A method, system, and medium for estimating the physical consistency of soil resistivity

By constructing a heterogeneous spatial map of the surface and subsurface and a graph convolutional network, and combining the spatial orientation of the strata with the electrical relationship of rocks, the problem of insufficient consistency in soil resistivity estimation was solved, and higher accuracy in three-dimensional resistivity prediction was achieved.

CN123113014APending Publication Date: 2026-11-17STATE GRID JIANGSU ELECTRIC POWER CO LTD RESEARCH INSTITUTE
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611409221.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-11
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to integrate the evolution of subsurface seepage caused by changes in surface hydrology with the anisotropic electrical characteristics of strata into the spatial information transmission and network training process in complex geological environments, resulting in insufficient consistency between the spatial distribution of soil resistivity and the actual physical evolution law.

Method used

By acquiring spatial feature data of the target area, a heterogeneous spatial map of the surface and subsurface is constructed. A graph convolutional network is used to predict the subsurface resistivity tensor. A physical constraint loss is constructed by combining the spatial orientation of the strata and the electrical relationship of the rocks. The network is trained by combining measured data to achieve the prediction of three-dimensional soil resistivity distribution.

Benefits of technology

This effectively solves the problem of insufficient physical consistency of three-dimensional resistivity prediction results caused by the lack of synergistic constraints between hydrological seepage evolution and the anisotropic rock electrical laws in existing soil resistivity estimation methods, and achieves higher physical consistency and accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123113014A_ABST
    Figure CN123113014A_ABST
Patent Text Reader

Abstract

This invention relates to the field of geophysical exploration technology, and particularly to a method, system, and medium for estimating the physical consistency of soil resistivity. The method includes: acquiring spatial characteristic data and measured resistivity data; constructing a heterogeneous spatial map of surface nodes and subsurface nodes; determining the seepage characteristics from surface nodes to subsurface nodes, determining cross-layer edge weights, and updating water saturation; inputting the heterogeneous spatial map into a graph convolutional network to obtain a predicted resistivity tensor; performing coordinate transformation based on the formation spatial orientation, and constructing a physical constraint loss based on water saturation and rock electrical relationships; training the graph convolutional network based on the physical constraint loss and data constraint loss to obtain a three-dimensional soil resistivity distribution. This invention effectively solves the problem of insufficient physical consistency in three-dimensional resistivity prediction results caused by the lack of synergistic constraints from hydrological seepage evolution and the anisotropic rock electrical laws of the formation in existing soil resistivity estimation methods.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration technology, and in particular to a method, system, and medium for estimating the physical consistency of soil resistivity. Background Technology

[0002] Soil resistivity characterizes the degree to which subsurface media impede the conduction of electric current. Its spatial distribution is influenced by factors such as water content, pore structure, and formation morphology, making it a crucial fundamental parameter for geophysical exploration and subsurface electrical property assessment. Existing technologies have employed neural networks to enhance the inversion capabilities of resistivity and formation parameters. For example, CN119129672A uses a physical-driven neural network to invert parameters such as formation resistivity and employs physical forward modeling error constraints for network training; CN119882088A substitutes the porosity and saturation output from the neural network into the Archie relation to obtain predicted resistivity and optimizes network parameters using the error between the predicted results and observed data.

[0003] However, existing technologies mainly use geophysical laws as constraints or forward relationships at the network output end, and it is still difficult to integrate the evolution of underground seepage caused by changes in surface hydrology and the anisotropic electrical characteristics of strata into the spatial information transmission and network training process. This results in insufficient consistency between the spatial distribution of resistivity obtained in complex geological environments and the actual physical evolution laws.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the general background of this disclosure and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] This invention provides a method, system, and medium for estimating the physical consistency of soil resistivity, which can effectively solve the problems in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for estimating the physical consistency of soil resistivity, the method comprising: Acquire spatial feature data of the target area and measured resistivity data of sample nodes, and construct a heterogeneous spatial map including surface nodes and underground nodes based on the spatial feature data; Based on the spatial feature data, the seepage characterization of the surface node to the underground node is determined, and the cross-layer edge weight of the heterogeneous spatial map is determined based on the seepage characterization and the water saturation of the underground node is updated. The heterogeneous spatial graph is input into a graph convolutional network for feature transfer to obtain the predicted resistivity tensor of the underground node; The predicted resistivity tensor is subjected to coordinate transformation based on the formation spatial attitude, and the physical constraint loss of the predicted resistivity tensor is constructed by combining the updated water saturation and rock electrical relationship. The graph convolutional network is trained based on the physical constraint loss and the data constraint loss determined by the predicted resistivity tensor and the measured resistivity data, and the three-dimensional soil resistivity distribution of the target area is obtained based on the trained graph convolutional network.

[0007] Furthermore, the spatial characteristic data includes elevation data, precipitation flux data, vertical permeability, porosity, node burial depth data, formation dip angle, formation azimuth angle, and initial water saturation.

[0008] Furthermore, the heterogeneous spatial map also includes surface-level edges connecting the surface nodes and underground-level edges connecting the underground nodes. The cross-level edges are constructed according to the spatial correspondence between the surface nodes and the underground nodes, and the surface-level edges and underground-level edges are constructed according to the spatial adjacency between nodes of the same level.

[0009] Furthermore, the hydraulic gradient between the surface node and the corresponding underground node is determined based on the precipitation flux data, elevation data and node burial depth data, and the dynamic seepage rate is determined in combination with the vertical permeability, and the dynamic seepage rate is used as the seepage characterization quantity. The dynamic seepage rate is nonlinearly mapped to obtain the cross-layer edge weights. The updated water saturation is determined based on the cumulative amount of the dynamic seepage rate and the initial water saturation.

[0010] Furthermore, the graph convolutional network performs weighted aggregation of node features transmitted from the surface node to the underground node according to the cross-layer edge weights, and maps the aggregated underground node features into multiple independent resistivity tensor components. Based on tensor symmetry, the multiple independent resistivity tensor components are reconstructed into second-order anisotropic tensors to obtain the predicted resistivity tensor.

[0011] Furthermore, the spatial attitude of the formation includes the dip angle and the azimuth angle, and a rotation matrix is ​​constructed based on the dip angle and the azimuth angle to transform from the global coordinate system to the local coordinate system of the formation. The predicted resistivity tensor is subjected to coordinate transformation according to the rotation matrix, and a first predicted component parallel to the ground plane and a second predicted component perpendicular to the ground plane are obtained from the transformed predicted resistivity tensor.

[0012] Furthermore, the rock electrical relationship is based on the anisotropic Archie's law, and the first theoretical resistivity parallel to the ground plane and the second theoretical resistivity perpendicular to the ground plane are determined according to the porosity and the updated water saturation, respectively. The first predicted component corresponding to the first theoretical resistivity and the second predicted component corresponding to the second theoretical resistivity are determined from the predicted resistivity tensor after coordinate transformation, and the physical constraint loss is determined based on the deviation between the first predicted component and the first theoretical resistivity and the deviation between the second predicted component and the second theoretical resistivity.

[0013] Further, the data constraint loss is determined based on the error between the predicted resistivity tensor of the sample node and the measured resistivity data. The physical constraint loss, the data constraint loss, and the network weight regularization term of the graph convolutional network are weighted to obtain the total loss function, and the graph convolutional network is trained based on the total loss function.

[0014] A soil resistivity physical consistency estimation system, the system comprising: The data acquisition and initialization module acquires spatial feature data of the target area and measured resistivity data of sample nodes, and constructs a heterogeneous spatial map including surface nodes and underground nodes based on the spatial feature data. The graph construction and seepage evolution module determines the seepage characterization from surface nodes to underground nodes based on spatial feature data, and determines the cross-layer edge weights of the heterogeneous spatial graph and updates the water saturation of underground nodes based on the seepage characterization. The network feature transfer module inputs the heterogeneous spatial graph into the graph convolutional network for feature transfer, and obtains the predicted resistivity tensor of the underground nodes. The tensor projection and physical constraint module performs coordinate transformation on the predicted resistivity tensor based on the formation spatial attitude, and constructs the physical constraint loss of the predicted resistivity tensor by combining the updated water saturation and rock electrical relationship. The network training and field inference module trains a graph convolutional network based on physical constraint loss and data constraint loss determined by the predicted resistivity tensor and measured resistivity data, and obtains the three-dimensional soil resistivity distribution of the target area based on the trained graph convolutional network.

[0015] A computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, can implement the soil resistivity physical consistency estimation method.

[0016] The technical solution of this invention can achieve the following technical effects: By acquiring multi-source spatial features of the target area and constructing a surface-subsurface heterogeneous spatial map, the method drives cross-layer feature transfer and water saturation update through hydrological seepage processes. It uses graph convolutional networks to predict subsurface resistivity tensors, then combines the spatial attitude of the strata to perform tensor coordinate transformation and introduces rock electrical relationships to construct physical constraint loss. Finally, it combines measured data to constrain and train the network to deduce the three-dimensional soil resistivity distribution of the target area. This effectively solves the problem of insufficient physical consistency of three-dimensional resistivity prediction results caused by the lack of synergistic constraints between hydrological seepage evolution and the anisotropic rock electrical laws of the strata in existing soil resistivity estimation methods.

[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the method for estimating the physical consistency of soil resistivity. Figure 2 This is a schematic diagram of the topological structure of the surface-subsurface heterogeneous space. Figure 3 A schematic diagram of a graph convolutional network structure driven by both physical and data constraints; Figure 4 A schematic diagram of tensor coordinate transformation and directional component extraction for predicting resistivity; Figure 5 A schematic diagram of the functional module structure of a soil resistivity physical consistency estimation system; Figure 6 This is a schematic diagram of the electronic device structure. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0022] Example 1; like Figure 1 As shown, this application provides a method for estimating the physical consistency of soil resistivity, the method comprising: S10: Obtain spatial feature data of the target area and measured resistivity data of sample nodes, and construct a heterogeneous spatial map including surface nodes and underground nodes based on the spatial feature data. S20: Determine the seepage characterization from surface nodes to underground nodes based on spatial feature data, determine the cross-layer edge weights of the heterogeneous spatial map based on the seepage characterization, and update the water saturation of underground nodes. S30: Input the heterogeneous spatial graph into the graph convolutional network for feature transfer to obtain the predicted resistivity tensor of the underground nodes; S40: Perform coordinate transformation on the predicted resistivity tensor based on the formation spatial attitude, and construct the physical constraint loss of the predicted resistivity tensor by combining the updated water saturation and rock electrical relationship; S50: A graph convolutional network is trained based on physical constraint loss and data constraint loss determined by the predicted resistivity tensor and measured resistivity data, and the three-dimensional soil resistivity distribution of the target area is obtained based on the trained graph convolutional network.

[0023] Specifically, this embodiment provides a method for estimating the physical consistency of soil resistivity. The method establishes a spatial relationship between the surface environment, underground medium and the spatial state of the strata based on the spatial characteristic data of the target area. It introduces the seepage process formed by changes in surface hydrology into the feature transfer of the graph convolutional network, and uses the electrical laws of the underground medium and measured resistivity data to jointly constrain the network training to obtain the three-dimensional soil resistivity distribution of the target area. The computing device acquires the spatial coordinate boundary information of the target area and discretizes the target area in three dimensions based on the spatial coordinate boundary. A two-dimensional grid is divided along the horizontal plane according to a preset grid resolution, forming surface nodes at the top of the three-dimensional space. The grid resolution ranges from 10 meters to 50 meters. Vertically extending downwards along the surface nodes, underground nodes are formed at preset depth intervals, ranging from 1 meter to 5 meters. Surface nodes carry surface environmental and hydrological data, while underground nodes carry underground media property data. Acquire spatial characteristic data corresponding to the target area; spatial characteristic data includes data reflecting the surface environment, underground media properties, and stratigraphic spatial state, including at least elevation data, precipitation flux data, vertical permeability, porosity, node burial depth data, stratigraphic dip angle, and stratigraphic azimuth angle; elevation data is extracted from digital elevation models, precipitation flux data is read from meteorological monitoring data, vertical permeability, porosity, stratigraphic dip angle, and stratigraphic azimuth angle are obtained from engineering geological exploration data, and node burial depth data is determined based on the surface node elevation and the corresponding underground node elevation; Acquire sample nodes with actual resistivity measurement results within the target area and read the measured resistivity data corresponding to each sample node; the sample nodes are located in underground nodes, and a correspondence is established between the spatial location of the sample nodes and the measured resistivity data, so as to calculate the error between the prediction results and the measured results during the network training process; Establish node connection relationships based on the spatial location of surface nodes and underground nodes to form a heterogeneous spatial map; the heterogeneous spatial map includes surface nodes and underground nodes, and establishes cross-layer edges from the surface to the underground between surface nodes and corresponding underground nodes, so that the surface hydrological information carried by the surface nodes can be transmitted to the underground nodes. The hydraulic state between surface nodes and corresponding underground nodes is determined based on elevation data, precipitation flux data, and node burial depth data. The dynamic seepage rate from surface nodes to underground nodes is determined by combining vertical permeability. The dynamic seepage rate is used as a seepage characteristic quantity to characterize the intensity of surface water migration to underground media. The dynamic seepage rate is nonlinearly mapped to obtain the cross-layer edge weights between surface nodes and underground nodes; the cross-layer edge weights change with the dynamic seepage rate and participate in the feature transfer from surface nodes to underground nodes. Simultaneously, the dynamic seepage rate corresponding to the surface node connected to the underground node is accumulated, and the water saturation of the underground node is updated based on the accumulation result and the initial water saturation of the underground node; the updated water saturation reflects the groundwater state formed after the continuous migration of surface water to the ground. The resulting heterogeneous spatial graph is input into a graph convolutional network, and features are propagated according to the connection relationships between nodes. Features of surface nodes are propagated to underground nodes through cross-layer edges, and the propagation intensity of corresponding features is adjusted by the weights of the cross-layer edges. Underground nodes aggregate the received surface node features and underground medium features to form hidden features of underground nodes. After forward operation of the graph convolutional network, the predicted resistivity tensor of each underground node in the global coordinate system is obtained. The predicted resistivity tensor is a second-order tensor that characterizes the anisotropic resistivity of the subsurface medium in three dimensions. The graph convolutional network generates independent tensor components based on the hidden features of the subsurface nodes and forms the predicted resistivity tensor corresponding to each subsurface node according to the symmetry relationship. Obtain the dip angle and azimuth angle of the strata at the corresponding locations of the underground nodes, and use the dip angle and azimuth angle to characterize the spatial attitude of the strata; establish the spatial transformation relationship between the global coordinate system and the local strata coordinate system based on the dip angle and azimuth angle, transform the predicted resistivity tensor in the global coordinate system to the local strata coordinate system, and obtain the predicted components parallel to the strata plane and the predicted components perpendicular to the strata plane from the transformed resistivity tensor. Combining the updated water saturation, porosity, and Archie's law of anisotropy, theoretical resistivity benchmark values ​​are calculated for directions parallel to and perpendicular to the ground plane. The errors between the predicted components and the corresponding theoretical resistivity benchmark values ​​in the two directions are calculated, and physical constraint losses are formed based on the errors in the two directions. The physical constraint loss characterizes the deviation between the predicted resistivity tensor and the electrical properties of the subsurface medium. Based on the spatial location of the sample nodes, the corresponding predicted resistivity tensor is extracted from the output of the graph convolutional network and compared with the corresponding measured resistivity data. The data constraint loss is formed based on the error between the predicted resistivity tensor and the measured resistivity data. The data constraint loss characterizes the deviation between the predicted results of the sample nodes and the actual measurement results. The network training objective is formed based on physical constraint loss and data constraint loss, and the graph convolutional network parameters are updated by gradient descent. During training, the gradient generated by physical constraint loss participates in the network parameter update, so that the prediction results are constrained by the electrical laws of the underground medium; the gradient generated by data constraint loss participates in the network parameter update, so that the prediction results are constrained by the measured resistivity data of the sample nodes. After the training reaches the preset convergence condition, the parameter update is stopped and the graph convolutional network parameters are locked. The feature data of the underground nodes to be predicted in the target area are input into the trained graph convolutional network. The forward feature is passed according to the node connection relationship in the heterogeneous spatial graph to obtain the predicted resistivity tensor of each underground node. Based on the spatial coordinates of the underground nodes and the corresponding predicted resistivity tensor, spatial inference is performed to form the three-dimensional soil resistivity distribution of the target area.

[0024] The technical solution of this invention obtains multi-source spatial features of the target area and constructs a surface-subsurface heterogeneous spatial map. It drives the cross-layer feature transfer and water saturation update through hydrological seepage processes, uses graph convolutional networks to predict subsurface resistivity tensors, performs tensor coordinate transformation in combination with the spatial attitude of the strata, and introduces rock electrical relationships to construct physical constraint loss. Finally, it combines measured data to constrain and train the network to deduce the three-dimensional soil resistivity distribution of the target area. This effectively solves the problem of insufficient physical consistency of three-dimensional resistivity prediction results caused by the lack of synergistic constraints between hydrological seepage evolution and the anisotropic rock electrical laws of the strata in existing soil resistivity estimation methods.

[0025] Furthermore, spatial characteristic data include elevation data, precipitation flux data, vertical permeability, porosity, node burial depth data, formation dip angle, formation azimuth angle, and initial water saturation.

[0026] As a preferred embodiment of the above, the spatial feature data includes at least elevation data, precipitation flux data, vertical permeability, porosity, node burial depth data, stratum dip angle, stratum azimuth angle, and initial water saturation. When acquiring spatial feature data, the elevation data and surface slope data of the target area are extracted from the digital elevation model file, precipitation flux data are read from the meteorological monitoring system, vertical permeability, porosity, stratum dip angle, and stratum azimuth angle are obtained from the engineering geological exploration data, and the node burial depth data is determined based on the difference between the surface node elevation and the corresponding underground node elevation. For each surface node, elevation data, surface slope data, precipitation flux data, and initial soil moisture index obtained from historical hydrological data are extracted. These four physical parameters are then concatenated and normalized to generate a feature vector for the surface node. The feature vector of the surface node is represented as follows: ; in, Indicates the first Feature vectors of each surface node Indicates the first Elevation data of each surface node Indicates the first Surface slope data for each surface node. Indicates the first Precipitation flux data for each surface node, Indicates the first The initial soil moisture index for each surface node was determined based on historical hydrological data. Indicates matrix transpose; For underground nodes, the vertical permeability, porosity, node burial depth, formation dip angle, formation azimuth angle, and initial water saturation determined based on core water content test results are extracted at the corresponding locations. These six physical parameters are then concatenated to generate the feature vector of the underground node. The feature vector of the underground node is represented as follows: ; in, Indicates the first Feature vectors of underground nodes Indicates the first Vertical permeability of the strata at the location of each underground node This indicates the porosity at the corresponding location. This represents the node burial depth data. Indicates the dip angle of the strata. Indicates the azimuth of the strata. This indicates the initial water saturation determined based on the core water content test results. Indicates matrix transpose; The feature vectors of surface nodes and underground nodes are stored according to the spatial coordinates of the corresponding nodes, and a correspondence is established with the nodes in the heterogeneous spatial graph, which serves as the node input data when the graph convolutional network performs spatial feature transfer. While initializing node features, sample nodes with measured data are determined based on historical logging coordinates within the target area. Anisotropic array induced resistivity curve data obtained from exploration instruments is read and parsed into a three-dimensional symmetric matrix to obtain the true resistivity tensor corresponding to each sample node. The true resistivity tensor is stored according to the spatial coordinates of the sample nodes and used during network training to calculate the data constraint loss between the predicted resistivity tensor and the true resistivity tensor.

[0027] Furthermore, such as Figure 2 As shown, the heterogeneous spatial map also includes surface-level edges connecting surface nodes and underground-level edges connecting underground nodes. Cross-level edges are constructed according to the spatial correspondence between surface nodes and underground nodes, while surface-level edges and underground-level edges are constructed according to the spatial adjacency between nodes of the same level.

[0028] As a preferred embodiment of the above, a heterogeneous spatial map is constructed based on the spatial discretization results of surface nodes and underground nodes. The heterogeneous spatial map includes cross-layer edges, surface same-layer edges, and underground same-layer edges. Cross-layer edges are used to connect surface nodes and underground nodes, surface same-layer edges are used to connect spatially adjacent surface nodes, and underground same-layer edges are used to connect spatially adjacent underground nodes. Traverse the surface nodes and underground nodes within the target area, and determine the spatial correspondence between surface nodes and underground nodes based on the three-dimensional spatial coordinates of the nodes; for any surface node, determine the underground node located directly below it along the vertical direction of the node, and establish a cross-layer edge between the surface node and the corresponding underground node; when a surface node corresponds to multiple underground nodes at different depths in the vertical direction, establish cross-layer edges from the surface node to each underground node respectively. The cross-layer edge adopts a unidirectional directional connection, with the connection direction from the surface node to the underground node, so that the surface hydrological characteristics carried by the surface node are transmitted to the underground node in the vertical direction. For the connection relationship between surface nodes, the horizontal Euclidean distance between nodes of the same layer is calculated based on the spatial coordinates of each surface node; when the horizontal Euclidean distance between two surface nodes is less than the preset topological adjacency threshold, a surface edge of the same layer is established between the two surface nodes. For the connection relationship between underground nodes, the horizontal Euclidean distance between nodes on the same level is calculated based on the spatial coordinates of each underground node; when the horizontal Euclidean distance between two underground nodes is less than the preset topological adjacency threshold, an underground edge on the same level is established between the two underground nodes; the preset topological adjacency threshold is set to 1.5 times the grid resolution, so that each node establishes a horizontal topological connection with the surrounding eight neighboring nodes. Both surface and underground same-layer edges are undirected connections. For each surface and underground same-layer edge, the three-dimensional spatial coordinates of the nodes at both ends of the edge are extracted and the Euclidean distance is calculated. The reciprocal of the Euclidean distance is normalized and the normalized results are used as the weights of the surface and underground same-layer edges, respectively. The adjacency matrix of the constructed heterogeneous spatial graph is stored in a compressed sparse row format, and no self-loop edges are added to the surface nodes and underground nodes. The heterogeneous spatial graph is formed by surface nodes, underground nodes, cross-layer edges, surface same-layer edges, and underground same-layer edges.

[0029] Furthermore, the hydraulic gradient between the surface node and the corresponding underground node is determined based on precipitation flux data, elevation data and node burial depth data, and the dynamic seepage rate is determined by combining the vertical permeability, with the dynamic seepage rate as the seepage characterization quantity. Nonlinear mapping of dynamic seepage rate is performed to obtain cross-layer edge weights; The updated water saturation is determined based on the cumulative amount of dynamic seepage rate and the initial water saturation.

[0030] As a preferred embodiment of the above, the precipitation flux data corresponding to the surface node is converted into an equivalent hydraulic head using a preset infiltration conversion coefficient; the infiltration conversion coefficient is determined based on historical hydrological logging data and ranges from 0.15 to 0.35. The hydraulic gradient from the surface node to the corresponding underground node is calculated by combining the elevation data and equivalent hydraulic head of the surface node and the burial depth data of the underground node; the hydraulic gradient is expressed as: ; in, Indicates the first The surface node points to the corresponding first Hydraulic gradient of each underground node Indicates the first Elevation data of each surface node Indicates the first The equivalent water head height of each surface node, Indicates the first Node burial depth data of each underground node Indicates the elevation of underground nodes; Get the The vertical permeability of the strata at the location of each underground node is multiplied by the hydraulic gradient to obtain the dynamic seepage rate from the surface node to the underground node: ; in, Indicates the first The surface node to the first The dynamic seepage rate of each underground node, Indicates the first Vertical permeability of the strata at the location of each underground node The corresponding hydraulic gradient is represented; the dynamic seepage rate is used as a measure of seepage from surface nodes to underground nodes. The dynamic seepage rate is input into a mapping layer containing learnable weight parameters and an activation function for nonlinear mapping to obtain the cross-layer edge weights corresponding to the cross-layer edges: ; in, Indicates the first The first surface node and the first Cross-layer edge weights between underground nodes This represents the learnable weight parameters of the mapping layer. This represents the bias term parameters of the mapping layer. This represents a non-linear activation function; the non-linear activation function uses the Sigmoid function, which compresses the mapping result to the interval between 0 and 1. For any underground node, extract all surface neighbor nodes connected to that underground node, and sum the dynamic seepage rates transmitted from each surface neighbor node; multiply the summation result by a preset time accumulation constant and saturation conversion coefficient to obtain the saturation increment, then add the saturation increment to the initial water saturation of the underground node, and truncate using the upper limit threshold of saturation to obtain the updated water saturation: ; in, Indicates the first Updated water saturation of each underground node Indicates the first Initial water saturation of each underground node, Indicates the relationship with the first The set of surface neighbor nodes connected to each underground node. This indicates the corresponding dynamic seepage rate. Represents the time accumulation constant. Represents the saturation conversion factor. This indicates the upper limit threshold for saturation. This indicates the minimum value operation; Precipitation flux data are based on daily data, in millimeters per day; time cumulative constant. The time frame is set to 72 hours, and a sliding window accumulation method is used. When calculating the dynamic seepage rate at the current moment, the precipitation flux data of the three consecutive days preceding the current moment are accumulated, and the daily precipitation flux data are summed with equal weights; saturation conversion coefficient. The calibration was performed based on empirical data on rock pore structure, with values ​​ranging from 0.05 to 0.15 cubic meters per second; the upper limit threshold for saturation was also determined. Set to 1.0.

[0031] Furthermore, such as Figure 3 As shown, the graph convolutional network performs weighted aggregation of node features transmitted from surface nodes to underground nodes based on cross-layer edge weights, and maps the aggregated underground node features into multiple independent resistivity tensor components. Based on tensor symmetry, the multiple independent resistivity tensor components are reconstructed into second-order anisotropic tensors to obtain the predicted resistivity tensor.

[0032] As a preferred embodiment of the above, the 4-dimensional feature vectors of surface nodes and the 6-dimensional feature vectors of underground nodes are input into a graph convolutional network. The graph convolutional network includes an input layer, three graph convolutional hidden layers, and an output layer. The input layer maps the 4-dimensional feature vectors of surface nodes and the 6-dimensional feature vectors of underground nodes to a 64-dimensional unified feature space through mutually independent linear transformation matrices, and inputs the mapped node features and edge weights in the heterogeneous spatial graph into the network operation module. The output dimension of each graph convolutional hidden layer is 128, and ReLU nonlinear activation function and batch normalization layer are set between layers. Batch processing uses full graph input of all nodes. Regarding the first For each underground node, the features of its surface neighbor nodes across layers and its underground neighbor nodes within the same layer are aggregated in each graph convolutional hidden layer. The single-layer feature update is represented as follows: ; in, Indicates the first The output of the first layer Hidden feature vectors of underground nodes; Indicates the relationship with the first A set of surface neighbor nodes connected to an underground node across multiple layers; Indicates the relationship with the first The set of underground neighbor nodes connected to each underground node on the same floor; Indicates the first The first surface node and the first Cross-layer edge weights between underground nodes; Indicates the first The underground neighbor node on the same floor and the first Weights of underground same-level edges between underground nodes; Indicates the first The first in the layer Hidden feature vectors of each surface node; Indicates the first The first in the layer Hidden feature vectors of neighboring nodes on the same underground floor; and They represent the first A learnable weight matrix for cross-layer feature transfer and intra-layer feature transfer in the subsurface; Indicates the first Layer bias terms; For heterogeneous feature transfer from surface nodes to underground nodes via cross-layer edges, the single-layer feature update is represented as: ; in, Indicates the relationship with the first The set of surface neighbor nodes connected to an underground node. Indicates the weight of cross-layer edges. Indicates the first The first in the layer The hidden feature vectors of each of the surface neighbor nodes, Indicates the first The learnable weight matrix of the layer, Indicates the first The bias term of the layer; after multi-layer graph convolution operation, the hidden features of the underground nodes are obtained by integrating spatial seepage information and underground medium properties; The hidden feature vectors of the underground nodes output from the last layer of the graph convolutional network are extracted and input into a fully connected linear layer. The fully connected linear layer maps the 128-dimensional hidden features into six independent tensor components, which are represented as follows: ; in, Indicates the first The set of six independent tensor components corresponding to each underground node. , and These represent the independent diagonal components along the three principal axes. , and These represent the independent off-diagonal components in different principal axis intersection directions. Indicates matrix transpose; Based on the symmetry relationship of the second-order anisotropic tensor, the six independent tensor components are mapped to their corresponding positions in the three-dimensional matrix, and the off-diagonal elements at mutually symmetrical positions on both sides of the main diagonal are set to the same value, forming the... Predicted resistivity tensor of each underground node in the global coordinate system: ; in, Indicates the first Predicted resistivity tensor of an underground node in the global coordinate system.

[0033] Furthermore, the spatial attitude of the stratigraphy includes the dip angle and the azimuth angle. Based on the dip angle and the azimuth angle, a rotation matrix is ​​constructed to transform the global coordinate system to the local stratigraphic coordinate system. The predicted resistivity tensor is subjected to coordinate transformation based on the rotation matrix, and a first predicted component parallel to the ground plane and a second predicted component perpendicular to the ground plane are obtained from the transformed predicted resistivity tensor.

[0034] As a preferred embodiment of the above, the dip angle and azimuth angle of the strata at the corresponding locations of each underground node are obtained, and a three-dimensional Euler rotation matrix is ​​constructed based on the dip angle and azimuth angle to transform from the global coordinate system to the local strata coordinate system; the global coordinate system uses geographic north, east and vertical directions as spatial directions, and the local strata coordinate system uses the ground surface as the reference plane; during coordinate transformation, the azimuth angle is first rotated around the global Z-axis, and then the dip angle is rotated around the rotated X-axis; No. The three-dimensional Euler rotation matrix corresponding to each underground node is represented as follows: ; in, Indicates the first The three-dimensional Euler rotation matrix corresponding to each underground node. Indicates the first The dip angle of the strata at the location corresponding to each underground node. Indicates the first The azimuth of the strata corresponding to the location of each underground node; Get the Predicted resistivity tensor of each underground node in the global coordinate system Multiplying the predicted resistivity tensor by the three-dimensional Euler rotation matrix on the left and then by the transpose of the three-dimensional Euler rotation matrix on the right, we obtain the resistivity tensor in the local coordinate system of the formation: ; in, Indicates the first Resistivity tensor of an underground node in the local coordinate system of the strata Indicates the first Predicted resistivity tensor of an underground node in the global coordinate system A transpose of a three-dimensional Euler rotation matrix; Extract the first row and first column elements and the second row and second column elements from the resistivity tensor in the local stratigraphic coordinate system. Sum the two elements and take the average to obtain the predicted component parallel to the stratigraphic plane: ; in, Indicates the first The predicted components of each underground node in the direction parallel to the ground plane. This represents the first row and first column element of the resistivity tensor in the local coordinate system. This refers to the element in the second row and second column. Extracting the third row and third column element from the resistivity tensor in the local formation coordinate system yields the predicted component perpendicular to the formation plane: ; in, Indicates the first The predicted component of each underground node perpendicular to the ground plane. This represents the element in the third row and third column of the resistivity tensor in the local coordinate system.

[0035] Furthermore, such as Figure 4 As shown, the rock electrical relationship is based on Archie's law of anisotropy. The first theoretical resistivity parallel to the ground plane and the second theoretical resistivity perpendicular to the ground plane are determined according to the porosity and the updated water saturation, respectively. The first predicted component corresponding to the first theoretical resistivity and the second predicted component corresponding to the second theoretical resistivity are determined from the predicted resistivity tensor after coordinate transformation. The physical constraint loss is determined based on the deviation between the first predicted component and the first theoretical resistivity and the deviation between the second predicted component and the second theoretical resistivity.

[0036] As a preferred embodiment of the above, obtaining the first Porosity of each underground node Updated water saturation and the inherent resistivity of pore fluid The theoretical resistivity in two directions was calculated using lithology coefficients, cementation indexes, and saturation indices parallel and perpendicular to the ground plane, respectively. The theoretical resistivity parallel to the ground plane is expressed as: ; in, Indicates the first The theoretical resistivity of each underground node parallel to the ground plane. Indicates the lithology coefficient in the parallel direction. Represents the inherent resistivity of pore fluid. Indicates the first Porosity of each underground node Indicates the parallel bonding index. Indicates the first Updated water saturation of each underground node Indicates the parallel saturation index; The theoretical resistivity perpendicular to the ground plane is expressed as: ; in, Indicates the first The theoretical resistivity of each underground node perpendicular to the ground plane. Indicates the lithology coefficient in the vertical direction. Indicates the vertical bonding index. Indicates the vertical saturation index; The lithology coefficient, cementation index, and saturation index in the parallel and perpendicular directions were calibrated based on core experimental data from the target area. Core samples from different depths within the target area were selected, and the water saturation of the core samples was controlled in the laboratory. The resistivity in the directions parallel and perpendicular to the ground plane under different water saturation conditions was measured, and the lithology coefficient, cementation index, and saturation index in the corresponding directions were obtained using nonlinear least squares fitting. The inherent resistivity of the pore fluid was also measured. Determined based on the conductivity test results of groundwater samples from the site; Lithology coefficient in parallel direction and vertical lithology coefficient The value ranges from 0.8 to 1.2; the parallel bonding index and vertical bonding index The value range is from 1.5 to 2.5; Parallel saturation index and vertical saturation index The value ranges from 1.8 to 2.2; the inherent resistivity of the pore fluid. The value ranges from 0.1 to 1.0 ohmmeter; Obtain the prediction components of each underground node in the direction parallel to the ground plane. and the predicted component perpendicular to the ground plane The squared differences between the two predicted components and their corresponding theoretical resistivity are calculated separately. The sum of these squared differences in both directions is then averaged over all underground nodes to obtain the physical constraint loss. ; in, Represents physical constraint loss. This represents the total number of underground nodes within the target area. Indicates the first The predicted components of each underground node in the direction parallel to the ground plane. This represents the corresponding parallel theoretical resistivity. Indicates the first The predicted component of each underground node perpendicular to the ground plane. This represents the corresponding vertical theoretical resistivity.

[0037] Furthermore, the data constraint loss is determined based on the error between the predicted resistivity tensor of the sample node and the measured resistivity data. The physical constraint loss, data constraint loss, and the network weight regularization term of the graph convolutional network are weighted to obtain the total loss function, and the graph convolutional network is trained based on the total loss function.

[0038] As a preferred embodiment of the above, based on the spatial coordinate index of the sample nodes, the predicted resistivity tensor corresponding to each sample node is extracted from the output of the graph convolutional network, and the true resistivity tensor corresponding to each sample node is read; the squared Frobenius norm between the predicted resistivity tensor and the true resistivity tensor is calculated, and the average is taken over all sample nodes to obtain the data constraint loss. ; in, Indicates data constraint loss. This represents the total number of sample nodes with measured resistivity data. Indicates the sample node index. Indicates the first The predicted resistivity tensor corresponding to each sample node Indicates the first The true resistivity tensor corresponding to each sample node Denotes the squared Frobenius norm of a matrix; Extract the learnable weight matrices from each layer of the graph convolutional network, calculate the squared Frobenius norm of each learnable weight matrix, and sum them to form the network weight regularization term. Linearly weight the physical constraint loss, data constraint loss, and network weight regularization term to obtain the total loss function. ; in, Represents the total loss function. Represents physical constraint loss. Indicates data constraint loss. This represents the total number of layers in the graph convolutional network. This represents the graph convolutional network. The learnable weight matrix of the layer, , and These represent the weighting coefficients corresponding to the physical constraint loss, data constraint loss, and network weight regularization term, respectively. Weighting coefficients corresponding to physical constraint loss The value ranges from 0.5 to 0.7, representing the weighting coefficient corresponding to the data constraint loss. The value ranges from 0.2 to 0.4, representing the weighting coefficients corresponding to the network weight regularization term. The value ranges from 0.01 to 0.05; The graph convolutional network was trained using the total loss function. The network training used the Adam optimizer, with an initial learning rate of 0.001. The learning rate was adjusted using a step decay method, and the learning rate was decayed to 0.5 of the current learning rate every 50 training epochs. The maximum number of training epochs was set to 2000. During training, the gradient of the total loss function with respect to the network parameters is calculated using an automatic differentiation mechanism. The updated water saturation is used as an intermediate variable in the computation graph. The gradient generated by the physical constraint loss is backpropagated to the nonlinear mapping layer of the cross-layer edge weights through chain differentiation. The learnable parameters in the cross-layer edge weight mapping layer, graph convolutional layer and linear fully connected layer are updated according to the gradient obtained from backpropagation. The network parameters are iteratively updated in the reverse direction of the gradient, and the total loss function is calculated after each training epoch. The network is considered closed when the total loss function converges to a preset threshold, and the decrease in the total loss function over 100 consecutive training epochs is less than a certain value. At this point, stop network training and fix the graph convolutional network parameters; The feature vectors of the underground nodes to be predicted in the target area are input into the trained graph convolutional network. The forward operation is performed in combination with the topology of the heterogeneous spatial graph to obtain the predicted resistivity tensor corresponding to each underground node. The three-dimensional spatial coordinates and corresponding predicted resistivity tensors of each underground node are extracted, and the spatial position between discrete underground nodes is three-dimensionally interpolated using inverse distance weighted interpolation. The interpolation power of inverse distance weighted interpolation is set to 2, and the search radius is set to 3 times the resolution of the 3D spatial grid. During the interpolation process, the global coordinates of the location to be interpolated are transformed to the local coordinate system of the stratum based on the 3D Euler rotation matrix corresponding to the dip angle. In the local coordinate system of the stratum, the distance weights between the location to be interpolated and the adjacent underground nodes are calculated to obtain the 3D soil resistivity continuous distribution tensor field of the target area. The 3D soil resistivity continuous distribution tensor field is then serialized and saved to the storage medium of the computing platform.

[0039] Example 2; Based on the same inventive concept as the soil resistivity physical consistency estimation method in the aforementioned embodiments, such as... Figure 5 As shown, the present invention also provides a soil resistivity physical consistency estimation system, the system comprising: The data acquisition and initialization module acquires spatial feature data of the target area and measured resistivity data of sample nodes, and constructs a heterogeneous spatial map including surface nodes and underground nodes based on the spatial feature data. The graph construction and seepage evolution module determines the seepage characterization from surface nodes to underground nodes based on spatial feature data, and determines the cross-layer edge weights of the heterogeneous spatial graph and updates the water saturation of underground nodes based on the seepage characterization. The network feature transfer module inputs the heterogeneous spatial graph into the graph convolutional network for feature transfer, and obtains the predicted resistivity tensor of the underground nodes. The tensor projection and physical constraint module performs coordinate transformation on the predicted resistivity tensor based on the formation spatial attitude, and constructs the physical constraint loss of the predicted resistivity tensor by combining the updated water saturation and rock electrical relationship. The network training and field inference module trains a graph convolutional network based on physical constraint loss and data constraint loss determined by the predicted resistivity tensor and measured resistivity data, and obtains the three-dimensional soil resistivity distribution of the target area based on the trained graph convolutional network.

[0040] The adjustment system described above in this invention can effectively realize a method for estimating the physical consistency of soil resistivity, and the technical effects it can achieve are as described in the above embodiments, and will not be repeated here.

[0041] Example 3; Based on the same inventive concept as the soil resistivity physical consistency estimation method in the aforementioned embodiments, such as... Figure 6 As shown, the present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, can realize a method for estimating the physical consistency of soil resistivity.

[0042] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of the application as defined herein, and are to be considered as covering any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Thus, if such modifications and modifications fall within the scope of this application and its equivalents, this application intends to include such modifications and modifications.

Claims

1. A method for estimating the physical consistency of soil resistivity, characterized in that, The method includes: Acquire spatial feature data of the target area and measured resistivity data of sample nodes, and construct a heterogeneous spatial map including surface nodes and underground nodes based on the spatial feature data; Based on the spatial feature data, the seepage characterization of the surface node to the underground node is determined, and the cross-layer edge weight of the heterogeneous spatial map is determined based on the seepage characterization and the water saturation of the underground node is updated. The heterogeneous spatial graph is input into a graph convolutional network for feature transfer to obtain the predicted resistivity tensor of the underground node; The predicted resistivity tensor is subjected to coordinate transformation based on the formation spatial attitude, and the physical constraint loss of the predicted resistivity tensor is constructed by combining the updated water saturation and rock electrical relationship. The graph convolutional network is trained based on the physical constraint loss and the data constraint loss determined by the predicted resistivity tensor and the measured resistivity data, and the three-dimensional soil resistivity distribution of the target area is obtained based on the trained graph convolutional network.

2. The method for estimating the physical consistency of soil resistivity according to claim 1, characterized in that, The spatial characteristic data includes elevation data, precipitation flux data, vertical permeability, porosity, node burial depth data, formation dip angle, formation azimuth angle, and initial water saturation.

3. The method for estimating the physical consistency of soil resistivity according to claim 1, characterized in that, The heterogeneous spatial map also includes surface-level edges connecting the surface nodes and underground-level edges connecting the underground nodes. The cross-level edges are constructed according to the spatial correspondence between the surface nodes and the underground nodes, and the surface-level edges and underground-level edges are constructed according to the spatial adjacency between nodes of the same level.

4. The method for estimating the physical consistency of soil resistivity according to claim 1, characterized in that: The hydraulic gradient between the surface node and the corresponding underground node is determined based on precipitation flux data, elevation data and node burial depth data, and the dynamic seepage rate is determined in combination with vertical permeability, with the dynamic seepage rate as the seepage characterization quantity. The dynamic seepage rate is nonlinearly mapped to obtain the cross-layer edge weights. The updated water saturation is determined based on the cumulative amount of the dynamic seepage rate and the initial water saturation.

5. The method for estimating the physical consistency of soil resistivity according to claim 1, characterized in that, The graph convolutional network performs weighted aggregation of node features transmitted from the surface node to the underground node according to the cross-layer edge weights, and maps the aggregated underground node features into multiple independent resistivity tensor components. Based on tensor symmetry, the multiple independent resistivity tensor components are reconstructed into second-order anisotropic tensors to obtain the predicted resistivity tensor.

6. The method for estimating the physical consistency of soil resistivity according to claim 1, characterized in that: The spatial attitude of the formation includes the dip angle and the azimuth angle. A rotation matrix is ​​constructed based on the dip angle and the azimuth angle to transform from the global coordinate system to the local coordinate system of the formation. The predicted resistivity tensor is subjected to coordinate transformation according to the rotation matrix, and a first predicted component parallel to the ground plane and a second predicted component perpendicular to the ground plane are obtained from the transformed predicted resistivity tensor.

7. The method for estimating the physical consistency of soil resistivity according to claim 1, characterized in that: The electrical relationship of the rock is based on the anisotropic Archie's law. The first theoretical resistivity parallel to the ground plane and the second theoretical resistivity perpendicular to the ground plane are determined according to the porosity and the updated water saturation, respectively. The first predicted component corresponding to the first theoretical resistivity and the second predicted component corresponding to the second theoretical resistivity are determined from the predicted resistivity tensor after coordinate transformation, and the physical constraint loss is determined based on the deviation between the first predicted component and the first theoretical resistivity and the deviation between the second predicted component and the second theoretical resistivity.

8. The method for estimating the physical consistency of soil resistivity according to claim 1, characterized in that, The data constraint loss is determined based on the error between the predicted resistivity tensor of the sample node and the measured resistivity data. The physical constraint loss, the data constraint loss, and the network weight regularization term of the graph convolutional network are weighted to obtain the total loss function, and the graph convolutional network is trained based on the total loss function.

9. A soil resistivity physical consistency estimation system, characterized in that, The system includes: The data acquisition and initialization module acquires spatial feature data of the target area and measured resistivity data of sample nodes, and constructs a heterogeneous spatial map including surface nodes and underground nodes based on the spatial feature data. The graph construction and seepage evolution module determines the seepage characterization from surface nodes to underground nodes based on spatial feature data, and determines the cross-layer edge weights of the heterogeneous spatial graph and updates the water saturation of underground nodes based on the seepage characterization. The network feature transfer module inputs the heterogeneous spatial graph into the graph convolutional network for feature transfer, and obtains the predicted resistivity tensor of the underground nodes. The tensor projection and physical constraint module performs coordinate transformation on the predicted resistivity tensor based on the formation spatial attitude, and constructs the physical constraint loss of the predicted resistivity tensor by combining the updated water saturation and rock electrical relationship. The network training and field inference module trains a graph convolutional network based on physical constraint loss and data constraint loss determined by the predicted resistivity tensor and measured resistivity data, and obtains the three-dimensional soil resistivity distribution of the target area based on the trained graph convolutional network.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, implement the soil resistivity physical consistency estimation method as described in any one of claims 1-8.

Citation Information

Patent Citations

  • Stratum parameter inversion method based on physically-driven neural network

    CN119129672A

  • Reservoir physical property parameter pre-stack seismic inversion method based on resistivity logging constraint

    CN119882088A