Shallow geothermal energy reserve monitoring system

Abstract

The invention discloses a shallow geothermal energy reserve monitoring system, and the system comprises a sensing layer which is used for deploying a multi-parameter sensor network, covering a target region, and obtaining a ground temperature field, hydrogeology, thermophysical parameters and environmental parameters; the transmission layer is used for carrying out data transmission by adopting a wired and wireless hybrid communication mode; the platform layer is used for performing data fusion and quality control on the data acquired by the sensing layer, constructing a three-dimensional geological map structure, taking sensor nodes as map vertexes, taking stratum permeability and lithology as edge attributes, realizing spatial feature aggregation by utilizing a GraphSAGE algorithm, introducing a time convolutional network to capture time sequence dependence of underground water flow, performing spatio-temporal joint interpolation, and obtaining a three-dimensional geologic map structure; calculating geothermal energy reserves of the target area, combining an underground water flow model and a heat transfer equation, fusing a multi-scale map neural network interpolation result, predicting long-term reserve changes, and performing development potential evaluation on the target area; and the application layer provides a user terminal, an early warning system and a management interface.

Description

Technical Field

[0001] The present invention relates to the technical field of shallow geothermal energy, and in particular to a shallow geothermal energy reserve monitoring system. Background Art

[0002] The shallow geothermal energy monitoring technology has been gradually mature, mainly relying on high-precision temperature measuring instruments (such as fiber Bragg gratings, platinum resistors, etc.), with a temperature measurement range covering -271°C to 1000°C and an accuracy of ±0.1°C to ±0.5°C. The monitoring content covers the dynamic changes of groundwater level, water volume, water quality, water temperature and ground temperature, and forms systematic data support through annual reports and achievement reports. Monitor the operation status of the underground heat exchange system, and grasp key indicators such as ground temperature changes and thermal efficiency in real time, prevent the phenomena of "heat accumulation" or "cold accumulation", and avoid energy waste and geological environment damage.

[0003] In the prior art, the Kriging interpolation method is used to generate a three-dimensional model of the regional temperature field / water level field to provide spatial distribution information for geothermal energy reserve assessment. However, the Kriging interpolation method has certain assumptions about the spatial distribution and correlation of data. If the actual data does not meet these assumptions, it may lead to inaccurate interpolation results. Therefore, a shallow geothermal energy reserve monitoring system is proposed. Summary of the Invention

[0004] The purpose of the present invention is to solve the deficiencies existing in the prior art, and a shallow geothermal energy reserve monitoring system is proposed.

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

[0006] A shallow geothermal energy reserve monitoring system includes:

[0007] Perception layer: Deploy a multi-parameter sensor network to cover the target area, which includes geothermal wells, soil layers, and aquifers. The sensor selection covers distributed temperature sensing (DTS) fiber optic temperature measurement cables, multi-parameter hydrological probes (integrated with pressure, temperature, and conductivity sensors), thermal response testers, and surface meteorological stations to obtain the ground temperature field, hydrogeology, thermal physical properties parameters, and environmental parameters;

[0008] Transmission layer: Adopt a hybrid communication method of wired (such as RS-485, fiber optic) and wireless (such as LoRa, NB-IoT, 5G) for data transmission;

[0009] Platform layer: It performs data fusion and quality control on the data collected by the perception layer, constructs a three-dimensional geological map structure, takes sensor nodes as graph vertices, and formation permeability and lithology as edge attributes. It uses the GraphSAGE algorithm to achieve spatial feature aggregation, introduces a temporal convolutional network (TCN) to capture the temporal dependence of groundwater flow, performs spatio-temporal joint interpolation, calculates the geothermal energy reserves in the target area, combines the groundwater flow model and the heat transfer equation, and incorporates the interpolation results of a multi-scale graph neural network (GNN) to predict long-term reserve changes and evaluate the development potential of the target area;

[0010] Application layer: It provides user terminals, warning systems, and management interfaces.

[0011] The above technical solution further includes:

[0012] Further, the platform layer constructs a three-dimensional geological map structure, takes sensor nodes as graph vertices, and formation permeability and lithology as edge attributes. The weight of the edge is set according to the differences in formation permeability and lithology. The greater the difference in formation permeability between node A and node B, the higher the weight of the edge between node A and node B is set, reflecting the impact of the difference on spatial feature aggregation.

[0013] Further, the implementation of spatial feature aggregation using the GraphSAGE algorithm includes the following steps:

[0014] Data preparation: Each sensor node collects the geothermal field, hydrogeology, thermal physical properties parameters, and environmental parameters at its own location;

[0015] Node sampling: For each sensor node, determine the sampling range of its neighbor nodes and determine the neighbor nodes. For example, set the sampling radius to 10 meters. For node A, find all other sensor nodes within 10 meters of it as neighbor nodes. Suppose 3 neighbor nodes B, C, and D are found within this range;

[0016] Feature extraction: Extract the features of each node and its neighbor nodes. The features include temperature, formation permeability, and lithology, and combine the features into a feature vector;

[0017] Feature aggregation: Use the aggregation function of the GraphSAGE algorithm to aggregate the features of the neighbor nodes. The aggregated feature of node A is calculated by the following formula:

[0018] Multi-layer iteration: The aggregated feature obtained by the above aggregation As a new feature of node A, sampling and aggregation operations are performed again. A multi-layer GraphSAGE algorithm is set up, and the above-mentioned sampling, feature extraction, and aggregation steps are repeated for each layer, continuously updating the embedded representation of the node. After multiple iterations, the final embedded representation of each node is obtained, and the embedded representation contains the spatial feature information of the node.

[0019] Furthermore, introducing a Temporal Convolutional Network (TCN) to capture the temporal dependence of groundwater flow includes the following steps;

[0020] Data preparation: Sensor nodes continuously collect time series data such as temperature and water level;

[0021] Data preprocessing: Preprocess the collected time series data, including normalization to eliminate the dimensional differences between different data;

[0022] Constructing the TCN model: The TCN mainly consists of causal convolution, dilated convolution, and residual connections. Causal convolution ensures that the output only depends on the data at the current time and before, avoiding information leakage. Dilated convolution expands the receptive field by increasing the spacing of the convolution kernel, enabling the model to capture longer-term dependencies. Residual connections help alleviate the vanishing gradient problem and accelerate model training. Let the TCN model have L layers, each layer has d filters, the convolution kernel size is k, and the dilation factor is d l , the dilation factor of the l-th layer;

[0023] Forward propagation calculation: For the input time series data T norm , after the causal convolution and dilated convolution operations of the first layer of TCN, the output is where W1 is the convolution kernel weight of the first layer, b1 is the bias term, *d l represents the convolution operation with a dilation factor of d l , ReLU is the activation function. After iterative calculations of multiple layers of TCN, the final output is h T .

[0024] Furthermore, the specific steps of the spatio-temporal joint interpolation:

[0025] Feature extraction: Use the GraphSAGE algorithm to obtain the spatial feature embedding h S , and obtain the temporal feature h through the TCN T ;

[0026] Feature fusion: Fuse the spatial feature embedding h S and the temporal feature h T , and use the concatenation operation to obtain the joint feature h = [h S ; h T ;

[0027] Determine weights: For the position (x, y, z) to be interpolated, determine the weights w of the surrounding sensor nodes. i , where the weights are determined according to the distance between the nodes and the interpolation position, and the inverse distance weighting method is adopted. The formula is d i is the distance between node i and the interpolation position (x, y, z), and p is a positive integer, taking p = 2;

[0028] Interpolation calculation: Use the fused features for interpolation calculation. According to the joint features of the surrounding sensor nodes, perform interpolation by weighted averaging. Let the joint feature of the surrounding node i be h i , and the weight be w i , then the interpolated temperature value T (x,y,z) is expressed as: T (x,y,z) = ∑ i w i ·f(h i ), where f() is a mapping function that maps the joint feature to the temperature value. In practical applications, a neural network can be trained to learn this mapping function.

[0029] Furthermore, calculating the shallow geothermal energy reserves based on the probability box (p-box) theory and the stochastic finite element method (SFEM) includes the following steps:

[0030] Determine the key parameters and their interval distributions: According to the geological exploration report and existing research, determine the interval distributions of the key parameters;

[0031] Construct a probability box (p-box) model: For each key parameter, construct a probability box model. The probability box model describes the probability distribution of the parameter within a given interval. Let the parameter k follow a uniform distribution within the interval [k min , k max , and its probability density function f(k) is

[0032] Establish a stochastic finite element model: Use the finite element method to establish a calculation model for the shallow geothermal energy reserves. Discretize the target area into a finite number of elements, and randomly assign the thermal physical properties (such as thermal conductivity, porosity) of each element according to the probability box model;

[0033] Calculate the geothermal energy reserves: According to the geothermal energy reserve calculation formula Q = ρ·c p ·V·ΔT·η, in the stochastic finite element model, for each element, calculate the stored thermal energy Q i , where ρ is the density of the rock and soil, c p is the specific heat capacity, V i$V_i$ is the effective heat storage volume of the $i$-th unit, $\Delta T$ is the available temperature difference, $\eta$ is the heat extraction efficiency coefficient, and the geothermal energy storage $Q$ in the entire target area is the sum of the storage of each unit;

[0034] Calculating the confidence interval of the storage through multiple simulations: Repeatedly establish a stochastic finite element model and calculate the geothermal energy storage multiple times (such as $m$ times) to obtain $m$ geothermal energy storage values. Conduct statistical analysis on these storage values, calculate their mean and standard deviation, and determine the confidence interval of the storage based on the normal distribution hypothesis.

[0035] Furthermore, the specific steps for predicting long-term storage changes by combining the groundwater flow model and the heat transfer equation and incorporating the interpolation results of the multi-scale graph neural network are as follows:

[0036] Data preparation: Collect the geothermal temperature field, hydrogeology, thermal physical properties parameters, and environmental parameters of the target area, and use the multi-scale graph neural network for interpolation to obtain the spatial distribution data;

[0037] Model establishment: Combine the groundwater flow model and the heat transfer equation to construct a coupled model. The groundwater flow model describes the flow of groundwater, and its basic equation is the continuity equation of groundwater flow where $K$ x $, K$ y $, K$ z are the permeability coefficients in the $x$, $y$, and $z$ directions respectively, $h$ is the hydraulic head, $Q$ is the source-sink term, $S$ s is the storage coefficient, $t$ is the time. The heat transfer equation describes the heat transfer process, and the heat transfer equation is where $\rho$ is the density of rock and soil, $c$ p is the specific heat capacity, $T$ is the temperature, $k$ is the thermal conductivity, $u$ is the groundwater seepage velocity, and $q$ is the heat source term;

[0038] Model parameter setting: According to the geological exploration data, set the permeability coefficients of different strata, use the interpolated thermal physical properties parameters, and set the groundwater level boundary and temperature boundary;

[0039] Simulation prediction: Input the set parameters into the coupled model, run the model for simulation, and obtain the changes in geothermal energy storage at different time points.

[0040] Furthermore, the specific steps for evaluating the development potential of the target area are as follows:

[0041] Analysis of storage changes: Analyze the change trend of geothermal energy storage based on the prediction results;

[0042] Combined with a real-time cost-benefit digital twin system: access the energy market API to obtain dynamic economic parameters such as electricity prices and carbon trading prices, and use Monte Carlo risk simulation to consider factors such as equipment depreciation rates and policy subsidy uncertainties to generate the probability distribution of the net present value (NPV) of reserve development;

[0043] Evaluate the development potential: According to the probability distribution of the net present value (NPV) of reserve development and the reserve change trend, evaluate the development potential. If the expected value of NPV in this area is relatively high and the reserve reduction rate is relatively slow, it has a relatively high development potential; otherwise, the development potential is relatively low.

[0044] The present invention has the following beneficial effects:

[0045] 1. In the present invention, in the platform layer, by constructing a three-dimensional geological map structure, taking the sensor nodes as the graph vertices and the formation permeability and lithology as the edge attributes, and using the GraphSAGE algorithm to achieve spatial feature aggregation, this processing method can more effectively integrate multi-source heterogeneous data, reduce data redundancy and noise interference, improve the accuracy and reliability of the data, introduce a temporal convolutional network to capture the temporal dependence of groundwater flow, and perform spatio-temporal joint interpolation, which can more accurately reflect the dynamic changes of the underground geological structure and improve the accuracy of data interpolation.

[0046] 2. In the present invention, by combining the groundwater flow model with the heat transfer equation to ensure the accuracy of reserve assessment and integrating the interpolation results of the multi-scale graph neural network, the platform layer can predict the long-term reserve changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a system block diagram of a shallow geothermal energy reserve monitoring system proposed by the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0049] Please refer to Figure 1 As shown, the present invention is a shallow geothermal energy reserve monitoring system, including:

[0050] Perception layer: Deploy a multi-parameter sensor network to cover the target area, which includes geothermal wells, soil layers, and aquifers. The sensor selection covers distributed temperature sensing (DTS) fiber optic temperature measurement cables, multi-parameter hydrographic probes (integrated with pressure, temperature, and conductivity sensors), thermal response testers, and surface meteorological stations to obtain geothermal fields, hydrogeology, thermal physical properties parameters, and environmental parameters;

[0051] Transport layer: Adopt a hybrid communication method of wired (such as RS-485, fiber optic) and wireless (such as LoRa, NB-IoT, 5G) for data transmission;

[0052] Platform layer: Perform data fusion and quality control on the data collected by the perception layer, construct a three-dimensional geological map structure, use sensor nodes as graph vertices, and formation permeability and lithology as edge attributes. Utilize the GraphSAGE algorithm to achieve spatial feature aggregation, introduce a temporal convolutional network (TCN) to capture the temporal dependence of groundwater flow, perform spatio-temporal joint interpolation, calculate the geothermal energy reserves in the target area, combine the groundwater flow model and the heat transfer equation, and integrate the interpolation results of the multi-scale graph neural network (GNN) to predict long-term reserve changes and evaluate the development potential of the target area;

[0053] Application layer: Provide user terminals, warning systems, and management interfaces.

[0054] In one embodiment, the platform layer constructs a three-dimensional geological map structure, uses sensor nodes as graph vertices, and formation permeability and lithology as edge attributes. The weight of the edge is set according to the differences in formation permeability and lithology. The greater the difference in formation permeability between node A and node B, the higher the weight of the edge between node A and node B is set, reflecting the influence of the difference on spatial feature aggregation.

[0055] In one embodiment, the implementation of spatial feature aggregation using the GraphSAGE algorithm includes the following steps:

[0056] Data preparation: Each sensor node collects geothermal field, hydrogeology, thermal physical properties parameters, and environmental parameters at its own location;

[0057] Node sampling: For each sensor node, determine the sampling range of its neighbor nodes and determine the neighbor nodes. For example, set the sampling radius to 10 meters. For node A, find all other sensor nodes within 10 meters of it as neighbor nodes. Assume that 3 neighbor nodes B, C, and D are found within this range;

[0058] Feature extraction: Extract the features of each node and its neighbor nodes. The features include temperature, formation permeability, and lithology, and combine the features into a feature vector;

[0059] Feature Aggregation: Use the aggregation function of the GraphSAGE algorithm to aggregate the features of neighboring nodes. The aggregated feature of node A is calculated by the following formula:

[0060] Multi-Layer Iteration: Use the above-aggregated feature as the new feature of node A, and perform sampling and aggregation operations again. Set up a multi-layer GraphSAGE algorithm, and repeat the above sampling, feature extraction, and aggregation steps for each layer to continuously update the embedding representation of the node. After multi-layer iteration, obtain the final embedding representation of each node, and the embedding representation contains the spatial feature information of the node.

[0061] In one embodiment, introducing a Temporal Convolutional Network (TCN) to capture the temporal dependence of groundwater flow includes the following steps;

[0062] Data Preparation: Sensor nodes continuously collect time series data such as temperature and water level;

[0063] Data Preprocessing: Preprocess the collected time series data, including normalization to eliminate the dimensional difference between different data;

[0064] Construct a TCN Model: TCN is mainly composed of causal convolution, dilated convolution, and residual connection. Causal convolution ensures that the output only depends on the data at the current time and before, avoiding information leakage. Dilated convolution expands the receptive field by increasing the interval of the convolution kernel, enabling the model to capture longer-term dependencies. Residual connection helps to alleviate the vanishing gradient problem and accelerate model training. Assume that the TCN model has L layers, each layer has d filters, the convolution kernel size is k, and the dilation factor is d l , the dilation factor of the l-th layer;

[0065] Forward Propagation Calculation: For the input time series data T norm , after the causal convolution and dilated convolution operations of the first layer of TCN, the output is where W1 is the convolution kernel weight of the first layer, b1 is the bias term, and *d l represents the convolution operation with a dilation factor of d l , ReLU is the activation function. After the iterative calculation of multiple layers of TCN, the final output is h T .

[0066] In one embodiment, the specific steps of spatio-temporal joint interpolation are as follows:

[0067] Feature Extraction: Use the GraphSAGE algorithm to obtain the spatial feature embedding h S , and obtain the temporal feature h T through TCN;

[0068] Feature fusion: Embed the spatial feature into h S and the temporal feature h T for fusion, and perform a concatenation operation to obtain the joint feature h = [h S ; h T ;

[0069] Determine weights: For the position (x, y, z) to be interpolated, determine the weights w i of its surrounding sensor nodes. The weights are determined according to the distance between the nodes and the interpolation position, and the inverse distance weighting method is adopted. The formula is d i is the distance between node i and the interpolation position (x, y, z), and p is a positive integer, taking p = 2;

[0070] Interpolation calculation: Use the fused feature for interpolation calculation. According to the joint features of the surrounding sensor nodes, perform interpolation by weighted average. Let the joint feature of the surrounding node i be h i , and the weight be w i , then the interpolated temperature value T (x,y,z) is expressed as: T (x,y,z) = ∑ i w i · f(h i ), where f() is a mapping function that maps the joint feature to the temperature value. In practical applications, this mapping function can be learned by training a neural network.

[0071] In one embodiment, calculating the shallow geothermal energy reserve based on the probability box (p-box) theory and the stochastic finite element method (SFEM) includes the following steps:

[0072] Determine the key parameters and interval distributions: According to the geological exploration report and existing research, determine the interval distributions of the key parameters;

[0073] Construct a probability box (p-box) model: For each key parameter, construct a probability box model. The probability box model describes the probability distribution of the parameter within a given interval. Let the parameter k follow a uniform distribution within the interval [k min , k max , and its probability density function f(k) is

[0074] Establish a stochastic finite element model: Use the finite element method to establish a calculation model for the shallow geothermal energy reserve, discretize the target area into a finite number of elements, and randomly assign the thermal physical properties (such as thermal conductivity, porosity) of each element according to the probability box model;

[0075] Calculating geothermal energy reserves: According to the geothermal energy reserve calculation formula Q = ρ·c p ·V·ΔT·η, in the stochastic finite element model, for each element, calculate the thermal energy Q stored in it i , where ρ is the rock and soil density, c p is the specific heat capacity, V i is the effective heat storage volume of the i-th element, ΔT is the available temperature difference, η is the heat extraction efficiency coefficient, and the geothermal energy reserve Q of the entire target area is the sum of the reserves of each element;

[0076] Calculating the confidence interval of reserves through multiple simulations: Repeatedly establish the stochastic finite element model and calculate the geothermal energy reserves multiple times (such as m times) to obtain m geothermal energy reserve values. Conduct statistical analysis on these reserve values, calculate their mean and standard deviation, and determine the reserve confidence interval based on the normal distribution hypothesis.

[0077] In one embodiment, the specific steps for predicting long-term reserve changes by combining the groundwater flow model with the heat transfer equation and incorporating the interpolation results of the multi-scale graph neural network are as follows:

[0078] Data preparation: Collect the geothermal temperature field, hydrogeology, thermal physical properties parameters, and environmental parameters of the target area, and use the multi-scale graph neural network for interpolation to obtain spatial distribution data;

[0079] Model establishment: Combine the groundwater flow model with the heat transfer equation to construct a coupled model. The groundwater flow model describes the flow of groundwater, and its basic equation is the continuity equation of groundwater flow where K x , K y , K z are the permeability coefficients in the x, y, and z directions respectively, h is the hydraulic head, Q is the source-sink term, S s is the storage coefficient, t is the time, and the heat transfer equation describes the heat transfer process. The heat transfer equation is where ρ is the rock and soil density, c p is the specific heat capacity, T is the temperature, k is the thermal conductivity, u is the groundwater seepage velocity, and q is the heat source term;

[0080] Model parameter setting: According to the geological exploration data, set the permeability coefficients of different strata, use the interpolated thermal physical properties parameters, and set the groundwater level boundary and temperature boundary;

[0081] Simulation prediction: Input the set parameters into the coupled model, run the model for simulation, and obtain the changes in geothermal energy reserves at different time points.

[0082] In one embodiment, the specific steps for evaluating the development potential of the target area are as follows:

[0083] Analysis of reserve changes: Analyze the changing trend of geothermal energy reserves according to the prediction results;

[0084] Combined with the real-time cost-benefit digital twin system: Access the energy market API to obtain dynamic economic parameters such as electricity prices and carbon trading prices, and use Monte Carlo risk simulation to consider factors such as equipment depreciation rates and policy subsidy uncertainties to generate the probability distribution of the net present value (NPV) of reserve development;

[0085] Evaluate the development potential: Evaluate the development potential based on the probability distribution of the net present value (NPV) of reserve development and the changing trend of reserves. If the expected value of NPV in this area is relatively high and the reserve reduction rate is relatively slow, then it has a high development potential; otherwise, the development potential is relatively low.

[0086] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A shallow geothermal energy storage monitoring system, characterized in that, Including: Perception layer: Deploy a multi-parameter sensor network to cover the target area, which includes geothermal wells, soil layers, and aquifers, and obtain the geothermal temperature field, hydrogeology, thermal physical properties parameters, and environmental parameters; Transport layer: Adopt a hybrid wired and wireless communication method for data transmission; Platform layer: Perform data fusion and quality control on the data collected by the perception layer, construct a three-dimensional geological map structure, use sensor nodes as graph vertices, and formation permeability and lithology as edge attributes. Utilize the GraphSAGE algorithm to achieve spatial feature aggregation, introduce a temporal convolutional network to capture the temporal dependence of groundwater flow, conduct spatio-temporal joint interpolation, calculate the geothermal energy reserves in the target area, combine the groundwater flow model and the heat transfer equation, and incorporate the interpolation results of the multi-scale graph neural network to predict long-term reserve changes and evaluate the development potential of the target area; Application layer: Provide user terminals, warning systems, and management interfaces.

2. The shallow geothermal energy storage monitoring system according to claim 1, characterized in that The platform layer constructs a three-dimensional geological map structure, uses sensor nodes as graph vertices, and formation permeability and lithology as edge attributes. The weight of the edge is set according to the differences in formation permeability and lithology. The greater the difference in formation permeability between node A and node B, the higher the weight of the edge between node A and node B is set, reflecting the impact of the difference on spatial feature aggregation.

3. A shallow geothermal energy storage monitoring system according to claim 1, characterized in that, The implementation of spatial feature aggregation using the GraphSAGE algorithm includes the following steps: Data preparation: Each sensor node collects the geothermal temperature field, hydrogeology, thermal physical properties parameters, and environmental parameters at its own location; Node sampling: For each sensor node, determine the sampling range of its neighbor nodes and determine the neighbor nodes; Feature extraction: Extract the features of each node and its neighbor nodes. The features include temperature, formation permeability, and lithology, and combine the features into a feature vector; Feature Aggregation: Aggregate the features of neighboring nodes using the aggregation function of the GraphSAGE algorithm. The aggregated features of node A are calculated by the following formula: Multi-layer iteration: Use the aggregated features obtained from the above aggregation as the new features of node A, and perform sampling and aggregation operations again. Set the multi-layer GraphSAGE algorithm, and repeat the above sampling, feature extraction, and aggregation steps for each layer to continuously update the embedding representation of the nodes. After multi-layer iteration, obtain the final embedding representation of each node, and the embedding representation contains the spatial feature information of the nodes.

4. A shallow geothermal energy storage monitoring system according to claim 3, characterized in that, The introduction of a temporal convolutional network to capture the temporal dependence of groundwater flow includes the following steps; Data preparation: Sensor nodes continuously collect time series data; Data preprocessing: Preprocess the collected time series data; Construct the TCN model: The TCN is mainly composed of causal convolution, dilated convolution, and residual connections. Assume that the TCN model has L layers, each layer has d filters, the convolution kernel size is k, and the dilation factor is d l , the dilation factor of the l-th layer; Forward propagation calculation: For the input time series data T norm , after the causal convolution and dilated convolution operations of the first layer of TCN, the output is where W1 is the convolutional kernel weight of the first layer, b1 is the bias term, and *d l represents the convolutional operation with a dilation factor of d l , ReLU is the activation function. After the iterative calculation of multiple layers of TCN, the final output is h T .

5. A shallow geothermal energy storage monitoring system according to claim 4, characterized in that, The specific steps of spatio-temporal joint interpolation: Feature extraction: Use the GraphSAGE algorithm to obtain the spatial feature embedding h S , and obtain the temporal feature h through TCN T ; Feature Fusion: Embed the spatial feature into h S and the temporal feature h T for fusion, and use the concatenation operation to obtain the joint feature h = [h S ; h T ; Determine the weight: For the position (x, y, z) to be interpolated, determine the weight w of the surrounding sensor nodes i , and the weight is determined according to the distance between the node and the interpolation position. The inverse distance weighting method is adopted, and the formula is d i is the distance between node i and the interpolation position (x, y, z), and p is a positive integer, taking p = 2; Interpolation calculation: Use the fused features for interpolation calculation. Based on the combined features of surrounding sensor nodes, perform interpolation by weighted average. Let the combined feature of surrounding node i be h i , with the weight being w i . Then the interpolated temperature value T (x,y,z) is expressed as: T (x,y,z) = ∑ i w i ·f(h i ), where f() is a mapping function that maps the combined feature to the temperature value.

6. The shallow geothermal energy storage monitoring system according to claim 1, wherein The calculation of shallow geothermal energy reserves based on the probability box theory and the stochastic finite element method includes the following steps: Determine the key parameters and interval distributions: According to the geological exploration report and existing research, determine the interval distributions of the key parameters; Construct a probability box model: For each key parameter, construct a probability box model. Let the parameter k follow a uniform distribution in the interval [k min , k mar , and its probability density function f(k) is Establish a stochastic finite element model: Use the finite element method to establish a calculation model for shallow geothermal energy reserves, discretize the target area into a finite number of elements, and randomly assign the thermal physical properties parameters of each element according to the probability box model; Calculating geothermal energy reserves: According to the geothermal energy reserve calculation formula Q = ρ·c p ·V·ΔT·η, in the stochastic finite element model, for each element, calculate the thermal energy Q stored in it i , where ρ is the density of rock and soil, c p is the specific heat capacity, V i is the effective heat storage volume of the i-th element, ΔT is the available temperature difference, η is the heat extraction efficiency coefficient, and the geothermal energy reserve Q of the entire target area is the sum of the reserves of each element; Simulate and calculate the reserve confidence interval multiple times: Repeatedly establish the stochastic finite element model and calculate the geothermal energy reserves multiple times to obtain m geothermal energy reserve values. Conduct statistical analysis on these reserve values, calculate their mean and standard deviation, and determine the reserve confidence interval according to the normal distribution hypothesis.

7. A shallow geothermal energy storage monitoring system according to claim 1, characterized in that, The specific steps of combining the groundwater flow model and the heat transfer equation, incorporating the interpolation results of the multi-scale graph neural network, and predicting long-term reserve changes: Data preparation: Collect the geothermal temperature field, hydrogeology, thermal physical properties parameters, and environmental parameters of the target area, and use the multi-scale graph neural network for interpolation to obtain spatial distribution data; Model establishment: Combine the groundwater flow model and the heat transfer equation to construct a coupled model. The groundwater flow model describes the flow of groundwater, and its basic equation is the continuity equation of groundwater flow where K x , K y , K z are the permeability coefficients in the x, y, and z directions respectively, h is the hydraulic head, Q is the source-sink term, S s is the storage coefficient, t is the time, and the heat transfer equation describes the heat transfer process. The heat transfer equation is where ρ is the density of rock and soil, c p is the specific heat capacity, T is the temperature, k is the thermal conductivity, u is the groundwater seepage velocity, and q is the heat source term; Model parameter setting: According to the geological exploration data, set the permeability coefficients of different strata, obtain the thermal physical properties parameters by interpolation, and set the groundwater level boundary and temperature boundary; Simulation prediction: Input the set parameters into the coupled model, run the model for simulation, and obtain the changes in geothermal energy reserves at different time points.

8. A shallow geothermal energy reserve monitoring system according to claim 7, characterized in that, The specific steps for evaluating the development potential of the target area are as follows: Reserve change analysis: Analyze the change trend of geothermal energy reserves according to the prediction results; Combined with the real-time cost-benefit digital twin system: Access the energy market API, obtain dynamic economic parameters, and use Monte Carlo risk simulation to generate the probability distribution of the net present value of reserve development; Evaluate the development potential: Evaluate the development potential according to the probability distribution of the net present value of reserve development and the reserve change trend.

Images