Joint gravity-magnetic inversion method, device and electronic equipment

Abstract

The application provides a gravity-magnetic joint inversion method and device and electronic equipment, and relates to the technical field of geological exploration, and comprises the following steps: obtaining surface observation data of a target underground medium; processing the surface observation data by using a target neural network model to obtain gravity-magnetic characteristic data of the target underground medium; the target neural network model is a neural network model based on gravity forward simulation constraints and magnetic forward simulation constraints; and remodeling the gravity-magnetic characteristic data according to a preset dimension to obtain a magnetization intensity model and a gravity density model of the target underground medium. The method is a data-driven artificial intelligence processing technology, and can realize accurate mapping from surface observation data to underground attribute distribution based on data-driven training. Moreover, the target neural network model is a model based on physical principle constraints, which is more robust for gravity and magnetic field inversion problems, and effectively alleviates the technical problem of poor accuracy of the inversion result of the existing gravity-magnetic inversion method.

Description

Technical Field

[0001] This invention relates to the technical field of geological exploration, and in particular to a gravity and magnetic joint inversion method, apparatus and electronic equipment. Background Technology

[0002] Existing gravity and magnetic inversion methods are mainly based on single or joint inversion of models. They involve least-squares fitting of potential field data generated by forward modeling with observed potential field data, and processing the data by solving a minimization problem.

[0003] However, single gravity or magnetic field inversion is typically limited to a single type of parameter, lacking more comprehensive constraints on the model. Furthermore, due to insufficient observational data and the influence of noise, it is a typical ill-posed problem, leading to non-unique and unstable inversion results. Conventional joint gravity and magnetic field inversion models are constructed based on mathematical physics equations, whose gradient functions exhibit strong nonlinearity. The algorithms commonly used to solve for model parameters—such as solving simplified, discretized, underdetermined linear equations or algebraic iteration methods—are unstable, resulting in significant deviations from the true values. In summary, existing gravity and magnetic field inversion methods suffer from poor accuracy. Summary of the Invention

[0004] The purpose of this invention is to provide a gravity and magnetic field combined inversion method, apparatus and electronic device to alleviate the technical problem of poor inversion accuracy in existing gravity and magnetic field inversion methods.

[0005] In a first aspect, the present invention provides a gravity and magnetic joint inversion method, comprising: acquiring surface observation data of a target subsurface medium; wherein the surface observation data includes gravity observation data and magnetic observation data; processing the surface observation data using a target neural network model to obtain gravity and magnetic characteristic data of the target subsurface medium; wherein the target neural network model is a neural network model based on gravity forward modeling constraints and magnetic forward modeling constraints; reshaping the gravity and magnetic characteristic data according to a preset dimension to obtain a magnetization intensity model and a gravity density model of the target subsurface medium; wherein the magnetization intensity model is used to characterize the magnetization intensity distribution of the target subsurface medium, and the gravity density model is used to characterize the gravity density distribution of the target subsurface medium.

[0006] In an optional implementation, the method further includes: acquiring a training sample set; wherein the training sample set includes multiple training samples, each training sample including: a true magnetization model, a true gravity density model, and corresponding surface observation data of the sample subsurface medium; inputting the surface observation data of the sample subsurface medium into an initial neural network model, and using the corresponding true magnetization model and true gravity density model as labels for the surface observation data of the sample subsurface medium to obtain predicted gravity and magnetic feature data of the sample subsurface medium; reshaping the predicted gravity and magnetic feature data according to a preset dimension to obtain the predicted magnetization model and predicted gravity density model of the sample subsurface medium; calculating the function value of a loss function based on a preset number of training samples and the predicted magnetization model and predicted gravity density model corresponding to each training sample; training the initial neural network model based on the function value of the loss function to obtain the target neural network model.

[0007] In an optional implementation, the loss function is expressed as: Where N represents the number of training samples in the batch, y_pred k Let y_pred represent the prediction medium model for the k-th training sample. k =(m′ g ,m′ m ) k ,m′ g This represents the predicted gravity density model, m′ m This represents the model for predicting magnetization, y_ture k Let y_ture represent the true medium model of the k-th training sample. k =(m g ,m m ) k m g Represents the true gravity density model, m m This represents the true magnetization model, where λ represents the preset regularization parameter, and d... k This represents the surface observation data of the k-th training sample. G1 represents the first discretized matrix characterized by the kernel function modeled by gravity forward modeling, and G2 represents the second discretized matrix characterized by the kernel function modeled by magnetic forward modeling.

[0008] In an optional implementation, the first discretization matrix is ​​solved using the following formula: Where (x,y,z) represents the location coordinates of the i-th surface observation point, i takes values ​​from 1 to 1, and 1 represents the total number of surface observation points. The sample subsurface medium is divided into J subsurface pixels according to the preset dimensions, and (ξ,η,ζ) represents the location coordinates of the j-th subsurface pixel, j takes values ​​from 1 to J. 1,ijThis represents the element in the i-th row and j-th column of the first discretization matrix G1; the second discretization matrix is ​​solved using the following formula: G 2,ij =G x cosIcosA+G y cosIsinA+G z sinI; where, I represents the local magnetic inclination, A represents the angle between the strike of the geological body and magnetic north, μ0 represents the free magnetic permeability, v represents the subsurface integration region, and r represents the distance between the i-th surface observation point and the j-th subsurface pixel, r = [(ξ-x)]. 2 +(η-y) 2 +(ζ-z) 2 ] 1 / 2 H x H represents the component of the Earth's background magnetic field intensity along the x-axis in a Cartesian coordinate system. y H represents the component of the Earth's background magnetic field intensity along the y-axis in a Cartesian coordinate system. z This represents the component of the Earth's background magnetic field intensity along the z-axis in a spatial rectangular coordinate system.

[0009] In optional implementations, the distribution of anomalies in the underground medium of the training sample set includes the following types: single rectangular anomaly, two rectangular anomalies distributed from left to right, two rectangular anomalies distributed from top to bottom, and stepped anomalies.

[0010] In an optional implementation, the target neural network model includes: a first input layer, a combination of at least one first convolutional layer and a first maximum pooling layer, a first flattening layer, a first fully connected layer, a second input layer, a combination of at least one second convolutional layer and a second maximum pooling layer, a second flattening layer, a second fully connected layer, and a splicing layer; the first input layer, the combination of at least one first convolutional layer and a first maximum pooling layer, the first flattening layer, and the first fully connected layer are sequentially connected; the second input layer, the combination of at least one second convolutional layer and a second maximum pooling layer, the second flattening layer, and the second fully connected layer are sequentially connected; the outputs of the first fully connected layer and the second fully connected layer are both connected to the splicing layer; the first input layer is used to receive the gravity observation data, the second input layer is used to receive the magnetic observation data, and the splicing layer is used to output the gravity and magnetic feature data.

[0011] In an optional implementation, the output data of the stitching layer is a one-dimensional vector, and the number of elements in the one-dimensional vector is twice the total number of underground pixels.

[0012] Secondly, the present invention provides a gravity and magnetic joint inversion device, comprising: a first acquisition module for acquiring surface observation data of a target subsurface medium; wherein the surface observation data includes gravity observation data and magnetic observation data; a processing module for processing the surface observation data using a target neural network model to obtain gravity and magnetic characteristic data of the target subsurface medium; wherein the target neural network model is a neural network model based on gravity forward modeling constraints and magnetic forward modeling constraints; and a first reshaping module for reshaping the gravity and magnetic characteristic data according to a preset dimension to obtain a magnetization intensity model and a gravity density model of the target subsurface medium; wherein the magnetization intensity model is used to characterize the magnetization intensity distribution of the target subsurface medium, and the gravity density model is used to characterize the gravity density distribution of the target subsurface medium.

[0013] Thirdly, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the steps of the gravity and magnetic joint inversion method described in any of the foregoing embodiments.

[0014] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the gravity and magnetic joint inversion method described in any of the foregoing embodiments.

[0015] The gravity and magnetic field joint inversion method provided by this invention relies on data-driven artificial intelligence processing technology, rather than traditional inversion methods that heavily depend on prior knowledge and assumptions. Based on data-driven training, it can achieve an accurate mapping from surface observation data to subsurface attribute distribution. Furthermore, the target neural network model used in this invention is a neural network model based on gravity forward modeling and magnetic forward modeling constraints. Therefore, the neural network can capture more complex data features, which not only improves the model's accuracy but also ensures that the model's prediction results conform to physical principles, preventing over-reliance on data-driven training. This makes it more robust to gravity and magnetic field inversion problems, effectively alleviating the technical problem of poor inversion accuracy in existing gravity and magnetic field inversion methods. Attached Figure Description

[0016] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1A flowchart of a gravity and magnetic field joint inversion method provided in an embodiment of the present invention;

[0018] Figure 2 This is a schematic diagram of the gravity density model corresponding to the anomalies in the underground medium of the sample under four spatial distributions;

[0019] Figure 3 This is a schematic diagram of the magnetization intensity model corresponding to the anomalies in the underground medium of the sample under four spatial distributions;

[0020] Figure 4 This is a schematic diagram of the network structure of a target neural network model provided in an embodiment of the present invention;

[0021] Figure 5 This is a data flow diagram illustrating how a target neural network model processes surface observation data, as provided in an embodiment of the present invention.

[0022] Figure 6 This is a schematic diagram of the inversion results of the gravity density model;

[0023] Figure 7 This is a schematic diagram of the inversion results of the magnetization intensity model;

[0024] Figure 8 This diagram illustrates how the loss function of a neural network model changes with the number of iterations during training.

[0025] Figure 9 This is a comparison chart of the inversion results of the method of this invention and the joint inversion method of gravity and magnetic field cross gradient;

[0026] Figure 10 A functional block diagram of a gravity and magnetic field joint inversion device provided in an embodiment of the present invention;

[0027] Figure 11 This is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0029] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0030] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0031] Example 1

[0032] Figure 1 A flowchart of a gravity and magnetic field joint inversion method provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method specifically includes the following steps:

[0033] Step S102: Obtain surface observation data of the target subsurface medium.

[0034] Specifically, to invert the medium model (magnetization intensity model and gravity density model) of the target subsurface medium, this embodiment of the invention requires surface observation data of the target subsurface medium. This surface observation data includes gravity observation data and magnetic observation data. The surface observation data is obtained from dedicated monitoring equipment located at surface observation points within the work area where the target subsurface medium is located. This embodiment of the invention does not specifically limit the number of surface observation points; users can set them according to actual needs.

[0035] Step S104: Process the surface observation data using the target neural network model to obtain gravity and magnetic characteristic data of the target subsurface medium.

[0036] The target neural network model is a neural network model based on gravity forward modeling constraints and magnetic forward modeling constraints. In other words, the method provided in this embodiment of the invention uses a neural network constrained by physical principles to jointly invert gravity and magnetic field data. Therefore, training the neural network model with a large amount of sample data ensures that the model's output predictions conform to physical principles and guarantees the accuracy of the model's output results.

[0037] In this embodiment of the invention, the output of the target neural network model is the gravity and magnetic feature data of the target underground medium, which is data that characterizes the gravity and magnetic field features of the target underground medium.

[0038] Step S106: Reshape the gravity and magnetic feature data according to a preset dimension to obtain the magnetization intensity model and gravity density model of the target underground medium.

[0039] Gravity and magnetic feature data are combined data of gravity and magnetic field characteristics. In this embodiment of the invention, the gravity and magnetic feature data are one-dimensional vectors. Therefore, in order to obtain the magnetization intensity distribution data and gravity density distribution data of the target underground medium, it is necessary to reshape the gravity and magnetic feature data of the target underground medium according to a preset dimension, so as to obtain the magnetization intensity model and gravity density model of the target underground medium. Among them, the magnetization intensity model is used to characterize the magnetization intensity distribution of the target underground medium, and the gravity density model is used to characterize the gravity density distribution of the target underground medium.

[0040] In this embodiment of the invention, the preset dimension is two-dimensional. This embodiment does not specifically limit the value of the preset dimension; users can set it according to their actual needs. For example, if there are 1600 data points in the gravity and magnetic feature data, it means that there are 800 magnetization intensity data points and 800 gravity density data points. That is, the underground medium is divided into 800 pixels, each corresponding to a magnetization intensity and a gravity density. If the preset dimension is 20×40, then the above two sets of 800 data points need to be reshaped according to the 20×40 dimension to obtain the two-dimensional magnetization intensity model and gravity density model of the target underground medium.

[0041] The gravity and magnetic field joint inversion method provided in this invention relies on data-driven artificial intelligence processing technology, rather than traditional inversion methods that heavily depend on prior knowledge and assumptions. Based on data-driven training, it can achieve an accurate mapping from surface observation data to subsurface attribute distribution. Furthermore, the target neural network model used in this invention is a neural network model based on gravity forward modeling constraints and magnetic forward modeling constraints. Therefore, the neural network can capture more complex data features, which not only improves the model's accuracy but also ensures that the model's prediction results conform to physical principles, preventing over-reliance on data-driven training. This makes it more robust to gravity and magnetic field inversion problems, effectively alleviating the technical problem of poor inversion accuracy in existing gravity and magnetic field inversion methods.

[0042] In one optional implementation, the method of this embodiment further includes the following steps:

[0043] Step S201: Obtain the training sample set.

[0044] The training sample set includes multiple training samples, each of which includes: a true magnetization model of the sample's underground medium, a true gravity density model, and corresponding surface observation data.

[0045] Specifically, when constructing training samples, the profile needs to be placed on the Earth's surface. For example, the total profile length is 1000 meters, and observation points are evenly spaced at 10-meter intervals, resulting in 101 observation points. The subsurface medium model covers an area of ​​1000 meters by 500 meters, and the subsurface two-dimensional space is discretized into 800 square cells (20 rows × 40 columns), each with a side length of 25 meters. It is assumed that each square cell has a uniformly distributed gravity density and magnetization, and that the magnetic anomaly is located within a uniform, non-magnetic substrate, neglecting the effects of demagnetization or remanence. Additionally, the magnetic inclination is set to 90° (exemplary value), and the magnetic declination is also set to 90° (exemplary value). Based on these configurations, multiple training samples can be obtained.

[0046] To enhance the diversity of the training sample set, anomalies can be categorized into different types based on their spatial distribution. Optionally, the distribution of anomalies within the subsurface medium of the training sample set includes the following types: single rectangular anomalies, two rectangular anomalies distributed from left to right, two rectangular anomalies distributed from top to bottom, and stepped anomalies. Figure 2 This is a schematic diagram of the gravity density models corresponding to four spatial distributions of anomalies within the sample's subsurface medium. Figure 3 This is a schematic diagram of the magnetization intensity model corresponding to the anomalies in the underground medium of the sample under four spatial distributions.

[0047] Furthermore, to make the feature extraction of the neural network model more obvious, the anomalies in the underground medium of the sample are made of pure metal with high density; for example, the density of the anomalies can be 10 g / cm³. 3 The magnetization is 50 A / m.

[0048] Step S202: Input the sample surface observation data of the sample subsurface medium into the initial neural network model, and use the corresponding real magnetization intensity model and real gravity density model as labels for the sample surface observation data to obtain the predicted gravity and magnetic characteristic data of the sample subsurface medium.

[0049] When training the model, the initial neural network model is used to process the surface observation data of the labeled samples. The labels of the samples are the true magnetization intensity model and the true gravity density model in the training samples, so as to ensure that the model can effectively fit the labeled data. The output of the neural network model is the gravity and magnetic characteristic data predicted based on the surface observation data of the sample subsurface medium.

[0050] Step S203: The predicted gravity and magnetic characteristic data are reshaped according to a preset dimension to obtain the predicted magnetization intensity model and the predicted gravity density model of the sample underground medium.

[0051] The method for dimensional reshaping the predicted gravity and magnetic characteristic data is the same as step S106 above, and will not be repeated here. Please refer to the above for details. After step S203 is completed, the predicted magnetization intensity model and the predicted gravity density model inverted from the sample surface observation data will be obtained.

[0052] Step S204: Based on a preset number of training samples, the prediction magnetization model and the prediction gravity density model corresponding to each training sample, calculate the function value of the loss function.

[0053] To ensure the stability of the training process, this embodiment of the invention adopts a batch training method. That is, each batch has a preset number of training samples. After obtaining the predicted magnetization intensity model and predicted gravity density model corresponding to each training sample through the neural network model, the function value of the loss function is calculated based on the preset number of sets of data. Each set of data consists of training samples and the corresponding model output results (predicted magnetization intensity model and predicted gravity density model).

[0054] Step S205: Train the initial neural network model based on the function value of the loss function to obtain the target neural network model.

[0055] Optionally, in this embodiment of the invention, the gradient descent algorithm is used to train the initial neural network model. Training can be stopped after the training reaches a preset number of iterations or the value of the loss function is less than a preset threshold, thereby obtaining the target neural network model.

[0056] Although the method provided in this embodiment of the invention requires a lot of computing resources during network training, once the neural network model is effectively trained, it will greatly reduce the amount of computation required for inversion. That is, by inputting surface observation data, the corresponding underground model with physical property distribution can be quickly calculated to achieve the inversion goal.

[0057] In one alternative implementation, the loss function is expressed as:

[0058] Where N represents the number of training samples in the batch, y_pred k Let y_pred represent the prediction medium model for the k-th training sample. k =(m′ g ,m′ m ) k ,m′ g This represents the predicted gravity density model, m′ m This represents the model for predicting magnetization, y_ture k Let y_ture represent the true medium model of the k-th training sample. k =(m g ,m m) k m g Represents the true gravity density model, m m This represents the true magnetization model, where λ represents the preset regularization parameter, and d... k This represents the surface observation data of the k-th training sample. G1 represents the first discretized matrix characterized by the kernel function modeled by gravity forward modeling, and G2 represents the second discretized matrix characterized by the kernel function modeled by magnetic forward modeling.

[0059] As can be seen from the above expression of the loss function, the loss function consists of two parts. Combining the meaning of each parameter in the expression, we can see that the first part is to calculate the mean square error between the predicted medium model and the real medium model. This part reflects the overall performance of the model and can ensure that the model can effectively fit the label data. The second part is to calculate the mean square error between the sample surface observation data and the data calculated based on the predicted medium model.

[0060] Specifically, the relationship between the known subsurface medium model and surface observation data can be expressed by the following formula: d = Gm, which can be expanded to obtain: Where, d g Represents gravity observation data, d g =G1m g d m d represents magnetic observation data. m =G2m m d g =(d g1 ,d g2 ,…,d gI ) T d m =(d m1 ,d m2 ,…,d mI ) T m g =(m g1 ,m g2 ,…,m gI ) T m m =(m m1 ,m m2 ,…,m mI ) T I represents the total number of surface observation points, and the dimensions of the first discretization matrix G1 and the second discretization matrix G2 are both I×J, where J represents the total number of underground pixels. Therefore, in the second part of the loss function, Gy_pred kIn other words, it is equivalent to forward modeling based on physical constraints (gravity and magnetic field). The process of generating data based on the prediction medium model is entirely based on physical forward modeling calculation. Therefore, by incorporating this into the calculation of the loss function, the neural network model can capture more complex data relationships, which not only improves the accuracy of the model, but also ensures that the prediction conforms to physical principles, thereby improving the overall performance and reliability of the neural network.

[0061] Furthermore, to balance the model's performance, the first and second parts of the loss function have equal weights and are linearly combined. Therefore, this loss function can prevent the model from becoming overly reliant on data-driven training, and under physical constraints, the model can better meet the requirements of the actual problem.

[0062] In one alternative implementation, the first discretization matrix is ​​solved using the following formula: Where (x,y,z) represents the location coordinates of the i-th surface observation point, i takes values ​​from 1 to I, and I represents the total number of surface observation points. The sample subsurface medium is divided into J subsurface pixels according to a preset dimension, and (ξ,η,ζ) represents the location coordinates of the j-th subsurface pixel, j takes values ​​from 1 to J. 1,ij This represents the element in the i-th row and j-th column of the first discretized matrix G1.

[0063] The second discretization matrix is ​​solved using the following formula: G 2,ij =G x cosIcosA+G y cosIsinA+G z sinI; where, I represents the local magnetic inclination, A represents the angle between the strike of the geological body and magnetic north, μ0 represents the free magnetic permeability, v represents the subsurface integration region, and r represents the distance between the i-th surface observation point and the j-th subsurface pixel, r = [(ξ-x)]. 2 +(η-y) 2 +(ζ-z) 2 ] 1 / 2 H x H represents the component of the Earth's background magnetic field intensity along the x-axis in a Cartesian coordinate system. y H represents the component of the Earth's background magnetic field intensity along the y-axis in a Cartesian coordinate system. z This represents the component of the Earth's background magnetic field intensity along the z-axis in a spatial rectangular coordinate system.

[0064] In one alternative implementation, the target neural network model includes: a first input layer, a combination of at least one first convolutional layer and a first maximum pooling layer, a first flattening layer, a first fully connected layer, a second input layer, a combination of at least one second convolutional layer and a second maximum pooling layer, a second flattening layer, a second fully connected layer, and a splicing layer.

[0065] A first input layer, a combination of at least one first convolutional layer and a first maximum pooling layer, a first flattening layer, and a first fully connected layer are connected in sequence.

[0066] A second input layer, a combination of at least one second convolutional layer and a second maximum pooling layer, a second flattening layer, and a second fully connected layer are connected in sequence.

[0067] The outputs of both the first fully connected layer and the second fully connected layer are connected to the splicing layer.

[0068] The first input layer is used to receive gravity observation data, the second input layer is used to receive magnetic observation data, and the stitching layer is used to output gravity and magnetic feature data.

[0069] Specifically, the target neural network model in this embodiment of the invention is used for the prediction (regression) task of spatial sequence data, and is referred to as GMNet below. Figure 4 This is a schematic diagram of the network structure of a target neural network model provided in an embodiment of the present invention. (Reference) Figure 4 The detailed description of the GMNet architecture is as follows:

[0070] The model has two input layers: one for receiving gravity observation data and the other for receiving magnetic observation data. Referring to the example of training samples above, if 101 observation points are set on the Earth's surface, the input layer receives spatial sequence data of size (101, 1).

[0071] After receiving gravity observation data and magnetic observation data, the model needs to process both types of data simultaneously. Specifically, at least one combination of a first convolutional layer and a first maximum pooling layer, a first flattening layer and a first fully connected layer are used to process the gravity observation data, and at least one combination of a second convolutional layer and a second maximum pooling layer, a second flattening layer and a second fully connected layer are used to process the magnetic observation data.

[0072] The following section uses one type of observation data processing as an example to specifically introduce the convolutional layer (Conv), max pooling layer (MaxPooling), flattening layer (Flatten), and dense fully connected layer (Dense). The convolutional layer and max pooling layer are the core feature extraction parts of the model. Optionally, this embodiment of the invention uses a 3*3 convolutional kernel, meaning that each kernel considers three consecutive spatial strides. This kernel size selection aims to capture local features in the spatial sequence, such as variations and trends. The convolutional layer is used to learn various local features, and the subsequent pooling layer performs a 2x downsampling, reducing the dimensionality of the feature map while preserving basic information. Optionally, the model includes a combination of three convolutional layers and max pooling layers, with each convolutional layer extracting higher-level features based on the output of the previous layer.

[0073] The flattening layer flattens the feature map generated by the last max pooling layer into a one-dimensional vector for input into the fully connected layer. The fully connected layer is used for further feature extraction and nonlinear transformations. The model consists of two fully connected layers, for example, with 256 and 800 neurons respectively; that is, the number of neurons in the last layer equals the number of sub-pixels. Furthermore, each neuron is introduced with nonlinearity through a rectified linear unit (ReLU) activation function, thereby enhancing the model's ability to learn complex data relationships.

[0074] The concatenation layer (tf.concat) does not use an activation function because its purpose is to directly concatenate the outputs of the first and second fully connected layers to output a regression prediction. This architecture is designed based on best practices and empirical knowledge in deep learning, aiming to capture key features from the input data and achieve high-performance regression predictions. Optionally, the output data of the concatenation layer is a one-dimensional vector, and the number of elements in the one-dimensional vector is twice the total number of subsurface pixels. That is, if there are 800 subsurface pixels, the one-dimensional vector output by the concatenation layer will have 1600 elements.

[0075] When using GMNet for machine learning, Figure 5 This is a data flow diagram illustrating how a target neural network model processes surface observation data according to an embodiment of the present invention. (Refer to...) Figure 5The first convolutional layers are conv1D_g1 and conv1D_m1, each containing 32 kernels of 3x3 size. The activation function is ReLU, and the padding method is 'same'. The first pooling layers are max_pooling1D_g1 and max_pooling1D_m1, performing max pooling with a pooling window size of 2, padding method 'same', and a stride of 2. Similarly, there are two more convolutional layers and two more pooling layers. The second and third convolutional layers have 64 and 128 kernels respectively, with all other parameters identical to the first convolutional layer. The pooling layer parameters are also identical to the first pooling layer. The flatten layer flattens the feature maps into a one-dimensional vector. The next two fully connected layers, Dense, have 256 and 800 neurons respectively. The final concatenation layer, tf.concat, merges the gravity and magnetic feature data into a single output. The model's output is a one-dimensional vector; by reshaping the output vector, the inversion result can be obtained.

[0076] In summary, traditional model-driven techniques are based solely on the model itself, and the inversion process is limited by insufficient observation data and affected by noise, resulting in strong ill-posedness and non-unique, unstable inversion results. Compared with conventional model-driven techniques, the data-driven deep learning technique based on physical constraints proposed in this invention has the following advantages: First, the loss function is constructed from the mean squared error between the predicted medium model and the real medium model, reflecting the overall performance of the model and ensuring that the model effectively fits the labeled data. Second, the loss function further incorporates the physical constraints of the data-driven convolutional neural network, ensuring that the model effectively predicts within a specific feature space.

[0077] To verify the performance of the method provided in this embodiment of the invention, the following simulation was performed: the model was trained using 1300 training samples, and its performance was evaluated using 68 validation samples. To accelerate convergence, the Adam optimizer was used with an initial learning rate set to 0.01, combined with a learning rate decay at a rate of 0.00001 to effectively fine-tune the model parameters. After each convolution operation, a Rectified Linear Unit (ReLU) activation function was used to introduce non-linearity, enabling the network to better capture complex patterns and features in the data.

[0078] To ensure a stable training process, the training samples were divided into mini-batches, each containing 16 samples. This method not only reduced memory consumption but also improved the efficiency of gradient descent. A total of 3000 training iterations were performed. Figure 6 This is a schematic diagram of the inversion results from the gravity density model. Figure 7This is a schematic diagram of the inversion results of the magnetization intensity model. Figure 6 and Figure 7 This demonstrates that the method provided in the embodiments of the present invention obtained joint gravity and magnetic inversion results on the verification dataset. Figure 6 and Figure 7 In the diagram, the actual location of the anomaly is highlighted with a thick-lined box. The inversion results show that most models achieve good agreement on the depth and location of the subsurface anomaly. The inverted values ​​of the anomaly density and magnetization are very close to the actual values ​​(density 10 g / cm³). 3 (Magnetic saturation is 50 A / m).

[0079] Figure 8 This diagram illustrates how the loss function of a neural network model changes with the number of iterations during training. Figure 8 As can be seen, the loss function value shows a gradually decreasing trend, indicating that the method provided by the embodiments of the present invention has a stable learning rate.

[0080] To evaluate the inversion performance of the gravity and magnetic inversion method provided in this embodiment of the invention, the method was applied to an ideal underground pure metal model, which includes two elements with a density of 10 g / cm³. 3 An anomaly with a magnetization of 50 A / m was identified. The results of this invention's method were compared with those of the gravity and magnetic field cross-gradient joint inversion method (a purely model-driven joint inversion algorithm). In the specific implementation, only its two-dimensional form was used, and the subsurface model area and the number of anomalies were expanded. The initial inversion model was set to a density of 1 g / cm³. 3 An underground space with a magnetization of 1 A / m. In the comparative experiment, the gravity and magnetic field cross-gradient joint inversion method underwent 300 iterations, taking approximately 5 minutes. Figure 9 This is a comparison diagram of the inversion results of the method of this invention and the joint inversion method of gravity and magnetic field cross gradient. The thick-lined boxes mark the true location of the anomaly. Figure 9 View A shows the inversion results obtained by the joint inversion method based on cross gradients, while Figure 9 View b shows the results obtained using the method of this invention. Clearly, the inversion results obtained by the joint inversion method based on cross-gradients show lower density and magnetization values ​​than the true values, accompanied by ambiguity in the location of the anomalous body. In contrast, the method of this invention demonstrates effective inversion results in less than one second.

[0081] Example 2

[0082] This invention also provides a gravity and magnetic joint inversion device, which is mainly used to execute the gravity and magnetic joint inversion method provided in Embodiment 1 above. The gravity and magnetic joint inversion device provided in this invention will be described in detail below.

[0083] Figure 10 This is a functional block diagram of a gravity and magnetic field joint inversion device provided in an embodiment of the present invention, as shown below. Figure 10 As shown, the device mainly includes: a first acquisition module 10, a processing module 20, and a first reshaping module 30, wherein:

[0084] The first acquisition module 10 is used to acquire surface observation data of the target subsurface medium; wherein, the surface observation data includes gravity observation data and magnetic observation data.

[0085] The processing module 20 is used to process the surface observation data using the target neural network model to obtain the gravity and magnetic characteristics data of the target subsurface medium; wherein, the target neural network model is a neural network model based on gravity forward modeling constraints and magnetic forward modeling constraints.

[0086] The first reshaping module 30 is used to reshape the gravity and magnetic feature data according to a preset dimension to obtain the magnetization intensity model and gravity density model of the target underground medium; wherein, the magnetization intensity model is used to characterize the magnetization intensity distribution of the target underground medium, and the gravity density model is used to characterize the gravity density distribution of the target underground medium.

[0087] The gravity and magnetic field joint inversion device provided in this invention uses data-driven artificial intelligence processing technology, rather than traditional inversion methods that heavily rely on prior knowledge and assumptions. Based on data-driven training, it can achieve accurate mapping from surface observation data to subsurface attribute distribution. Furthermore, the target neural network model used in this invention is a neural network model based on gravity forward modeling constraints and magnetic forward modeling constraints. Therefore, the neural network can capture more complex data features, which not only improves the model's accuracy but also ensures that the model's prediction results conform to physical principles, preventing over-reliance on data-driven training. This makes it more robust to gravity and magnetic field inversion problems, effectively alleviating the technical problem of poor inversion accuracy in existing gravity and magnetic field inversion methods.

[0088] Optionally, the device further includes:

[0089] The second acquisition module is used to acquire a training sample set; wherein, the training sample set includes multiple training samples, and each training sample includes: the true magnetization intensity model of the sample's underground medium, the true gravity density model, and the corresponding sample surface observation data.

[0090] The input module is used to input the sample surface observation data of the sample subsurface medium into the initial neural network model, and use the corresponding real magnetization intensity model and real gravity density model as labels for the sample surface observation data to obtain the predicted gravity and magnetic characteristic data of the sample subsurface medium.

[0091] The second reshaping module is used to reshape the predicted gravity and magnetic characteristic data according to a preset dimension to obtain the predicted magnetization intensity model and the predicted gravity density model of the sample underground medium.

[0092] The calculation module is used to calculate the value of the loss function based on a preset number of training samples, the prediction magnetization model and the prediction gravity density model corresponding to each training sample.

[0093] The training module is used to train an initial neural network model based on the function value of the loss function to obtain the target neural network model.

[0094] Alternatively, the loss function can be expressed as: Where N represents the number of training samples in the batch, y_pred k Let y_pred represent the prediction medium model for the k-th training sample. k =(m′ g ,m′ m ) k ,m′ g This represents the predicted gravity density model, m′ m This represents the model for predicting magnetization, y_ture k Let y_ture represent the true medium model of the k-th training sample. k =(m g ,m m ) k m g Represents the true gravity density model, m m This represents the true magnetization model, where λ represents the preset regularization parameter, and d... k This represents the surface observation data of the k-th training sample. G1 represents the first discretized matrix characterized by the kernel function modeled by gravity forward modeling, and G2 represents the second discretized matrix characterized by the kernel function modeled by magnetic forward modeling.

[0095] Alternatively, the first discretization matrix is ​​solved using the following formula: Where (x,y,z) represents the location coordinates of the i-th surface observation point, i takes values ​​from 1 to I, and I represents the total number of surface observation points. The sample subsurface medium is divided into J subsurface pixels according to a preset dimension, and (ξ,η,ζ) represents the location coordinates of the j-th subsurface pixel, j takes values ​​from 1 to J. 1,ij This represents the element in the i-th row and j-th column of the first discretized matrix G1.

[0096] The second discretization matrix is ​​solved using the following formula: G 2,ij =G x cosIcosA+G y cosIsinA+Gz sinI; where, I represents the local magnetic inclination, A represents the angle between the strike of the geological body and magnetic north, μ0 represents the free magnetic permeability, v represents the subsurface integration region, and r represents the distance between the i-th surface observation point and the j-th subsurface pixel, r = [(ξ-x)]. 2 +(η-y) 2 +(ζ-z) 2 ] 1 / 2 H x H represents the component of the Earth's background magnetic field intensity along the x-axis in a Cartesian coordinate system. y H represents the component of the Earth's background magnetic field intensity along the y-axis in a Cartesian coordinate system. z This represents the component of the Earth's background magnetic field intensity along the z-axis in a spatial rectangular coordinate system.

[0097] Optionally, the distribution of anomalies in the underground medium of the training sample set includes the following types: single rectangular anomaly, two rectangular anomalies distributed from left to right, two rectangular anomalies distributed from top to bottom, and stepped anomalies.

[0098] Optionally, the target neural network model includes: a first input layer, a combination of at least one first convolutional layer and a first maximum pooling layer, a first flattening layer, a first fully connected layer, a second input layer, a combination of at least one second convolutional layer and a second maximum pooling layer, a second flattening layer, a second fully connected layer, and a splicing layer.

[0099] A first input layer, a combination of at least one first convolutional layer and a first max pooling layer, a first flattening layer and a first fully connected layer are connected in sequence; a second input layer, a combination of at least one second convolutional layer and a second max pooling layer, a second flattening layer and a second fully connected layer are connected in sequence; the outputs of the first fully connected layer and the second fully connected layer are both connected to the splicing layer.

[0100] The first input layer is used to receive gravity observation data, the second input layer is used to receive magnetic observation data, and the stitching layer is used to output gravity and magnetic feature data.

[0101] Optionally, the output data of the stitching layer is a one-dimensional vector, and the number of elements in the one-dimensional vector is twice the total number of underground pixels.

[0102] Example 3

[0103] See Figure 11This invention provides an electronic device, which includes a processor 60, a memory 61, a bus 62, and a communication interface 63. The processor 60, the communication interface 63, and the memory 61 are connected via the bus 62. The processor 60 is used to execute executable modules, such as computer programs, stored in the memory 61.

[0104] The memory 61 may include high-speed random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 63 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc.

[0105] Bus 62 can be an ISA bus, PCI bus, or EISA bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 11 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0106] The memory 61 is used to store programs. After receiving an execution instruction, the processor 60 executes the program. The method executed by the apparatus defined by the process disclosed in any of the foregoing embodiments of the present invention can be applied to the processor 60 or implemented by the processor 60.

[0107] Processor 60 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 60 or by instructions in software form. Processor 60 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 61. Processor 60 reads the information in memory 61 and, in conjunction with its hardware, completes the steps of the above method.

[0108] The computer program product of the gravity and magnetic field joint inversion method, apparatus and electronic device provided in the embodiments of the present invention includes a computer-readable storage medium storing non-volatile program code executable by a processor. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0109] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0110] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0111] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0112] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. In addition, the terms "first," "second," "third," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0113] Furthermore, terms such as "horizontal," "vertical," and "sag" do not imply that components must be absolutely horizontal or suspended, but rather that they can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal relative to "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0114] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for joint gravity and magnetic inversion, characterized in that, include: Acquire surface observation data of the target subsurface medium; wherein, the surface observation data includes: gravity observation data and magnetic observation data; The surface observation data are processed using a target neural network model to obtain gravity and magnetic characteristic data of the target subsurface medium; wherein, the target neural network model is a neural network model based on gravity forward modeling constraints and magnetic forward modeling constraints; The gravity and magnetic feature data are reshaped according to a preset dimension to obtain the magnetization intensity model and gravity density model of the target underground medium; wherein, the magnetization intensity model is used to characterize the magnetization intensity distribution of the target underground medium, and the gravity density model is used to characterize the gravity density distribution of the target underground medium. The method further includes: Obtain a training sample set; wherein the training sample set includes multiple training samples, each training sample including: a true magnetization intensity model of the sample's underground medium, a true gravity density model, and corresponding sample surface observation data; The sample surface observation data of the subsurface medium is input into the initial neural network model, and the corresponding real magnetization intensity model and real gravity density model are used as labels for the sample surface observation data to obtain the predicted gravity and magnetic characteristic data of the sample subsurface medium. The predicted gravity and magnetic feature data are reshaped according to a preset dimension to obtain the predicted magnetization intensity model and the predicted gravity density model of the underground medium of the sample. Based on a preset number of training samples, the prediction magnetization model and the prediction gravity density model corresponding to each training sample, the function value of the loss function is calculated. The initial neural network model is trained based on the function value of the loss function to obtain the target neural network model; The loss function is expressed as: ; in, This indicates the number of training samples in the batch. This represents the prediction medium model for the k-th training sample. , This represents a model for predicting gravity density. This represents the model for predicting magnetization. This represents the true medium model of the k-th training sample. , This represents a model of true gravity density. This represents the true magnetization model. This indicates the preset regularization parameters. This represents the surface observation data of the k-th training sample. , Let represent the first discretized matrix characterized by the kernel function modeled by gravity forward modeling. This represents the second discretized matrix characterized by the kernel function modeled by the magnetic forward modeling.

2. The gravity and magnetic field joint inversion method according to claim 1, characterized in that, The first discretization matrix is ​​solved using the following formula: ;in, This represents the location coordinates of the i-th surface observation point, where i ranges from 1 to... , This represents the total number of surface observation points. The sample subsurface medium is divided into J subsurface pixels according to the preset dimensions. This represents the location coordinates of the j-th underground pixel, where j ranges from 1 to J. Represents the first discretization matrix The element in the i-th row and j-th column; The second discretization matrix is ​​solved using the following formula: ;in, , , , Indicates the local magnetic inclination. Indicates the angle between the strike of the geological body and magnetic north. Represents the permeability of free space. Indicates the underground integration region. This represents the distance between the i-th surface observation point and the j-th underground pixel. , This represents the component of the Earth's background magnetic field intensity along the x-axis in a Cartesian coordinate system. This represents the component of the Earth's background magnetic field intensity along the y-axis in a Cartesian coordinate system. This represents the component of the Earth's background magnetic field intensity along the z-axis in a spatial rectangular coordinate system.

3. The gravity and magnetic field joint inversion method according to claim 1, characterized in that, The distribution of anomalies in the underground medium of the training sample set includes the following types: single rectangular anomaly, two rectangular anomalies distributed from left to right, two rectangular anomalies distributed from top to bottom, and stepped anomalies.

4. The gravity and magnetic field joint inversion method according to claim 1, characterized in that, The target neural network model includes: a first input layer, a combination of at least one first convolutional layer and a first maximum pooling layer, a first flattening layer, a first fully connected layer, a second input layer, a combination of at least one second convolutional layer and a second maximum pooling layer, a second flattening layer, a second fully connected layer, and a splicing layer; The first input layer, the combination of at least one first convolutional layer and a first max pooling layer, the first flattening layer and the first fully connected layer are connected in sequence; The second input layer, the combination of the at least one second convolutional layer and the second maximum pooling layer, the second flattening layer, and the second fully connected layer are connected in sequence. The outputs of the first fully connected layer and the second fully connected layer are both connected to the splicing layer; The first input layer is used to receive the gravity observation data, the second input layer is used to receive the magnetic observation data, and the splicing layer is used to output the gravity and magnetic feature data.

5. The gravity and magnetic field joint inversion method according to claim 4, characterized in that, The output data of the splicing layer is a one-dimensional vector, and the number of elements in the one-dimensional vector is twice the total number of underground pixels.

6. A gravity and magnetic field joint inversion device, characterized in that, include: The first acquisition module is used to acquire surface observation data of the target subsurface medium; wherein, the surface observation data includes: gravity observation data and magnetic observation data; The processing module is used to process the surface observation data using a target neural network model to obtain gravity and magnetic characteristic data of the target subsurface medium; wherein, the target neural network model is a neural network model based on gravity forward modeling constraints and magnetic forward modeling constraints; The first reshaping module is used to reshape the gravity and magnetic feature data according to a preset dimension to obtain the magnetization intensity model and gravity density model of the target underground medium; wherein, the magnetization intensity model is used to characterize the magnetization intensity distribution of the target underground medium, and the gravity density model is used to characterize the gravity density distribution of the target underground medium. The device also includes: The second acquisition module is used to acquire a training sample set; wherein, the training sample set includes multiple training samples, and each training sample includes: a true magnetization intensity model of the sample's underground medium, a true gravity density model, and corresponding sample surface observation data; The input module is used to input the sample surface observation data of the sample subsurface medium into the initial neural network model, and use the corresponding real magnetization intensity model and real gravity density model as labels for the sample surface observation data to obtain the predicted gravity and magnetic characteristic data of the sample subsurface medium. The second reshaping module is used to reshape the predicted gravity and magnetic feature data according to a preset dimension to obtain the predicted magnetization intensity model and the predicted gravity density model of the sample underground medium. The calculation module is used to calculate the function value of the loss function based on a preset number of training samples, the prediction magnetization intensity model and the prediction gravity density model corresponding to each training sample; The training module is used to train the initial neural network model based on the function value of the loss function to obtain the target neural network model; The loss function is expressed as: ; in, This indicates the number of training samples in the batch. This represents the prediction medium model for the k-th training sample. , This represents a model for predicting gravity density. This represents the model for predicting magnetization. This represents the true medium model of the k-th training sample. , This represents a model of true gravity density. This represents the true magnetization model. This indicates the preset regularization parameters. This represents the surface observation data of the k-th training sample. , Let represent the first discretized matrix characterized by the kernel function modeled by gravity forward modeling. This represents the second discretized matrix characterized by the kernel function modeled by the magnetic forward modeling.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the gravity and magnetic field joint inversion method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the gravity and magnetic joint inversion method according to any one of claims 1 to 5.

Images