the error between the predicted data after downward extension and the data after downward extension of its corresponding label. After obtaining the measured potential field data, the trained physical information neural network is used to perform downward extension prediction on the measured potential field data to ensure that the output downward extension value conforms to the physical laws, thereby solving the problem of low downward extension accuracy of the existing method and realizing high-precision downward extension of the potential field data. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 This is an application environment diagram of a method for downward extension of potential field data in one embodiment of the present application;

[0018] Figure 2 A schematic diagram of a process of downward extension of potential field data provided in one embodiment of the present application;

[0019] Figure 3 The gravity anomaly plane diagram and main cross-sectional diagram of three shallow spherical models provided in one embodiment of the present application; wherein, Figure 3 (a) is the plane diagram of gravity anomalies of three spherical models in the shallow part; Figure 3 (b) is the main profile of gravity anomaly of three spherical models in the shallow part;

[0020] Figure 4 The present invention provides an embodiment of the present invention using the potential field data downward extension method and integral iteration method, Taylor series expansion method, Tikhonov regularization method provided by the present invention to Figure 3 The predicted downward extension data plane map and main profile map are obtained by extending the noise-free data and noise-containing data of the shallow three spherical models downward by 50m (5 times the point spacing); among them, Figure 4 (a) in Figure 3The contour map of theoretical gravity anomaly data generated by three spherical models at 0m in the shallow part; Figure 4 (b) in Figure 3 The contour map of the theoretical gravity anomaly data generated by the three spherical models at 0m in the shallow part with 10% noise added; Figure 4 (c) in Figure 3 The contour map of theoretical gravity anomaly data generated by three spherical models at 50m in the shallow part; Figure 4 (d) is the method for downward extension of the potential field data provided by this application. Figure 3 The contour map of the PINN prediction value obtained by extending the noise-free data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4 (e) is the method for downward extension of the potential field data provided by this application. Figure 3 The contour map of the PINN prediction value obtained by extending the gravity anomaly containing noise data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4 (f) is the method for downward extension of the potential field data provided by this application. Figure 3 Schematic diagram of the comparison between the PINN predicted values ​​and the true values ​​obtained by extending the noise-free data and noise-containing data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4 (g) is the Taylor series expansion method for Figure 3 The contour map of the predicted values ​​obtained by extending the noise-free data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4 (h) is the Taylor series expansion method for Figure 3 The contour map of the predicted values ​​obtained by extending the gravity anomaly containing noise data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4 (i) is the Taylor series expansion method for Figure 3 Schematic diagram of the comparison between the PINN predicted values ​​and the true values ​​obtained by extending the noise-free data and noise-containing data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4 (j) in the figure is the Tikhonov regularization method. Figure 3 The contour map of the predicted values ​​obtained by extending the noise-free data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4 (k) is the Tikhonov regularization method used for Figure 3 The contour map of the predicted values ​​obtained by extending the gravity anomaly containing noise data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4(l) in the figure is the Tikhonov regularization method. Figure 3 Schematic diagram of the comparison between the PINN predicted values ​​and the true values ​​obtained by extending the noise-free data and noise-containing data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4 The (m) in the equation is the integral iteration method. Figure 3 The contour map of the predicted values ​​obtained by extending the noise-free data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4 The (n) in the formula is the integral iteration method. Figure 3 The contour map of the predicted values ​​obtained by extending the gravity anomaly containing noise data of the shallow three spherical models downward by 50m (5 times the point spacing); Figure 4 The (o) in the figure is the integral iteration method. Figure 3 Schematic diagram of the comparison between the PINN predicted values ​​and the true values ​​obtained by extending the noise-free data and noise-containing data of the shallow three spherical models downward by 50m (5 times the point spacing);

[0021] Figure 5 The gravity anomaly plane diagram and main cross-sectional diagram of three spherical models in the deep provided by an embodiment of the present application; wherein, Figure 5 (a) is the plane diagram of gravity anomalies of three spherical models in the deep; Figure 5 (b) is the main profile of gravity anomaly of three spherical models in the deep area;

[0022] Figure 6 To use the potential field data downward extension method and Taylor series expansion method provided by this application to Figure 3 The plan view and main profile of the shallow three spherical model gravity anomaly data are calculated by extending downward 200m (20 times the point spacing); among them, Figure 6 (a) in Figure 3 The contour map of theoretical gravity anomaly data generated by three spherical models at 0m in the shallow part; Figure 6 (b) in Figure 3 The contour map of theoretical gravity anomaly data generated by three spherical models at 200m in the shallow part; Figure 6 (c) is the method for downward extension of the potential field data provided by this application. Figure 3 The contour map of the PINN downward extension prediction value obtained by extending the shallow three spherical model gravity anomaly data downward by 200m (20 times the point spacing); Figure 6 (d) is the method for downward extension of the potential field data provided by this application. Figure 3Schematic diagram of the comparison between the PINN downward continuation prediction value and the actual value obtained by performing a 200m (20 times the point spacing) downward continuation calculation on the shallow three spherical model gravity anomaly data; Figure 6 (e) is the Taylor series expansion method for Figure 3 The contour map of the Taylor series expansion result obtained by extending the shallow three spherical model gravity anomaly data downward by 200m (20 times the point spacing); Figure 6 (f) is the Taylor series expansion method for Figure 3 Schematic diagram of the comparison between the Taylor series expansion method extension result and the true value obtained by extending the shallow three spherical model gravity anomaly data downward by 200m (20 times the point spacing);

[0023] Figure 7 A schematic diagram of functional modules of a device for downward extension of potential field data provided in one embodiment of the present application;

[0024] Figure 8 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0026] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0027] The method for downward extension of potential field data provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be set up separately, or it can be integrated on the server 104, or it can be placed on the cloud or other servers. The terminal 102 can send the measured potential field data to the server 104. After the server 104 receives the measured potential field data, for the measured potential field data, the server 104 inputs the measured potential field data into the trained physical information neural network to obtain the predicted downward extension data. The server 104 can feed back the obtained predicted downward extension data for the measured potential field data to the terminal 102. In addition, in some embodiments, the method for extending the potential field data downward can also be implemented separately by the server 104 or the terminal 102, such as the terminal 102 can directly extend the potential field data downward for the measured potential field data, or the server 104 can obtain the measured potential field data from the data storage system, and extend the potential field data downward for the measured potential field data.

[0028] The terminal 102 may be, but is not limited to, various desktop computers, laptop computers, smart phones, tablet computers, IoT devices, and portable wearable devices. The IoT devices may be smart speakers, smart TVs, smart air conditioners, smart vehicle-mounted devices, etc. The portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server 104 may be implemented as an independent server or a server cluster consisting of multiple servers, or may be a cloud server.

[0029] In an exemplary embodiment, Figure 2 As shown, a method for downward extension of potential field data is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server 104 in the example is used as an example to illustrate, including the following steps 201 to 202. Among them:

[0030] Step 201: Acquire measured potential field data; the measured potential field data is gravity field data or magnetic field data.

[0031] Step 202: Input the measured potential field data into the trained physical information neural network to obtain the predicted downward extended data; the trained physical information neural network is trained by importing the sample potential field data into the physical information neural network containing the total loss function; the total loss function includes partial differential structure loss, boundary value loss and true value loss; when the predicted downward extended data is gravity field data, the partial differential structure loss is the error between the Laplace operator and the gravity product of the predicted downward extended data; the gravity product is the product of the gravity field and the density field; when the predicted downward extended data is magnetic field data, the partial differential structure loss is the error between the Laplace operator and the magnetic product of the predicted downward extended data; the magnetic product is the product of the magnetic field and the magnetization intensity; the boundary value loss is calculated by the predicted downward extended data; the true value loss is the error between the predicted downward extended data and the label downward extended data corresponding to the predicted downward extended data.

[0032] Implement the above-mentioned steps 201 to 202, and use the trained physical information neural network to perform downward extension prediction on the measured potential field data to ensure that the output downward extension value conforms to the laws of physics, solve the problem of low downward extension accuracy of the existing method, and achieve high-precision downward extension of the potential field data. As a universal approximator, the physical information neural network (Physics-informed Neural Networks, PINNs) can approximate any function as long as there are nonlinear layers and neurons. In view of the pain point that the optimal formula cannot be found in the downward extension, it is expected to solve the problem of inaccurate downward extension by continuously approximating with the help of neural networks. The physical information neural network adds physical constraints (loss function) to the neural network, which can ensure that the neural network prediction conforms to the laws of physics, the results are more accurate, and the accuracy and stability of the downward extension are improved.

[0033] In another exemplary embodiment of the present application, before the measured potential field data is input into the trained physical information neural network to obtain the predicted downward extended data, the method further includes: training the physical information neural network, specifically including the following steps 301 to 304:

[0034] Step 301: Acquire a sample data set; the sample data set includes a number of sample potential field data and data after downward extension of a label corresponding to each sample potential field data; the sample potential field data is gravity field data or magnetic field data.

[0035] Step 302: Input the sample potential field data into the physical information neural network to obtain the predicted downward extended data.

[0036] Step 303: Calculate the total loss function value based on the predicted downward extended data and the sample potential field data.

[0037] Step 304: Determine whether the set stop condition is met. If not, update the weights in the physical information neural network according to the total loss function value, and return to the step of "inputting the sample potential field data into the physical information neural network". If yes, end the training and obtain the trained physical information neural network. The set stop condition is that the current number of iterations reaches the maximum number of iterations or the total loss function value is less than the set error value.

[0038] Obtain sample potential field data, which includes gravity field data or magnetic field data, build a physical information neural network, set appropriate physical constraints for the neural network, continuously adjust relevant parameters to obtain the best training model, use the sample data set for model training, and use the verification data for downward extension and data verification.

[0039] In a specific embodiment, the physical information neural network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer and an output layer connected in sequence. The number of neurons in the first hidden layer, the second hidden layer, the third hidden layer and the fourth hidden layer are 512, 256, 128 and 64 respectively. The selection of a neural network with a continuously decreasing number of neurons such as 512, 256, 128 and 64 ensures that a larger nonlinear transformation is performed after the data is input into the physical information neural network, and the best effect can be maintained at the end.

[0040] The input of the physical information neural network is potential field data, actual or simulated data before extension. The data before extension can be gravity field data or magnetic field data that needs to be processed (downward extension), or it can be theoretical data generated by a sphere, or it can be actual gravity field or magnetic field data of a certain place in reality. The output of the physical information neural network is the data after downward extension.

[0041] This application designs the loss of the physical information neural network based on the specific problem of downward continuation. The following takes the prediction of gravity field data after downward continuation as an example to specifically explain the loss function:

[0042] (1) Partial differential structure loss

[0043] The partial differential structure loss first calculates the gradient of the predicted downward extended data to the density based on the gravity field value in the predicted downward extended data:

[0044]

[0045] Among them, g(x) is the gravity field data, is the gradient of the gravity field data, is the gradient of density v, is the partial derivative of the gravity field data g(x) with respect to the density ρ.

[0046] The Laplace operator of the gravity field is then calculated according to the following formula:

[0047]

[0048] in, is the Laplace operator of the gravity field data, is the second-order partial derivative of the gravity field data g(x) with respect to density ρ, are the partial derivatives of density in the x, y, and z directions respectively.

[0049] The difference between the Laplace operator and the product of the gravity field and the density is used as the residual term of the Poisson equation, and the partial differential loss is calculated accordingly. Therefore, when the data after downward extension is predicted to be gravity field data, the calculation formula of the partial differential structure loss is as follows:

[0050]

[0051] Among them, L PDE1 is the partial differential structure loss corresponding to the gravity field, i represents the data after the i-th prediction is extended downward, i=1, 2, 3...N, N is the total number of data after the prediction is extended downward.

[0052] The partial differential loss is used to calculate the Laplace operator of the gravitational potential field output by the model. The difference between the product of the density field ρg(x) and the density field ρg(x). By minimizing this loss, the model can be made to better fit the physical equations.

[0053] When the predicted downward extension data is magnetic field data, the calculation formula of partial differential structure loss is as follows:

[0054]

[0055] Among them, L PDE2 is the partial differential structure loss corresponding to the magnetic field, H(x) is the magnetic field data, and M is the magnetization intensity.

[0056] (2) Boundary value loss

[0057] The downward extension is to solve the boundary problem shown in the following formula:

[0058]

[0059] In the formula, is the second-order partial derivative of the gravity field data g(x) along the z direction. According to the above formula, the boundary value loss can be set. Therefore, when the data after downward extension is predicted to be gravity field data, the calculation formula of the boundary value loss is as follows:

[0060]

[0061] Among them, L BC1 is the boundary value loss corresponding to the gravity field; is the second-order partial derivative of the gravity field data g(x) along the x direction, indicating the change of curvature in this direction; It is the second-order partial derivative of the gravity field data g(x) along the y direction, indicating the change of curvature in this direction.

[0062] When the predicted downward extended data is magnetic field data, the calculation formula of the boundary value loss is as follows:

[0063]

[0064] Where, L BC2 is the boundary value loss corresponding to the magnetic field; is the second-order partial derivative of the magnetic field data H(x) along the z direction; is the second-order partial derivative of the magnetic field data H(x) along the x direction, indicating the change of curvature in this direction; It is the second-order partial derivative of the magnetic field data H(x) along the y direction, indicating the change of curvature in this direction.

[0065] By continuously training to reduce the value of boundary loss, the boundary problem can be fitted.

[0066] (3) True value loss

[0067] The calculation formula of the true value loss is as follows:

[0068]

[0069] Among them, L Data is the true value loss, Observed iData The label corresponding to the i-th prediction after the downward extension of the data is extended downward, Predicted iData is the data after the i-th prediction is extended downward, i represents the data after the i-th prediction is extended downward, i=1, 2, 3...N, N is the total number of data after the prediction is extended downward.

[0070] The total loss function includes partial differential structure loss, boundary value loss and true value loss. The total loss function value is calculated by taking weighted sum of the partial differential structure loss, boundary value loss and true value loss. Specifically, the partial differential structure loss, boundary value loss and true value loss can be added with a weight of 1:1:1 to obtain the total loss function value.

[0071] Select different learning rates: 0.01, 0.001, Adam (0.0001), Adam (0.0005), Adam (0.00001), and different activation functions: Tanh, ReLU, SeLU. The maximum number of iterations is 50,000. Add the early stopping algorithm to train the model. Select the best learning rate and activation function after comparison and verification.

[0072] After the neural network is trained, the total loss function value of the current iteration is observed. When the total loss function value of the current iteration is less than the set error value or the current iteration reaches the maximum iteration number, the training ends and a trained physical information neural network is obtained.

[0073] The following is a comparative analysis of the results of the downward extension method using the physical information neural network in this application with the traditional integral iteration method, Taylor series expansion, and Tikhonov regularization. The results obtained in the above steps are plotted to intuitively display the downward extended data predicted by the physical information neural network.

[0074] In an exemplary embodiment of the present application, three spheres with different burial depths and different radii are used to calculate the gravity data at 0m (original measurement plane) and 50m (extended downward by 5 times the point distance) according to the sphere gravity field formula. The calculation results of the present application are also compared with the results of the integral iteration method, Taylor series expansion, and Tikhonov regularization method. Figure 3 The gravity data anomaly diagram of three spherical models is shown. The sphere radius is 40m, 50m, and 80m, the burial depth is 180m, 200m, and 220m, and the center coordinates are (0, 300, 180), (0, 0, 200), and (0, -300, 220), respectively. The residual density is 1g / cm3. The present application and the integral iteration method, Taylor series expansion, and Tikhonov regularization method are used to extend downward by 50m (5 times the point spacing). In order to verify the robustness of the physical information neural network to noise, Figure 3 10% white noise is added to the medium gravity anomaly data, and 50m (5 times the point spacing) downward continuation is performed using the potential field data downward continuation method and integral iteration method, Taylor series expansion method, and Tikhonov regularization method provided in this application.

[0075] The formula for the gravitational field of a sphere is as follows

[0076]

[0077] Among them, g(x, y, z) is the gravitational field at a certain point in space; G is the gravitational constant, which represents the strength of gravity; Δp is the residual density, which represents the density difference between the sphere and the surrounding medium; h is the height (depth) of the center of the sphere, that is, the coordinate of the center of the sphere on the z-axis.

[0078] In order to further compare the extension effects of several methods, the section at x = 0m is compared with the theoretical model section.

[0079] Figure 4 Comparison results of different methods calculated under noise-free data and noisy data.

[0080] pass Figure 4 It can be seen that PINNs can predict high-precision results for both noise-free and noise-containing data, indicating that physical information neural networks can better solve the problem of downward continuation. Under noise-free conditions, the Taylor series expansion method can better fit the actual value of downward continuation, but in the case of noisy data, it cannot predict the actual value well, and its results are seriously disturbed by noise. The Tikhonov regularization method performs similarly under noisy and non-noisy data, and the trend of the predicted value is similar to the actual value, which proves that it has good robustness to noise and can effectively remove the interference of noise. However, in the continuation calculation, while removing the noise, the calculation results are too smooth, removing the detailed features of the actual data. The integral iteration method predicts the results of downward continuation well in the absence of noise, with a small error, but the effect is average in the case of noisy data, and its results are seriously affected by noise, and the noise cannot be effectively removed. From the comparison, it can be seen that the overall effect of using PINN for downward extension is better, the error is minimal, and the consistency with the theoretical data is the highest, which proves that PINN has strong robustness and can effectively remove the interference of noise on the data. This advantage is of great significance in obtaining key underground information in practical applications.

[0081] In order to verify the effect of physical information neural network in deep large point spacing extension, three spheres with different burial depths and radii are used to calculate gravity data, and the calculation results of this application are also compared with the results of Taylor series expansion method. Figure 5 The gravity data anomaly diagrams of three spherical models are shown. The sphere radii are 40m, 50m, and 80m, respectively. The burial depths are 420m, 400m, and 380m, respectively. The center coordinates are (0, 300, 420), (0, 0, 400), and (0, -300, 380), respectively. The residual density is 1g / cm3. The present application and Taylor series expansion are used to extend downward by 200m (20 times the point spacing). The following is obtained: Figure 6 From the comparison results shown, it can be seen that the downward continuation method of potential field data provided in this application obtains better values, which are more consistent with theoretical values, and the downward continuation of potential field data has higher accuracy. Figure 4 and Figure 6 The true value in is the data after the label is extended downward, and the PINN downward extended predicted value is the data after the prediction is extended downward.

[0082] The present application also provides an application scenario, which applies the above-mentioned method for downward extension of potential field data. Specifically: the method for downward extension of potential field data provided in this embodiment can be applied in geophysical geological resource exploration scenarios. The geophysical geological resource exploration scenario includes a data acquisition link and a potential field data downward extension link; the measured potential field data enters the potential field data downward extension link from the data acquisition link, and obtains the corresponding predicted downward extended data through human-computer collaboration. The method for downward extension of potential field data provided in this embodiment belongs to the potential field data downward extension link. Specifically, in the process of the potential field data downward extension link for the measured potential field data, the measured potential field data can be input into a trained physical information neural network to obtain predicted downward extended data.

[0083] This application uses the measured potential field data as input, imports the physical information neural network for training, and uses the computer to automatically find the best downward extension function to obtain a better downward extension model. At the same time, the formulated physical constraints are used to ensure that the output downward extension value conforms to the physical laws, thereby realizing the accurate downward extension of the potential field data. Compared with the traditional integral iteration method, Taylor series expansion, and Tikhonov regularization method, the physical information neural network ignores the difference between the frequency domain and the space domain, and there is no need to consider operations such as data transformation, which greatly reduces the complexity of downward extension. This application has higher resolution and more stable results, and can effectively remove noise in noisy data to obtain better results. In the case of divergence of traditional methods in the extension depth with large point spacing, this application can obtain better results. It is conducive to promotion and popularization, and has practical significance for improving the accuracy of potential field exploration interpretation.

[0084] Based on the same inventive concept, the embodiment of the present application also provides a potential field data downward extension device for implementing the potential field data downward extension method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more potential field data downward extension device embodiments provided below can refer to the limitations of the potential field data downward extension method above, and will not be repeated here.

[0085] In an exemplary embodiment, Figure 7 As shown, a device for downward extension of potential field data is provided, comprising:

[0086] The measured potential field data acquisition module T1 is used to: acquire the measured potential field data; the measured potential field data is gravity field data or magnetic field data.

[0087] The downward extended data prediction module T2 is used to: input the measured potential field data into the trained physical information neural network to obtain the predicted downward extended data; the trained physical information neural network is trained by importing the sample potential field data into the physical information neural network containing the total loss function; the total loss function includes partial differential structure loss, boundary value loss and true value loss; when the predicted downward extended data is gravity field data, the partial differential structure loss is the error between the Laplace operator and the gravity product of the predicted downward extended data; the gravity product is the product of the gravity field and the density field; when the predicted downward extended data is magnetic field data, the partial differential structure loss is the error between the Laplace operator and the magnetic product of the predicted downward extended data; the magnetic product is the product of the magnetic field and the magnetization intensity; the boundary value loss is calculated by predicting the downward extended data; the true value loss is the error between the predicted downward extended data and the label downward extended data corresponding to the predicted downward extended data.

[0088] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store bit field data downward extension processing data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for downward extension of bit field data is implemented.

[0089] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0090] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0091] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0092] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0093] Those of ordinary skill 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 the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0094] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0095] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0096] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for downward extension of potential field data, characterized in that: The method for downward extension of the potential field data comprises: Acquiring measured potential field data; the measured potential field data is gravity field data or magnetic field data; The measured potential field data are input into a trained physical information neural network to obtain predicted downward extended data; the trained physical information neural network is trained by importing sample potential field data into a physical information neural network containing a total loss function; the total loss function includes partial differential structure loss, boundary value loss and true value loss; when the predicted downward extended data are gravity field data, the partial differential structure loss is the error between the Laplace operator and the gravity product of the predicted downward extended data; the gravity product is the product of the gravity field and the density field; when the predicted downward extended data are magnetic field data, the partial differential structure loss is the error between the Laplace operator and the magnetic product of the predicted downward extended data; the magnetic product is the product of the magnetic field and the magnetization intensity; the boundary value loss is calculated by predicting the downward extended data; the true value loss is the error between the predicted downward extended data and the label downward extended data corresponding to the predicted downward extended data.

2. The method for downward extension of potential field data according to claim 1, characterized in that: Before inputting the measured potential field data into the trained physical information neural network to obtain the predicted downward extended data, the method further includes: training the physical information neural network, specifically including: Acquire a sample data set; the sample data set includes a number of sample potential field data and data after the label corresponding to each sample potential field data is extended downward; the sample potential field data is gravity field data or magnetic field data; Input the sample potential field data into the physical information neural network to obtain the predicted downward extended data; Calculate the total loss function value based on the predicted downward extended data and the sample potential field data; Determine whether the set stop condition is met. If not, update the weights in the physical information neural network according to the total loss function value, and return to the step of "inputting the sample potential field data into the physical information neural network". If so, end the training and obtain the trained physical information neural network.

3. The method for downward extension of potential field data according to claim 1, characterized in that: When the predicted downward extension data is gravity field data, the calculation formula of partial differential structure loss is as follows: Among them, L PDE1 is the partial differential structure loss corresponding to the gravity field data, g(x) is the gravity field data, is the Laplace operator of the gravity field data, ρ is the density, i represents the data after the i-th prediction is extended downward, i=1, 2, 3…N, N is the total number of data after the prediction is extended downward.

4. The method for downward extension of potential field data according to claim 1, characterized in that: When the predicted downward extension data is gravity field data, the calculation formula of boundary value loss is as follows: Among them, L BC1 is the boundary value loss corresponding to the gravity field, is the second-order partial derivative of the gravity field data g(x) with respect to the density ρ.

5. The method for downward extension of potential field data according to claim 1, characterized in that: The calculation formula of the true value loss is as follows: Among them, L Data is the true value loss, Observed iData The label corresponding to the i-th prediction after the downward extension of the data is extended downward, Predicted iData is the data after the i-th prediction is extended downward, i represents the data after the i-th prediction is extended downward, i=1, 2, 3...N, N is the total number of data after the prediction is extended downward.

6. The method for downward extension of potential field data according to claim 2, characterized in that: The stopping condition is set as the current number of iterations reaches the maximum number of iterations or the total loss function value is less than the set error value.

7. The method for downward extension of potential field data according to claim 1, characterized in that: The physical information neural network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, a fourth hidden layer and an output layer which are connected in sequence.

8. A device for downward extension of potential field data, characterized in that: The bit field data downward extension device comprises: The measured potential field data acquisition module is used to: acquire the measured potential field data; the measured potential field data is gravity field data or magnetic field data; The downward extended data prediction module is used for: inputting the measured potential field data into the trained physical information neural network to obtain the predicted downward extended data; the trained physical information neural network is trained by importing the sample potential field data into the physical information neural network containing the total loss function; the total loss function includes partial differential structure loss, boundary value loss and true value loss; when the predicted downward extended data is gravity field data, the partial differential structure loss is the error between the Laplace operator and the gravity product of the predicted downward extended data; the gravity product is the product of the gravity field and the density field; when the predicted downward extended data is magnetic field data, the partial differential structure loss is the error between the Laplace operator and the magnetic product of the predicted downward extended data; the magnetic product is the product of the magnetic field and the magnetization intensity; the boundary value loss is calculated by predicting the downward extended data; the true value loss is the error between the predicted downward extended data and the label downward extended data corresponding to the predicted downward extended data.

9. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method for downward extension of bit field data according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for downward extension of the bit field data described in any one of claims 1 to 7 is implemented.

Images