Geothermal abnormal point identification system based on remote sensing detection

Abstract

The invention discloses a geothermal abnormal point identification system based on remote sensing detection. The system comprises a data layer which collects thermal infrared remote sensing data, multispectral / hyperspectral remote sensing data, radar data, geological and topographic data, meteorological data and geophysical prospecting data; the processing layer is used for extracting key features, capturing dynamic changes of geothermal anomalies, distinguishing continuous geothermal anomalies and transient events, combining hyperspectral data and magnetotelluric sounding data, and constructing'earth surface-deep part 'associated features; the analysis layer is used for learning a diffusion rule of geothermal anomalies along a fault zone through a graph propagation model, taking a geothermal field heat conduction equation as a loss function constraint by utilizing a physical information neural network, jointly training a temperature field prediction model, constructing a deep three-dimensional geological structure based on gravity and geophysical prospecting data, and constructing a geologic structure model; constructing a geothermal anomaly detection model through the features provided by the processing layer to recognize geothermal anomaly points; and the output layer is used for generating a geothermal anomaly probability graph.

Description

Technical Field

[0001] The present invention relates to the technical field of geothermal resource management, and in particular to a geothermal anomaly point identification system based on remote sensing detection. Background Art

[0002] With the rapid development of the global economy and the continued growth in energy demand, the depletion of traditional fossil energy sources and the resulting environmental problems are becoming increasingly prominent. Geothermal resources, as a clean, renewable energy source, offer advantages such as large reserves, wide distribution, and high utilization efficiency, making them an important option for alleviating energy pressures and reducing carbon emissions. Remote sensing technology, which acquires information about the Earth's surface through platforms such as satellites and aircraft, offers advantages such as wide coverage, rapid acquisition, and low cost. In recent years, with the continuous advancement of remote sensing technology, its application in geothermal resource exploration has become increasingly widespread. Remote sensing can obtain multi-source information such as surface temperature, thermal infrared radiation, and mineral spectral characteristics, providing rich data support for the identification of geothermal anomalies. Leveraging this multi-source information obtained through remote sensing, combined with prior knowledge from geology and geophysics, an efficient geothermal anomaly identification model is constructed. This model should accurately identify the spatial distribution and characteristics of geothermal anomalies, providing decision support for geothermal resource exploration and development.

[0003] Existing technologies for identifying geothermal anomalies rely on data-driven approaches, such as traditional machine learning algorithms or simple statistical models. These methods often overlook the physical laws underlying geothermal phenomena when tackling the complex task of identifying geothermal anomalies. For example, the diffusion of geothermal anomalies, the heat conduction process, and the interaction between geothermal fluids and surrounding rocks all follow specific physical laws. This lack of consideration for these physical laws can lead to an inability to accurately capture the true distribution and characteristics of geothermal anomalies when identifying geothermal anomalies, resulting in illogical or inaccurate identification results. Therefore, this paper proposes a geothermal anomaly identification system based on remote sensing detection. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a geothermal anomaly point identification system based on remote sensing detection.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A geothermal anomaly identification system based on remote sensing detection, comprising:

[0007] Data layer: Collection includes thermal infrared remote sensing data, multispectral / hyperspectral remote sensing data, radar data, geological and topographic data, meteorological data, and geophysical data;

[0008] Processing layer: Preprocess the collected multi-source data to extract key features. Use temporal convolutional networks to analyze multi-phase thermal infrared sequences, capture the dynamic changes of geothermal anomalies, distinguish persistent geothermal anomalies from transient events, and combine hyperspectral data with magnetotelluric sounding data to construct "surface-deep" correlation features.

[0009] Analysis layer: Utilizes a physically constrained graph neural network with a fault network as the graph structure, embedding thermal anomalies, minerals, and terrain features in nodes. A graph propagation model is used to learn the diffusion patterns of geothermal anomalies along fault zones. A physical information neural network is used to constrain the heat conduction equation of the geothermal field as a loss function. A temperature field prediction model is jointly trained. The temperature field prediction model predicts future temperature field distribution based on known geothermal anomaly information and geological conditions. A deep three-dimensional geological structure is constructed based on gravity and geophysical data. A geothermal anomaly detection model is constructed using features provided by the processing layer to identify geothermal anomalies. The identified geothermal anomalies are then graded and evaluated based on the dynamics of geothermal anomalies, the diffusion patterns of geothermal anomalies along fault zones, and the deep three-dimensional geological structure.

[0010] Output layer: Generates geothermal anomaly probability map, displays the identified geothermal anomaly points in the form of probability map, marks high-probability areas as potential target areas, and combines DEM and geological data to generate a 3D thermal anomaly distribution map.

[0011] The above technical solution further includes:

[0012] Furthermore, the processing layer preprocessing operations include radiation correction, atmospheric correction, geometric alignment and data fusion, and the key feature extraction includes surface temperature inversion, thermal infrared radiation anomaly index calculation, thermal alteration mineral spectral feature extraction, terrain feature extraction and fault density analysis.

[0013] Furthermore, the specific steps of using a temporal convolutional network to analyze multi-phase thermal infrared sequences, capture the dynamic changes of geothermal anomalies, and distinguish between persistent geothermal anomalies and transient events are as follows:

[0014] Data preparation: Collect multi-temporal thermal infrared remote sensing data and preprocess the data, including radiation correction, atmospheric correction, and geometric registration;

[0015] Constructing a temporal convolutional network: Designing a temporal convolutional network architecture, including convolutional layers, residual blocks, causal convolutions, and dilated convolutions, determining hyperparameters, initializing network weights, selecting Adam and mean square error (MSE), and expressing the causal convolution in the residual block of the temporal convolutional network as follows:

[0016]

[0017] Among them, y t is the output at time step t, x t-iis the input of time step ti, w i is the convolution kernel weight, k is the convolution kernel size;

[0018] Training the temporal convolutional network model: Divide the preprocessed thermal infrared data into a training set and a validation set, input the data into the temporal convolutional network model, perform forward propagation to calculate the output, calculate the mean square error (MSE), and update the network weights through backpropagation. Repeat the training process until the model achieves satisfactory performance on the validation set.

[0019] Capturing dynamic changes in geothermal anomalies: Applying the trained temporal convolutional network model to new multi-temporal thermal infrared data, analyzing the model output, identifying dynamic patterns in geothermal anomalies, and distinguishing persistent geothermal anomalies from transient events by comparing outputs at different time steps.

[0020] Evaluation and Validation: Verify the prediction results of the temporal convolutional network model, calculate evaluation metrics, evaluate model performance, and adjust model parameters or improve model architecture based on the evaluation results.

[0021] Furthermore, the method of combining hyperspectral data with magnetotelluric sounding data to construct “surface-deep” correlation features includes the following steps:

[0022] Determine the associated characteristic parameters: Select characteristic parameters related to hydrothermal activity from the hyperspectral data;

[0023] Establishing a correlation model: Linear regression is used to establish a correlation model between hyperspectral features and MT features.

[0024] Furthermore, the physical constraint-based graph neural network uses a fault network as a graph structure, embeds thermal anomalies, minerals, and terrain features in nodes, and learns the diffusion law of geothermal anomalies along fault zones through a graph propagation model. The specific steps are as follows:

[0025] Constructing the graph structure:

[0026] Define nodes: The fault intersections, endpoints, and significant geothermal anomaly points within the study area are used as nodes of the graph;

[0027] Define edges: Define edges based on fault connectivity and the spatial proximity between geothermal anomalies. If there is a direct fault connection between two nodes or the spatial distance is less than a certain threshold, an edge is established between the two nodes.

[0028] Node feature embedding:

[0029] Thermal anomaly feature: Extract the thermal infrared radiation anomaly index at each node as the thermal anomaly feature;

[0030] Mineral characteristics: Use hyperspectral data to identify the mineral composition of the node area, and use the ratio of the main mineral types as the mineral characteristics;

[0031] Terrain features: Extract the terrain elevation and slope of the node as terrain features;

[0032] Graph propagation model:

[0033] Define the graph convolution layer: Use the graph convolution network layer to propagate and aggregate node features. The calculation formula of the graph convolution layer is Among them, H (l) is the node feature matrix of the lth layer, is the normalized adjacency matrix, W (l) is a learnable weight matrix, σ is the activation function;

[0034] Multi-layer graph convolution: stacking multiple graph convolution layers to learn node representations and the diffusion patterns of geothermal anomalies along fault zones;

[0035] Learn about the diffusion patterns of geothermal anomalies along fault zones:

[0036] Define the loss function: Use the mean square error loss function to measure the difference between the geothermal anomaly diffusion probability predicted by the model and the actual observed value;

[0037] Training model: Use training data to train the graph propagation model, and update the model parameters through the backpropagation algorithm to minimize the loss function;

[0038] Prediction and Evaluation: Predicting the diffusion of geothermal anomalies: Use the trained graph propagation model to predict the probability of geothermal anomaly diffusion for all nodes in the study area.

[0039] Furthermore, the physical information neural network is used to take the geothermal field heat conduction equation as a loss function constraint, and jointly train the temperature field prediction model. The temperature field prediction model predicts the future temperature field distribution based on known geothermal anomaly information and geological conditions. Specific steps are as follows:

[0040] Data preparation: Collect geothermal anomaly information and geological condition data. If there is observation data, prepare temperature data of the observation points;

[0041] Constructing PINN: Design a neural network with input (x, y, t) and output T pred (x, y, t), define the loss function, including physical loss and data loss, the physical loss calculation is expressed as

[0042]

[0043] Where N is the number of sampling points, (x i ,yi ,t i ) are the coordinates and time of the sampling points;

[0044] The data loss calculation is expressed as

[0045]

[0046] Where M is the number of observation points, T ots (x j ,y j ,t j ) is the temperature at the observation point;

[0047] The total loss is expressed as L = L physics +λL data , where λ is the weight of data loss;

[0048] Training PINN: Use Adam to minimize the total loss L. During the training process, continuously update the parameters of the neural network so that the predicted temperature field T pred (x, y, t) satisfies the heat conduction equation and fits the observed data. Heat conduction is described by Fourier's heat conduction law, and its two-dimensional form is Where T is temperature, t is time, α is the thermal diffusivity, and Q(x,y,t) is the heat source term;

[0049] Predicting the future temperature field: After the training is completed, use the trained PINN model to predict the future temperature field distribution, and input the future time t future And spatial coordinates (x, y), output predicted temperature T pred (x,y,t future ).

[0050] Furthermore, the method of constructing a deep three-dimensional geological structure based on gravity and geophysical data and identifying geothermal anomaly points by constructing a geothermal anomaly detection model through the features provided by the processing layer includes the following steps:

[0051] Constructing deep 3D geological structure: Using density anomalies obtained from gravity data and resistivity distribution obtained from geophysical data, combined with geological prior knowledge, a deep 3D geological structure model is constructed. Kriging interpolation is used to fuse gravity and MT data to obtain the 3D distribution of different underground lithologies and structures, which is expressed as Among them, Z(x) is the geological attribute value of the point to be estimated, Z(x i ) is the geological attribute value of the known point, λ i is the weight coefficient, obtained by minimizing the estimated variance;

[0052] Feature extraction: Extracting features related to geothermal anomalies from the constructed 3D geological structure model;

[0053] Fusion with processing layer features: Fusion of the extracted geological features with the features provided by the processing layer to form a comprehensive feature vector;

[0054] Constructing a geothermal anomaly detection model: Using the fused comprehensive feature vector as input and the known geothermal anomaly points as labels, the support vector machine model is trained. The support vector machine model is represented as

[0055]

[0056] sty i (w·x i +m)≥1-ξ i ,ξ i ≥0;

[0057] Among them, w is the weight vector, b is the bias term, ξ i is the slack variable, C is the penalty parameter, x i is the input feature vector, y i It’s a label;

[0058] Identification and verification of geothermal anomaly points: Use the trained geothermal anomaly detection model to predict unknown areas and obtain the probability distribution of geothermal anomalies. By comparing with actual geothermal exploration results, the accuracy and reliability of the model can be verified.

[0059] Furthermore, the specific steps of performing graded assessment on the identified geothermal anomaly points in combination with the geothermal anomaly dynamics, the diffusion pattern of geothermal anomalies along the fault zone, and the deep three-dimensional geological structure are as follows:

[0060] Assume that the dynamic change characteristic vector F of geothermal anomaly dynamic ;

[0061] Assume that the diffusion characteristic vector F of geothermal anomaly along the fault zone is diffusion ;

[0062] Assume that the geological structure characteristic vector F geology ;

[0063] Combined with the geothermal anomaly dynamic characteristic vector F dynamic , the diffusion characteristic vector F of geothermal anomaly along the fault zone diffusion and geological structure characteristic vector F geology , calculate the abnormal intensity score S intensity , the calculation formula is expressed as S intensity =w1·F dynamic +w2·F diffusion +w3·F geology , where w1, w2, w3 are weight coefficients;

[0064] Combined with the geothermal anomaly dynamic characteristic vector Fdynamic , the diffusion characteristic vector F of geothermal anomaly along the fault zone diffusion and geological structure characteristic vector F geology , calculate the geological stability score S stability , the calculation formula is expressed as S stability =1-(w4·F diffusion +w5·F geology ), where w4, w5 are weight coefficients;

[0065] Combined with the abnormal strength score S intensity and geological stability score S stability , and other development-related factors to calculate the development potential score S potential , the calculation formula is expressed as S potential =w6·S intensity +w7·S stability +..., where w6, w7... are weight coefficients;

[0066] According to the abnormal intensity score S intensity , geological stability score S stability and development potential score S potential , the identified geothermal anomalies are graded and evaluated, and the geothermal anomalies are divided into three levels: high, medium and low, corresponding to different score ranges:

[0067] Height: S intensity >0.3, S stability >0.7, S potential >0.4;

[0068] Medium: 0.2<S intensity ≤0.3, 0.5<S stability ≤0.7, 0.2<S potential ≤0.4;

[0069] Low: S intensity ≤0.2, S stability ≤0.5, S potential ≤0.2.

[0070] The present invention has the following beneficial effects:

[0071] In this invention, we focus on the deep fusion and cross-modal analysis of multi-source remote sensing data and geophysical data. This fusion not only improves the accuracy of geothermal anomaly identification, but also enhances the matching accuracy of surface anomalies and deep heat sources by constructing "surface-deep" correlation features. By introducing physically constrained graph neural networks (GNNs) and physical information neural networks (PINNs), these models can learn the diffusion laws of geothermal anomalies along fault zones and ensure that the temperature field prediction results comply with the law of conservation of energy, thereby improving the rationality and accuracy of the identification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a system block diagram of a geothermal anomaly point identification system based on remote sensing detection proposed by the present invention. DETAILED DESCRIPTION

[0073] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0074] See also Figure 1 As shown, the present invention is a geothermal anomaly identification system based on remote sensing detection, comprising:

[0075] Data layer: Collection includes thermal infrared remote sensing data (such as Landsat-8TIRS, ASTER, and MODIS) (used to invert land surface temperature (LST) and identify areas with abnormal surface temperature), multispectral / hyperspectral remote sensing data (such as AVIRIS-NG) (used to identify surface mineral composition (such as silicates and clay minerals) and thermal alteration characteristics, assist in the identification of geothermal anomalies, and perform cross-modal fusion with geophysical data), radar data (SAR) (used to monitor micro-surface deformation and assist in the identification of geothermal activity areas), geological and topographic data (such as DEM and geological survey data) (used to provide information such as fault distribution, volcanic activity records, and terrain elevation, providing geological background support), meteorological data (used to correct thermal infrared data), and geophysical data (such as magnetotelluric sounding (MT) and gravity measurement data) (used to perform cross-modal fusion with remote sensing data);

[0076] Processing layer: Preprocess the collected multi-source data to extract key features. Use temporal convolutional networks (TCNs) to analyze multi-phase thermal infrared sequences, capture the dynamic changes of geothermal anomalies, distinguish persistent geothermal anomalies from transient events, and combine hyperspectral data with magnetotelluric (MT) data (a type of geophysical data) to construct "surface-deep" correlation features.

[0077] Analysis layer: Using a physically constrained graph neural network (GNN) with a fault network as the graph structure, nodes are embedded with thermal anomalies, minerals, and terrain features. The diffusion patterns of geothermal anomalies along fault zones are learned through a graph propagation model. A physical information neural network (PINN) is used to constrain the heat conduction equation of the geothermal field as a loss function. A temperature field prediction model is jointly trained. The temperature field prediction model predicts the future temperature field distribution based on known geothermal anomaly information and geological conditions. A deep three-dimensional geological structure is constructed based on gravity and geophysical data (magnetotelluric sounding MT). A geothermal anomaly detection model is constructed using the features provided by the processing layer to identify geothermal anomalies. The identified geothermal anomalies are graded and evaluated in terms of anomaly intensity, geological stability, and development potential, combining geothermal anomaly dynamics, the diffusion patterns of geothermal anomalies along fault zones, and the deep three-dimensional geological structure.

[0078] Output layer: Generates geothermal anomaly probability map, displays the identified geothermal anomaly points in the form of probability map, marks high-probability areas as potential target areas, and combines DEM and geological data to generate a 3D thermal anomaly distribution map.

[0079] In one embodiment, the processing layer preprocessing operations include radiation correction, atmospheric correction, geometric alignment and data fusion. The key feature extraction includes surface temperature inversion, thermal infrared radiation anomaly index calculation, thermal alteration mineral spectrum feature extraction, terrain feature extraction and fault density analysis. The surface temperature inversion uses a split window algorithm to invert the surface temperature from thermal infrared remote sensing data. The formula is expressed as LST=T i +C1(T i -T j )+C2(T i -T j ) 2 +C0+ΔT, where T i and T j is the brightness temperature of two adjacent thermal infrared bands, C1, C2, C0 are algorithm coefficients, ΔT is the atmospheric correction term, and the purpose of calculating the thermal infrared radiation anomaly index is to quantify the degree of surface temperature anomaly. The TIAN index calculation formula is expressed as Among them, LST mean is the average land surface temperature in the study area, LST std is the standard deviation of the surface temperature. The purpose of thermal alteration mineral spectral feature extraction is to identify the surface mineral composition and its thermal alteration characteristics. For example, the spectral angle θ between the reference spectrum (known mineral spectrum) and the pixel spectrum is calculated, which is expressed as Among them, R i is the reflectance of the reference spectrum in the i-th band, P iis the reflectance of the pixel spectrum in the i-th band, n is the number of bands, and a threshold is set. If θ is less than the threshold, the pixel is considered to contain the mineral component corresponding to the reference spectrum. For example, if the reference spectrum is a thermally altered mineral spectrum, and the spectral angle θ calculated between the pixel spectrum and the reference spectrum is 10°, and the threshold is set to 15°, the pixel may contain the thermally altered mineral.

[0080] In one embodiment, the specific steps of using a temporal convolutional network (TCN) to analyze multi-temporal thermal infrared sequences, capture dynamic changes in geothermal anomalies, and distinguish persistent geothermal anomalies from transient events are as follows:

[0081] Data preparation: Collect multi-temporal thermal infrared remote sensing data, ensure that the data covers the target area and is collected at an appropriate time interval (such as weekly or monthly), and preprocess the data, including radiometric correction, atmospheric correction, and geometric registration, to eliminate sensor errors, atmospheric effects, and geometric distortion.

[0082] Build a Temporal Convolutional Network (TCN): Design the TCN architecture, including convolutional layers, residual blocks, causal convolutions, and dilated convolutions. Determine hyperparameters such as the input sequence length (e.g., 12 months), convolution kernel size, and dilation factor. Initialize the network weights, select Adam and mean square error (MSE). The causal convolution in the residual block of the TCN is expressed as:

[0083]

[0084] Among them, y t is the output at time step t, x t-i is the input of time step ti, w i is the convolution kernel weight, k is the convolution kernel size;

[0085] Training the Temporal Convolutional Network (TCN) model: Divide the preprocessed thermal infrared data into a training set and a validation set. Input the data into the Temporal Convolutional Network (TCN) model, perform forward propagation to calculate the output, calculate the mean square error (MSE), and update the network weights through backpropagation. Repeat the training process until the model achieves satisfactory performance on the validation set.

[0086] Capturing the dynamic changes of geothermal anomalies: Applying the trained temporal convolutional network (TCN) model to new multi-temporal thermal infrared data, analyzing the model output, identifying the dynamic change patterns of geothermal anomalies, and distinguishing persistent geothermal anomalies from transient events by comparing the outputs at different time steps.

[0087] Consider a new 12-month thermal infrared data series that is fed into a trained TCN model. The model outputs a surface temperature prediction for each time step. Temperature changes can be observed by plotting the time series. If temperatures in a region are consistently elevated for multiple consecutive months, this could indicate a persistent geothermal anomaly. However, if temperatures are elevated only in a single month, this could be a transient event.

[0088] Evaluation and Validation: Use ground truth data or other reliable data sources to verify the prediction results of the temporal convolutional network (TCN) model, calculate evaluation metrics (such as accuracy, recall, F1 score, etc.), evaluate model performance, and adjust model parameters or improve model architecture based on the evaluation results.

[0089] In one embodiment, the hyperspectral data is combined with magnetotelluric (MT) data (which belongs to geophysical data) to construct a "surface-deep" correlation feature, including the following steps:

[0090] Determine the associated characteristic parameters: Select characteristic parameters related to hydrothermal activity from the hyperspectral data, such as the silicification intensity index SI. The silicification intensity index is calculated by the reflectivity of the characteristic band of silicified minerals in the hyperspectral data, for example: Among them, R silicate is the reflectivity of the characteristic band of silicified minerals, R background is the reflectivity of the background features. Characteristic parameters related to the hydrothermal channel are selected from the MT data, such as the hydrothermal channel resistivity R channel , by analyzing the MT apparent resistivity profile, the resistivity value of the area where the hydrothermal channel is located is determined;

[0091] Establish a correlation model: Use linear regression to establish a correlation model between hyperspectral features and MT features. For example, the silicification intensity index SI and the hydrothermal channel resistivity R are obtained through linear regression analysis. channel The relationship between them is: R channel =a×SI+b, where a and b are regression coefficients, solved by the least squares method.

[0092] In one embodiment, the physical constraint-based graph neural network (GNN) uses a fault network as a graph structure, embeds thermal anomalies, minerals, and terrain features in nodes, and learns the diffusion law of geothermal anomalies along fault zones through a graph propagation model. The specific steps are as follows:

[0093] Constructing the graph structure:

[0094] Define nodes: The fault intersections, endpoints, and significant geothermal anomaly points within the study area are used as nodes of the graph;

[0095] Define edges: Define edges based on fault connectivity and the spatial proximity between geothermal anomalies. If there is a direct fault connection between two nodes or the spatial distance is less than a certain threshold, an edge is established between the two nodes.

[0096] Suppose there are three fault intersections V1, V2, and V3 in the study area, and a significant geothermal anomaly V4. Through geological survey and analysis, it is found that there is a direct fault connection between V1 and V2, a direct fault connection between V2 and V3, and the spatial distance between V3 and V4 is less than a set threshold (such as 1km). The constructed graph structure contains nodes V1, V2, V3, and V4, and edges (V1, V2), (V2, V3), and (V3, V4).

[0097] Node feature embedding:

[0098] Thermal anomaly feature: Extract the thermal infrared radiation anomaly index (such as TIAN index) at each node as the thermal anomaly feature;

[0099] Mineral characteristics: Use hyperspectral data to identify the mineral composition of the node area, and use the ratio of the main mineral types as the mineral characteristics;

[0100] Terrain features: Extract the terrain elevation and slope of the node as terrain features;

[0101] Graph propagation model:

[0102] Define the graph convolution layer: Use the graph convolution network (GCN) layer to propagate and aggregate node features. The calculation formula of the graph convolution layer is Among them, H (l) is the node feature matrix of the lth layer, is the normalized adjacency matrix, W (l) is a learnable weight matrix, σ is the activation function;

[0103] Multi-layer graph convolution: stacking multiple graph convolution layers to learn more complex node representations and the diffusion patterns of geothermal anomalies along fault zones;

[0104] Suppose a graph propagation model with two graph convolution layers is constructed. The input of the first graph convolution layer is the node feature matrix H (0) , after the first layer of graph convolution calculation, we get H (1) , and then H (1) As the input of the second graph convolution layer, the final node representation H is calculated (2) ;

[0105] Learn about the diffusion patterns of geothermal anomalies along fault zones:

[0106] Define the loss function: Use the mean square error loss function to measure the difference between the geothermal anomaly diffusion probability predicted by the model and the actual observed value;

[0107] Training model: Use training data to train the graph propagation model, and update the model parameters through the backpropagation algorithm to minimize the loss function;

[0108] Prediction and evaluation: Predicting the spread of geothermal anomalies: Use the trained graph propagation model to predict the probability of geothermal anomaly spread for all nodes in the study area;

[0109] Evaluate model performance: Use test set data to evaluate model performance, such as calculating accuracy, recall, F1 value, and other indicators.

[0110] In one embodiment, the physical information neural network (PINN) is used to use the geothermal field heat conduction equation as a loss function constraint to jointly train a temperature field prediction model. The temperature field prediction model predicts the future temperature field distribution based on known geothermal anomaly information and geological conditions. The specific steps are as follows:

[0111] Data preparation: Collect geothermal anomaly information and geological condition data, including the geological structure of the geothermal field, heat source distribution, initial temperature field, etc. If there is observation data, prepare temperature data of the observation point;

[0112] Constructing PINN: Design a neural network with input (x, y, t) and output T pred (x, y, t), define the loss function, including physical loss and data loss (if there is observation data), the physical loss calculation is expressed as

[0113]

[0114] Where N is the number of sampling points, (x i ,y i ,t i ) are the coordinates and time of the sampling points;

[0115] The data loss calculation is expressed as

[0116]

[0117] Where M is the number of observation points, T ots (x j ,y j ,t j ) is the temperature at the observation point;

[0118] The total loss is expressed as L = L physics +λL data , where λ is the weight of data loss;

[0119] Training PINN: Use Adam to minimize the total loss L. During the training process, continuously update the parameters of the neural network so that the predicted temperature field T pred (x, y, t) satisfies the heat conduction equation and fits the observed data. Heat conduction is described by Fourier's heat conduction law, and its two-dimensional form is Where T is temperature, t is time, α is the thermal diffusivity, and Q(x,y,t) is the heat source term;

[0120] Predicting the future temperature field: After the training is completed, use the trained PINN model to predict the future temperature field distribution, and input the future time t future And spatial coordinates (x, y), output predicted temperature T pred (x,y,t future ).

[0121] In one embodiment, the method of constructing a deep three-dimensional geological structure based on gravity and geophysical data (magnetotelluric sounding MT), and constructing a geothermal anomaly detection model by processing the features provided by the layer to identify geothermal anomalies, includes the following steps:

[0122] Constructing deep 3D geological structure: Using density anomalies obtained from gravity data and resistivity distribution obtained from geophysical data (magnetotelluric sounding MT data), combined with geological prior knowledge, a deep 3D geological structure model is constructed. Kriging interpolation is used to fuse gravity and MT data to obtain the 3D distribution of different underground lithologies and structures, which is expressed as Among them, Z(x) is the geological attribute value of the point to be estimated (such as density, resistivity), Z(x i ) is the geological attribute value of the known point, λ i is the weight coefficient, obtained by minimizing the estimated variance;

[0123] Feature extraction: Extracting features related to geothermal anomalies from the constructed 3D geological structure model, such as fault density, lithology changes, resistivity anomalies, etc.

[0124] Fusion with processing layer features: Fusion of the extracted geological features with the features provided by the processing layer (such as surface temperature inversion, thermal infrared radiation anomaly index, thermal alteration mineral spectral features, etc.) to form a comprehensive feature vector;

[0125] Constructing a geothermal anomaly detection model: Using the fused comprehensive feature vector as input and the known geothermal anomaly points as labels, the support vector machine (SVM) model is trained. The support vector machine (SVM) model is represented as

[0126]

[0127] sty i(w·x i +m)≥1-ξ i ,ξ i ≥0;

[0128] Among them, w is the weight vector, b is the bias term, ξ i is the slack variable, C is the penalty parameter, x i is the input feature vector, y i It’s a label;

[0129] Identification and verification of geothermal anomaly points: Use the trained geothermal anomaly detection model to predict unknown areas and obtain the probability distribution of geothermal anomalies. By comparing with actual geothermal exploration results, the accuracy and reliability of the model can be verified.

[0130] In one embodiment, the specific steps of performing a graded assessment of the anomaly intensity, geological stability, and development potential of the identified geothermal anomaly points based on the geothermal anomaly dynamics, the diffusion pattern of the geothermal anomaly along the fault zone, and the deep three-dimensional geological structure are as follows:

[0131] Assume that the dynamic change characteristic vector F of geothermal anomaly dynamic ;

[0132] Assume that the diffusion characteristic vector F of geothermal anomaly along the fault zone is diffusion ;

[0133] Assume that the geological structure characteristic vector F geology ;

[0134] Combined with the geothermal anomaly dynamic characteristic vector F dynamic , the diffusion characteristic vector F of geothermal anomaly along the fault zone diffusion and geological structure characteristic vector F geology , calculate the abnormal intensity score S intensity , the calculation formula is expressed as S intensity =w1·F dynamic +w2·F diffusion +w3·F geology , where w1, w2, w3 are weight coefficients;

[0135] Combined with the geothermal anomaly dynamic characteristic vector F dynamic , the diffusion characteristic vector F of geothermal anomaly along the fault zone diffusion and geological structure characteristic vector F geology , calculate the geological stability score S stability , the calculation formula is expressed as S stability =1-(w4·F diffusion +w5·F geology ), where w4, w5 are weight coefficients;

[0136] Combined with the abnormal strength score Sintensity and geological stability score S stability , and other development-related factors (such as surface conditions, development costs, etc.), calculate the development potential score S potential , the calculation formula is expressed as S potential =w6·S intensity +w7·S stability +..., where w6, w7... are weight coefficients;

[0137] According to the abnormal intensity score S intensity , geological stability score S stability and development potential score S potential , the identified geothermal anomalies are graded and evaluated, and the geothermal anomalies are divided into three levels: high, medium and low, corresponding to different score ranges:

[0138] Height: S intensity >0.3, S stability >0.7, S potential >0.4;

[0139] Medium: 0.2<S intensity ≤0.3, 0.5<S stability ≤0.7, 0.2<S potential ≤0.4;

[0140] Low: S intensity ≤0.2, S stability ≤0.5, S potential ≤0.2.

[0141] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A geothermal anomaly identification system based on remote sensing detection, characterized in that: include: Data layer: Collection includes thermal infrared remote sensing data, multispectral / hyperspectral remote sensing data, radar data, geological and topographic data, meteorological data, and geophysical data; Processing layer: Preprocesses the collected multi-source data to extract key features. A temporal convolutional network is used to analyze multi-phase thermal infrared sequences to capture the dynamic changes of geothermal anomalies, distinguish between persistent geothermal anomalies and transient events, and combine hyperspectral data with magnetotelluric sounding data to construct "surface-deep" correlation features. Analysis layer: Utilizes a physically constrained graph neural network with a fault network as the graph structure, embedding thermal anomalies, minerals, and terrain features in nodes. A graph propagation model is used to learn the diffusion patterns of geothermal anomalies along fault zones. A physical information neural network is used to constrain the heat conduction equation of the geothermal field as a loss function. A temperature field prediction model is jointly trained. The temperature field prediction model predicts future temperature field distribution based on known geothermal anomaly information and geological conditions. A deep three-dimensional geological structure is constructed based on gravity and geophysical data. A geothermal anomaly detection model is constructed using features provided by the processing layer to identify geothermal anomalies. The identified geothermal anomalies are then graded and evaluated based on the dynamics of geothermal anomalies, the diffusion patterns of geothermal anomalies along fault zones, and the deep three-dimensional geological structure. Output layer: Generates geothermal anomaly probability map, displays the identified geothermal anomaly points in the form of probability map, marks high-probability areas as potential target areas, and combines DEM and geological data to generate a 3D thermal anomaly distribution map.

2. The geothermal anomaly identification system based on remote sensing detection according to claim 1 is characterized in that: The processing layer preprocessing operations include radiation correction, atmospheric correction, geometric alignment and data fusion, and the key feature extraction includes surface temperature inversion, thermal infrared radiation anomaly index calculation, thermal alteration mineral spectrum feature extraction, terrain feature extraction and fault density analysis.

3. The geothermal anomaly identification system based on remote sensing detection according to claim 1 is characterized in that: The specific steps of using a temporal convolutional network to analyze multi-temporal thermal infrared sequences, capture the dynamic changes of geothermal anomalies, and distinguish between persistent geothermal anomalies and transient events are as follows: Data preparation: Collect multi-temporal thermal infrared remote sensing data and preprocess the data, including radiation correction, atmospheric correction, and geometric registration; Constructing a temporal convolutional network: Designing a temporal convolutional network architecture, including convolutional layers, residual blocks, causal convolutions, and dilated convolutions, determining hyperparameters, initializing network weights, selecting Adam and mean square error (MSE), and expressing the causal convolution in the residual block of the temporal convolutional network as follows: Among them, y t is the output at time step t, x t-i is the input of time step ti, w i is the convolution kernel weight, k is the convolution kernel size; Training the temporal convolutional network model: Divide the preprocessed thermal infrared data into a training set and a validation set, input the data into the temporal convolutional network model, perform forward propagation to calculate the output, calculate the mean square error (MSE), and update the network weights through backpropagation. Repeat the training process until the model achieves satisfactory performance on the validation set. Capturing dynamic changes in geothermal anomalies: Applying the trained temporal convolutional network model to new multi-temporal thermal infrared data, analyzing the model output, identifying dynamic patterns in geothermal anomalies, and distinguishing persistent geothermal anomalies from transient events by comparing outputs at different time steps. Evaluation and Validation: Verify the prediction results of the temporal convolutional network model, calculate evaluation metrics, evaluate model performance, and adjust model parameters or improve model architecture based on the evaluation results.

4. The geothermal anomaly identification system based on remote sensing detection according to claim 1 is characterized in that: The combined hyperspectral data and magnetotelluric sounding data construct the "surface-deep" correlation feature. The following steps are included: Determine the associated characteristic parameters: Select characteristic parameters related to hydrothermal activity from the hyperspectral data; Establishing a correlation model: Linear regression is used to establish a correlation model between hyperspectral features and MT features.

5. The geothermal anomaly identification system based on remote sensing detection according to claim 4 is characterized in that: The physical constraint-based graph neural network uses a fault network as a graph structure, embeds thermal anomalies, minerals, and terrain features in nodes, and learns the diffusion law of geothermal anomalies along fault zones through a graph propagation model. The specific steps are as follows: Constructing the graph structure: Define nodes: The fault intersections, endpoints, and significant geothermal anomaly points within the study area are used as nodes of the graph; Define edges: Define edges based on fault connectivity and the spatial proximity between geothermal anomalies. If there is a direct fault connection between two nodes or the spatial distance is less than a certain threshold, an edge is established between the two nodes. Node feature embedding: Thermal anomaly feature: Extract the thermal infrared radiation anomaly index at each node as the thermal anomaly feature; Mineral characteristics: Use hyperspectral data to identify the mineral composition of the node area, and use the ratio of the main mineral types as the mineral characteristics; Terrain features: Extract the terrain elevation and slope of the node as terrain features; Graph propagation model: Define the graph convolution layer: Use the graph convolution network layer to propagate and aggregate node features. The calculation formula of the graph convolution layer is Among them, H (l) is the node feature matrix of the lth layer, is the normalized adjacency matrix, W (l) is a learnable weight matrix, σ is the activation function; Multi-layer graph convolution: stacking multiple graph convolution layers to learn node representations and the diffusion patterns of geothermal anomalies along fault zones; Learn about the diffusion patterns of geothermal anomalies along fault zones: Define the loss function: Use the mean square error loss function to measure the difference between the geothermal anomaly diffusion probability predicted by the model and the actual observed value; Training model: Use training data to train the graph propagation model, and update the model parameters through the backpropagation algorithm to minimize the loss function; Prediction and Evaluation: Predicting the diffusion of geothermal anomalies: Use the trained graph propagation model to predict the probability of geothermal anomaly diffusion for all nodes in the study area.

6. The geothermal anomaly identification system based on remote sensing detection according to claim 1 is characterized in that: The specific steps of using the physical information neural network to use the geothermal field heat conduction equation as a loss function constraint and jointly train the temperature field prediction model, which predicts the future temperature field distribution based on known geothermal anomaly information and geological conditions are as follows: Data preparation: Collect geothermal anomaly information and geological condition data. If there is observation data, prepare temperature data of the observation points; Construct PINN: Design a neural network with input (x, y, t) and output Tpred(x, y, t). Define the loss function, including physical loss and data loss. The physical loss calculation is expressed as Where N is the number of sampling points, (x i ,y i ,t i ) are the coordinates and time of the sampling points; The data loss calculation is expressed as Where M is the number of observation points, T ots (x j ,y j ,t j ) is the temperature at the observation point; The total loss is expressed as L = L physics +λL data , where λ is the weight of data loss; Training PINN: Use Adam to minimize the total loss L. During the training process, continuously update the parameters of the neural network so that the predicted temperature field T pred (x, y, t) satisfies the heat conduction equation and fits the observed data. Heat conduction is described by Fourier's heat conduction law, and its two-dimensional form is Where T is temperature, t is time, α is the thermal diffusivity, and Q(x,y,t) is the heat source term; Predicting the future temperature field: After the training is completed, use the trained PINN model to predict the future temperature field distribution, and input the future time t future And spatial coordinates (x, y), output predicted temperature T pred (x,y,t future ).

7. The geothermal anomaly identification system based on remote sensing detection according to claim 1 is characterized in that: The method of constructing a deep three-dimensional geological structure based on gravity and geophysical data and constructing a geothermal anomaly detection model to identify geothermal anomaly points by processing the features provided by the layer includes the following steps: Constructing deep 3D geological structure: Using density anomalies obtained from gravity data and resistivity distribution obtained from geophysical data, combined with geological prior knowledge, a deep 3D geological structure model is constructed. Kriging interpolation is used to fuse gravity and MT data to obtain the 3D distribution of different underground lithologies and structures, which is expressed as Among them, Z(x) is the geological attribute value of the point to be estimated, Z(x i ) is the geological attribute value of the known point, λ i is the weight coefficient, obtained by minimizing the estimated variance; Feature extraction: Extracting features related to geothermal anomalies from the constructed 3D geological structure model; Fusion with processing layer features: Fusion of the extracted geological features with the features provided by the processing layer to form a comprehensive feature vector; Constructing a geothermal anomaly detection model: Using the fused comprehensive feature vector as input and the known geothermal anomaly points as labels, the support vector machine model is trained. The support vector machine model is represented as s.t.y i (w·x i +m)≥1-ξ i ,ξ i ≥0; Among them, w is the weight vector, b is the bias term, ξ i is the slack variable, C is the penalty parameter, x i is the input feature vector, y i It’s a label; Identification and verification of geothermal anomaly points: Use the trained geothermal anomaly detection model to predict unknown areas and obtain the probability distribution of geothermal anomalies. By comparing with actual geothermal exploration results, the accuracy and reliability of the model can be verified.

8. The geothermal anomaly identification system based on remote sensing detection according to claim 1 is characterized in that: The specific steps of performing a graded assessment of the identified geothermal anomaly points based on the geothermal anomaly dynamics, the diffusion patterns of the geothermal anomaly along the fault zone, and the deep three-dimensional geological structure are as follows: Assume that the dynamic change characteristic vector F of geothermal anomaly dynamic ; Assume that the diffusion characteristic vector F of geothermal anomaly along the fault zone is diffusion ; Assume that the geological structure characteristic vector F geology ; Combined with the geothermal anomaly dynamic characteristic vector F dynamic , the diffusion characteristic vector F of geothermal anomaly along the fault zone diffusion and geological structure characteristic vector F geology , calculate the abnormal intensity score S intensity , the calculation formula is expressed as S intensity =w1·F dynamic +w2·F diffusion +w3·F geology , where w1, w2, w3 are weight coefficients; Combined with the geothermal anomaly dynamic characteristic vector F dynamic , the diffusion characteristic vector F of geothermal anomaly along the fault zone diffusion and geological structure characteristic vector F geology , calculate the geological stability score S stability , the calculation formula is expressed as S stability =1-(w4·F diffusion +w5·F geology ), where w4, w5 are weight coefficients; Combined with the abnormal strength score S intensity and geological stability score S stability , and other development-related factors to calculate the development potential score S potential , the calculation formula is expressed as S potential =w6·S intensity +w7·S stability +..., where w6, w7... are weight coefficients; According to the abnormal intensity score S intensity , geological stability score S stability and development potential score S potential , the identified geothermal anomalies are graded and evaluated, and the geothermal anomalies are divided into three levels: high, medium and low, corresponding to different score ranges: Height: S intensity >0.3, S stability >0.7, S potential >0.4; Medium: 0.2<S intensity ≤0.3, 0.5<S stability ≤0.7, 0.2<S potential ≤0.4; Low: S intensity ≤0.2, S stability ≤0.5, S potential ≤0.2.

Images