Middle-deep layer geothermal pipe group multi-field coupling simulation method based on deep learning

Abstract

The invention discloses a deep learning-based middle-deep layer geothermal pipe group multi-field coupling simulation method, which comprises the following steps: carrying out numerical simulation cross mutual verification to obtain a middle-deep layer buried pipe heat pump system model, and carrying out system configuration by adopting a Monte Carlo method to obtain a first parameter environment set; screening the first parameter environment set according to a parameter environment function to obtain a second parameter environment set, performing data augmentation to obtain mid-deep layer geothermal pipe group simulation data, constructing a mid-deep layer geothermal pipe group multi-field coupling deep learning model, and obtaining simulation heat exchange data and prediction coupling heat exchange data; and adaptive grid optimization is carried out to obtain an optimized grid size, and the optimized grid size is adopted to adjust the mid-deep layer buried pipe heat pump system model to output the mid-deep layer buried pipe multi-field coupling model. According to the method, accurate and rapid simulation of the multi-field coupling process of the medium-deep geothermal pipe group can be realized under complex geological conditions, and the method has important practical significance for promoting large-scale development and utilization of medium-deep geothermal resources.

Description

Technical Field

[0001] This invention relates to the field of dynamic simulation technology for energy systems, and in particular to a multi-field coupling simulation method for medium-deep geothermal pipe groups based on deep learning. Background Technology

[0002] As a core component of renewable energy heating, the medium-deep buried geothermal system is at a critical stage of transition from single-pipe to pipe group mode. Many technical bottlenecks need to be overcome. The transition from single-pipe to pipe group-level modeling faces bottlenecks such as the heterogeneity of the strata, the computational time of large-scale array 3D modeling, and the progressively increasing uncertainty of parameters. Further scaling up the modeling scale is limited by the mutually exclusive relationship between computational efficiency and accuracy. There is an urgent need to develop new modeling methods that integrate multi-source data.

[0003] Existing multi-field coupled modeling of medium-deep geothermal pipe systems suffers from three main problems: First, the unclear multi-source data correlation mechanism leads to significant deviations between data-driven pattern mining and physical mechanism models. Second, there is a trade-off between model accuracy and computational efficiency; traditional numerical methods, limited by fixed mesh partitioning, struggle to achieve efficient modeling under complex geometries and dynamic boundary conditions, and cannot achieve adaptive topology optimization for pipe distribution. Third, the synergy between data-physics hybrid modeling is insufficient; purely data-driven models lack thermodynamic conservation law constraints, while traditional physical models have poor generalization capabilities and struggle to achieve dynamic feedback from real-time monitoring data. This invention proposes a deep learning-based multi-field coupled simulation method for medium-deep geothermal pipe systems. Combining measured data from large-scale medium-deep geothermal projects with high-precision numerical simulations, it deeply analyzes the interactions of multiple physical fields, including thermodynamics, fluid mechanics, and geomechanics, and optimizes the multi-field coupled simulation method for medium-deep geothermal pipe systems. This provides a novel approach and methodological support for high-precision simulation of medium-deep geothermal systems and numerical simulation of complex energy systems. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-field coupling simulation method for medium-deep geothermal pipe groups based on deep learning.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution:

[0006] This invention includes the following steps:

[0007] Measured heat exchange data of medium-deep geothermal pipe groups were obtained, and a buried pipe heat exchanger model was constructed. Numerical simulation and cross-validation were carried out to obtain a medium-deep buried pipe heat pump system model.

[0008] The Monte Carlo method is used to configure the system and obtain the first parameter environment set. The first parameter environment set is then filtered according to the parameter environment function to obtain the second parameter environment set.

[0009] Data augmentation is performed based on the second parameter environment set and the medium-deep geothermal pipe heat pump system model to obtain simulation data of medium-deep geothermal pipe groups. Based on the simulation data of medium-deep geothermal pipe groups, a multi-field coupled deep learning model of medium-deep geothermal pipe groups is constructed.

[0010] The basic parameters of the medium-deep geothermal pipe group project to be simulated are input into the medium-deep buried pipe heat pump system model and the medium-deep geothermal pipe group multi-field coupled deep learning model respectively to obtain simulated heat transfer data and predicted coupled heat transfer data.

[0011] Based on the simulated heat transfer data and the predicted coupled heat transfer data, adaptive mesh optimization is performed to obtain the optimized mesh size. The optimized mesh size is then used to adjust the model of the medium-deep buried pipe heat pump system, outputting the medium-deep buried pipe multi-field coupled model.

[0012] Furthermore, the method for obtaining the model of the medium-deep buried pipe heat pump system includes:

[0013] Acquire measured heat exchange data for a medium-deep geothermal pipe network project. This measured heat exchange data includes user-side data, heat source-side data, geothermal environment data, and buried pipe network design data. User-side data includes user-side circulation flow rate, supply and return water temperatures, user-side water system pressure distribution, and corresponding equipment inlet and outlet water temperatures. Heat source-side data includes heat source-side circulation flow rate, inlet and outlet water temperatures, heat source-side water system pressure distribution, and corresponding equipment inlet and outlet water temperatures. The geothermal environment includes soil and rock temperature distribution, soil and rock thermal properties, and outdoor air temperature. The buried pipe network design includes buried pipe network configuration and power. The buried pipe network configuration includes buried pipe specifications, number of buried pipes, buried pipe depth, buried pipe spacing, and operating parameters. The buried pipe network power includes heat pump units and delivery water pumps.

[0014] Based on the design of the underground pipe group, a model of the underground pipe heat exchanger was constructed, and the parameters of the background module of the energy system dynamic simulation software TRNSYS were set. The background module parameters include heat pump unit module, water pump module, building load module, hourly electricity price input module, outdoor meteorological parameters module for typical meteorological year, and optimization calculation module.

[0015] The buried pipe heat exchanger model is embedded into the energy system dynamic simulation software TRNSYS, and the background module is called to perform heat exchange simulation to obtain simulated heat exchange data. The element type, mesh type and TRNSYS boundary conditions of the buried pipe heat exchanger model are adjusted according to the deviation between the measured heat exchange data and the simulated heat exchange data until the deviation is minimized, and then the medium-deep buried pipe heat pump system model is output.

[0016] Furthermore, the method for obtaining the second parameter environment set includes:

[0017] Based on the measured heat exchange data of deep geothermal pipe group projects in history, the design range of the buried pipe group and the range of geothermal properties of the soil and rock are determined. The first parameter environment is obtained by random and uniform sampling within the design range of the buried pipe group and the range of geothermal properties of the soil and rock using the Monte Carlo method. The first parameter environment set is composed of the first parameter environment.

[0018] The statistical indicators for the configuration of the buried pipe group in the first parameter environment are determined based on the measured heat exchange data of historical deep geothermal pipe group projects; the statistical indicators include historical frequency, equal division distance, data confidence level, and mode;

[0019] Substitute statistical indicators into the parameter environment function to calculate the parameter environment score, and select the first parameter environment whose parameter environment score is less than the parameter environment score threshold as the second parameter environment to form the second parameter environment set.

[0020] The expression for the parameter environment function is:

[0021]

[0022] Where Env is the parameter environment score, N1 is the number of segments in the geothermal pipe group, and w i For segmented weights, w v For flow rate weighting, Configure parameter j to take the value x for section i of the geothermal pipe group ij Historical frequency, x j,l The parameters for configuring the type j geothermal pipe group are taken at the equally divided points l within the design range, d j,l Configure the geothermal pipe group with equal intervals of parameter j, P ij For x ij confidence level, x j,most The mode of the parameters configured for type j geothermal pipe groups. Let v be the flow rate of the exchange fluid in section i. i Historical frequency, v l Let d be the flow velocity of the exchange fluid at point l, which is an equal division point within the design range. v,l The distance that divides the exchange fluid flow rate equally. For v i confidence level, v most It is the mode of the exchange fluid flow rate.

[0023] Furthermore, the method for constructing a multi-field coupled deep learning model for medium-deep geothermal pipe groups includes:

[0024] Based on the second parameter environment, the buried pipe heat exchanger model of the medium-deep buried pipe heat pump system model is adjusted according to the design of the centralized buried pipe group. The background module of the medium-deep buried pipe heat pump system model is adjusted according to the corresponding geothermal properties to perform heat exchange simulation and obtain simulation data of the medium-deep geothermal pipe group.

[0025] The simulation data of the medium-deep geothermal pipe group and the second parameter environment are combined into a comprehensive training set. The comprehensive training set is randomly divided into a training set and a test set in a ratio of 6:4. The training set is used to train the deep learning model of the medium-deep geothermal pipe group with multiple fields coupled, and the test set is used to evaluate the performance of the deep learning model of the medium-deep geothermal pipe group with multiple fields coupled.

[0026] The multi-field coupling deep learning model for the medium-deep geothermal pipe group includes an input layer, a feature extraction layer, a physical constraint coupling layer, a parallel prediction layer, a policy fusion layer, and an output layer.

[0027] The feature extraction layer is used to extract the second parameter environmental features, including the ANN path, CNN path, RNN path and CNN-LSTM path; the ANN path uses a fully connected network to extract global features, the CNN path uses a convolutional network to extract spatial features, the RNN path uses a recurrent network to extract temporal features, and the CNN-LSTM path uses a convolutional-long short-term memory network to extract spatiotemporal features.

[0028] The physical constraint coupling layer updates the interaction between the temperature field, seepage field, and stress field through coupling equations, and outputs the coupled eigenvectors. The coupling equation expression is as follows:

[0029]

[0030] Where T' is the update of the temperature field characteristic T, P' is the update of the seepage field characteristic P, σ' is the update of the stress field characteristic σ, and α1, α2, β1, β2, γ, and δ are learnable coupling coefficients, determined by the range method of orthogonal experiments. For the partial derivative of the stress field characteristic σ with respect to time t, Let v be the convection term of the fluid velocity v and the temperature field characteristic T, representing the flow of the fluid carrying heat. Let ∈ be the partial derivative of the soil strain field characteristic with respect to time t, and k(T) be the temperature-dependent permeability. For the pressure gradient, E stone Here, ΔT represents the elastic modulus of the rock, and ΔT represents the temperature change.

[0031] The parallel prediction layer uses parallel prediction channels to predict the feature vectors coupled with physical constraints to obtain multi-field prediction results, including an ANN prediction layer, a CNN prediction layer, an RNN prediction layer, and a CNN-LSTM prediction layer. The ANN prediction layer uses a fully connected network to predict the distribution of temperature, pressure, and stress fields. The CNN prediction layer uses a convolutional network to predict the field distribution with spatial feature enhancement. The RNN prediction layer uses a recurrent network to predict the field distribution with temporal evolution. The CNN-LSTM prediction layer uses a convolutional-long short-term memory network to predict the field distribution with spatiotemporal coupling.

[0032] The strategy fusion layer performs weighted fusion of the multi-field prediction results from the parallel prediction layer to output a simulation coupled heat transfer feature vector. The weighted fusion weight expression is as follows:

[0033]

[0034] Where w i Let η be the weighted fusion weight of the i-th prediction layer, η be the positive activation coefficient, and θ be the negative penalty coefficient. MAPE is the ratio coefficient between the predicted value and the actual value of the i-th prediction layer. i Let RMSE be the mean absolute percentage error of the i-th prediction layer. i Let be the root mean square error of the i-th prediction layer.

[0035] Furthermore, the method for obtaining simulated heat transfer data and predicted coupled heat transfer data includes:

[0036] The basic parameters of the medium-deep geothermal pipe group project to be simulated are input into the medium-deep buried pipe heat pump system model to obtain simulated heat exchange data; the basic parameters include geothermal environment and buried pipe group design; the simulated heat exchange data includes overall simulated heat exchange data and local simulated heat exchange data;

[0037] The basic parameters of the deep geothermal pipe group project to be simulated are input into the deep geothermal pipe group multi-field coupling deep learning model to obtain the predicted coupled heat transfer data; the predicted coupled heat transfer data is related to the deep geothermal pipe group to be simulated as a whole.

[0038] Furthermore, the method for obtaining the optimized mesh size includes:

[0039] The overall mesh scaling factor is calculated based on the deviation between the overall simulated heat transfer data and the predicted coupled heat transfer data. The regional mesh scaling factor is calculated based on the deviation between the local simulated heat transfer data and the overall simulated heat transfer data. The expression is as follows:

[0040]

[0041] Where S g ω is the overall mesh scaling factor, β<0 is the scaling sensitivity factor, and ω is the scaling factor. T ω v ω P These represent the weights for temperature deviation, flow velocity deviation, and pressure deviation, respectively; N1 is the number of segments in the geothermal pipe group; and ΔT... i Let Δv be the deviation between the simulated exchange fluid temperature and the predicted coupled exchange fluid temperature of the i-segment geothermal pipe group. Let sign(·) be the sign function, taking +1 for positive deviation and -1 for negative deviation. i Let ΔP be the velocity deviation of the exchange fluid in section i of the geothermal pipe group, max(v) be the maximum velocity, and ΔP be the velocity deviation. iσ represents the water pressure deviation of section i of the geothermal pipe group. P The standard deviation of water pressure distribution. Let α be the scaling factor for the local grid at point i in section j of the geothermal pipe group, where α < 0 is the deviation scaling factor, and D ij Let be the local simulation deviation factor at point i in section j of the geothermal pipe group, max(D) be the maximum local simulation deviation factor, and γ be the stress coupling coefficient. Δσ represents the element compliance sensitivity output from the finite element analysis of a local point i in section j of the geothermal pipe group. ij This represents the deviation between the local simulated stress and the overall simulated stress at the corresponding location;

[0042] The optimized mesh size is obtained by adaptive mesh optimization based on the overall mesh scaling factor and the regional mesh scaling factor.

[0043] The beneficial effects of this invention are:

[0044] This invention is a deep learning-based multi-field coupling simulation method for medium-deep geothermal pipe groups. Compared with existing technologies, this invention has the following technical advantages:

[0045] This invention significantly improves the accuracy and efficiency of cross-scale simulations by using multi-physics coupled dynamic mesh generation technology, dynamically densifying / sparsening the pore-reservoir mesh, and combining it with a deep learning framework constrained by thermodynamic conservation laws. It successfully breaks through the dilemma of traditional fixed meshes between accuracy and computational cost, providing new ideas and methodological support for high-precision simulation of medium-deep geothermal systems and numerical simulation of complex energy systems. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating the steps of the deep learning-based multi-field coupling simulation method for medium-deep geothermal pipe groups in this invention. Detailed Implementation

[0047] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0048] The deep learning-based multi-field coupling simulation method for medium-deep geothermal pipe groups of this invention includes the following steps:

[0049] like Figure 1 As shown, this embodiment includes the following steps:

[0050] Measured heat exchange data of medium-deep geothermal pipe groups were obtained, and a buried pipe heat exchanger model was constructed. Numerical simulation and cross-validation were carried out to obtain a medium-deep buried pipe heat pump system model.

[0051] The Monte Carlo method is used to configure the system and obtain the first parameter environment set. The first parameter environment set is then filtered according to the parameter environment function to obtain the second parameter environment set.

[0052] Data augmentation is performed based on the second parameter environment set and the medium-deep geothermal pipe heat pump system model to obtain simulation data of medium-deep geothermal pipe groups. Based on the simulation data of medium-deep geothermal pipe groups, a multi-field coupled deep learning model of medium-deep geothermal pipe groups is constructed.

[0053] The basic parameters of the medium-deep geothermal pipe group project to be simulated are input into the medium-deep buried pipe heat pump system model and the medium-deep geothermal pipe group multi-field coupled deep learning model respectively to obtain simulated heat transfer data and predicted coupled heat transfer data.

[0054] Based on the simulated heat transfer data and the predicted coupled heat transfer data, adaptive mesh optimization is performed to obtain the optimized mesh size. The optimized mesh size is then used to adjust the model of the medium-deep buried pipe heat pump system, outputting the medium-deep buried pipe multi-field coupled model.

[0055] In this embodiment, the method for obtaining a model of a medium-deep buried pipe heat pump system includes:

[0056] Acquire measured heat exchange data for a medium-deep geothermal pipe network project. This measured heat exchange data includes user-side data, heat source-side data, geothermal environment data, and buried pipe network design data. User-side data includes user-side circulation flow rate, supply and return water temperatures, user-side water system pressure distribution, and corresponding equipment inlet and outlet water temperatures. Heat source-side data includes heat source-side circulation flow rate, inlet and outlet water temperatures, heat source-side water system pressure distribution, and corresponding equipment inlet and outlet water temperatures. The geothermal environment includes soil and rock temperature distribution, soil and rock thermal properties, and outdoor air temperature. The buried pipe network design includes buried pipe network configuration and power. The buried pipe network configuration includes buried pipe specifications, number of buried pipes, buried pipe depth, buried pipe spacing, and operating parameters. The buried pipe network power includes heat pump units and delivery water pumps.

[0057] Based on the design of the underground pipe group, a model of the underground pipe heat exchanger was constructed, and the parameters of the background module of the energy system dynamic simulation software TRNSYS were set. The background module parameters include heat pump unit module, water pump module, building load module, hourly electricity price input module, outdoor meteorological parameters module for typical meteorological year, and optimization calculation module.

[0058] The buried pipe heat exchanger model is embedded into the energy system dynamic simulation software TRNSYS, and the background module is called to perform heat exchange simulation to obtain simulated heat exchange data. The element type, mesh type and TRNSYS boundary conditions of the buried pipe heat exchanger model are adjusted according to the deviation between the measured heat exchange data and the simulated heat exchange data until the deviation is minimized, and then the medium-deep buried pipe heat pump system model is output.

[0059] In practical evaluation, taking the cross-verification of numerical simulation of a buried pipe heat exchanger model in a certain project as an example, the measured heat exchange data of a medium-deep geothermal pipe group project were obtained, namely, geothermal environment (user indoor air temperature, outdoor air temperature, and soil temperature distribution), user-side data (user-side heating network supply water temperature, heating network return water temperature, user-side circulation flow rate, and user-side water system pressure distribution), and heat source-side data (heat source-side circulation flow rate, inlet and outlet water temperatures, heat source-side water system pressure, and heat exchanger inlet temperature). The calculation includes: heat exchanger outlet temperature, heat exchanger side circulation flow rate, heat pump unit evaporator inlet water temperature, heat pump unit evaporator outlet water temperature, heat pump unit condenser inlet water temperature, heat pump unit condenser outlet water temperature; design of underground pipe network (underground pipe specifications, number of underground pipes, spacing between underground pipes, depth of underground pipes, power consumption of water pumps on the heat source side, power consumption of water pumps on the user side, power consumption of heat pump units), and calculation of heat output of deep well heat exchanger, heat exchange on the user side, COP of heat pump unit, and COPs of heating system;

[0060] Based on the specifications, quantity, spacing, and operating parameters (temperature and flow rate) of the buried pipes for the corresponding project, a suitable element type and mesh type are selected to construct a buried pipe heat exchanger model. Hourly load data at the terminal, heat pump system capacity configuration, equipment selection parameters, energy price information, and geotechnical thermal properties (thermal conductivity, specific heat capacity, and thermal diffusivity) are input to call the background module for heat exchange simulation to obtain simulated heat exchange data. The element type, mesh type, and TRNSYS boundary conditions of the buried pipe heat exchanger model are adjusted based on the deviation between the measured and simulated heat exchange data. The element type and mesh type with the smallest deviation are selected to adjust the buried pipe heat exchanger model, and the TRNSYS boundary conditions with the smallest deviation are used as the TRNSYS operating conditions to obtain a medium-deep buried pipe heat pump system model.

[0061] In this embodiment, the method for obtaining the second parameter environment set includes:

[0062] Based on the measured heat exchange data of deep geothermal pipe group projects in history, the design range of the buried pipe group and the range of geothermal properties of the soil and rock are determined. The first parameter environment is obtained by random and uniform sampling within the design range of the buried pipe group and the range of geothermal properties of the soil and rock using the Monte Carlo method. The first parameter environment set is composed of the first parameter environment.

[0063] The statistical indicators for the configuration of the buried pipe group in the first parameter environment are determined based on the measured heat exchange data of historical deep geothermal pipe group projects; the statistical indicators include historical frequency, equal division distance, data confidence level, and mode;

[0064] Substitute statistical indicators into the parameter environment function to calculate the parameter environment score, and select the first parameter environment whose parameter environment score is less than the parameter environment score threshold as the second parameter environment to form the second parameter environment set.

[0065] The expression for the parameter environment function is:

[0066]

[0067] Where Env is the parameter environment score, N1 is the number of segments in the geothermal pipe group, and w i For segmented weights, w v For flow rate weighting, Configure parameter j to take the value x for section i of the geothermal pipe group ij Historical frequency, x j,l The parameters for configuring the type j geothermal pipe group are taken at the equally divided points l within the design range, d j,l Configure the geothermal pipe group with equal intervals of parameter j, P ij For x ij confidence level, x j,most The mode of the parameters configured for type j geothermal pipe groups. Let v be the flow rate of the exchange fluid in section i. i Historical frequency, v l Let d be the flow velocity of the exchange fluid at point l, which is an equal division point within the design range. v,l The distance that divides the exchange fluid flow rate equally. For v i confidence level, v most The mode of the exchange fluid flow rate;

[0068] In practical assessments, the Monte Carlo method was used within the design range of the buried pipe network [buried pipe specifications 25-40mm, quantity 10-50 pipes, depth 1000-3000m, spacing 5-20m, exchange fluid flow velocity 0.5-2.0m / s] and within the range of geothermal properties of the soil and rock [thermal conductivity 1.5-3.5W / (m·K), specific heat capacity 0.8-1.2kJ / (kg·K), thermal diffusivity 0.5e]. -6 -1.5e -6 m 2 [Randomly selected first parameter environment (this buried pipe group is divided into 5 sections, and this set of data is one of the sections) [32mm, 20 pipes, 2000m, 10m, 1.2m / s, 2.5W / (m·K), 1.0kJ / (kg·K), 1.0e] -6 m 2 [ / s], obtain the corresponding statistical indicators (historical frequency, equal division distance, confidence level, mode) for buried pipe specifications / number of buried pipes / depth of buried pipes / spacing of buried pipes / flow velocity of exchange fluid: (50, 1.5mm, 0.9, 30mm) / (40, 4 pipes, 0.85, 25 pipes) / (60, 200m, 0.95, 1800m) / (55, 1.5m, 0.88, 12m) / (70, 0.15m / s, 0.92, 1.0m / s);

[0069] Take the flow rate weight w v The first parameter environment score is 28.14, which is less than the parameter environment score threshold of 50. This set of data is used as the second parameter environment. Multiple sets of first parameter environments are selected to form the second parameter environment set.

[0070] In this embodiment, the method for constructing a multi-field coupled deep learning model for medium-deep geothermal pipe groups includes:

[0071] Based on the second parameter environment, the buried pipe heat exchanger model of the medium-deep buried pipe heat pump system model is adjusted according to the design of the centralized buried pipe group. The background module of the medium-deep buried pipe heat pump system model is adjusted according to the corresponding geothermal properties to perform heat exchange simulation and obtain simulation data of the medium-deep geothermal pipe group.

[0072] The simulation data of the medium-deep geothermal pipe group and the second parameter environment are combined into a comprehensive training set. The comprehensive training set is randomly divided into a training set and a test set in a ratio of 6:4. The training set is used to train the deep learning model of the medium-deep geothermal pipe group with multiple fields coupled, and the test set is used to evaluate the performance of the deep learning model of the medium-deep geothermal pipe group with multiple fields coupled.

[0073] The multi-field coupling deep learning model for the medium-deep geothermal pipe group includes an input layer, a feature extraction layer, a physical constraint coupling layer, a parallel prediction layer, a policy fusion layer, and an output layer.

[0074] The feature extraction layer is used to extract the second parameter environmental features, including the ANN path, CNN path, RNN path and CNN-LSTM path; the ANN path uses a fully connected network to extract global features, the CNN path uses a convolutional network to extract spatial features, the RNN path uses a recurrent network to extract temporal features, and the CNN-LSTM path uses a convolutional-long short-term memory network to extract spatiotemporal features.

[0075] The physical constraint coupling layer updates the interaction between the temperature field, seepage field, and stress field through coupling equations, and outputs the coupled eigenvectors. The coupling equation expression is as follows:

[0076]

[0077] Where T' is the update of the temperature field characteristic T, P' is the update of the seepage field characteristic P, σ' is the update of the stress field characteristic σ, and α1, α2, β1, β2, γ, and δ are learnable coupling coefficients, determined by the range method of orthogonal experiments. For the partial derivative of the stress field characteristic σ with respect to time t, Let v be the convection term of the fluid velocity v and the temperature field characteristic T, representing the flow of the fluid carrying heat. Let ∈ be the partial derivative of the soil strain field characteristic with respect to time t, and k(T) be the temperature-dependent permeability. For the pressure gradient, E stone Here, ΔT represents the elastic modulus of the rock, and ΔT represents the temperature change.

[0078] The parallel prediction layer uses parallel prediction channels to predict the feature vectors coupled with physical constraints to obtain multi-field prediction results, including an ANN prediction layer, a CNN prediction layer, an RNN prediction layer, and a CNN-LSTM prediction layer. The ANN prediction layer uses a fully connected network to predict the distribution of temperature, pressure, and stress fields. The CNN prediction layer uses a convolutional network to predict the field distribution with spatial feature enhancement. The RNN prediction layer uses a recurrent network to predict the field distribution with temporal evolution. The CNN-LSTM prediction layer uses a convolutional-long short-term memory network to predict the field distribution with spatiotemporal coupling.

[0079] The strategy fusion layer performs weighted fusion of the multi-field prediction results from the parallel prediction layer to output a simulation coupled heat transfer feature vector. The weighted fusion weight expression is as follows:

[0080]

[0081] Where w i Let η be the weighted fusion weight of the i-th prediction layer, η be the positive activation coefficient, and θ be the negative penalty coefficient. MAPE is the ratio coefficient between the predicted value and the actual value of the i-th prediction layer. i Let RMSE be the mean absolute percentage error of the i-th prediction layer. i Let be the root mean square error of the i-th prediction layer;

[0082] In actual evaluation, the data input dimensions in the input layer are 8, specifically including pipe specifications / quantity / depth / spacing / flow rate, soil thermal conductivity / specific heat capacity / thermal diffusivity;

[0083] In the feature extraction layer, the input / output dimension of the ANN path is 8 / 128, the fully connected layer is [8→64→128], and the ReLU activation function is used; the input / output dimension of the CNN path is 8(4×2) / 128, the convolution kernel is 3×1×16→3×1×32, and the stride is 1; the input / output dimension of the RNN path is (8,10) / 128, the LSTM unit is 128-dimensional, and the time step is 10; the input / output dimension of the CNN-LSTM path is (8,10)→(4×2,10) / 128, which is expanded from Conv1D(16) to LSTM(128);

[0084] In the physical constraint coupling layer, the data input / output dimension is 4×128 / 384, and the learnable coupling coefficients α1, α2, β1, β2, γ, and δ are 0.18, 0.12, 0.25, 1.05, 0.32, and 0.08, respectively.

[0085] In the strategy fusion layer, the weighted fusion weights of the prediction layer are 0.31, 0.24, 0.35, and 0.40, respectively, the positive incentive coefficient η is 0.7, and the negative penalty coefficient θ is 0.3.

[0086] In this embodiment, the method for obtaining simulated heat transfer data and predicted coupled heat transfer data includes:

[0087] The basic parameters of the medium-deep geothermal pipe group project to be simulated are input into the medium-deep buried pipe heat pump system model to obtain simulated heat exchange data; the basic parameters include geothermal environment and buried pipe group design; the simulated heat exchange data includes overall simulated heat exchange data and local simulated heat exchange data;

[0088] The basic parameters of the deep geothermal pipe group project to be simulated are input into the deep geothermal pipe group multi-field coupling deep learning model to obtain the predicted coupled heat transfer data; the predicted coupled heat transfer data is related to the deep geothermal pipe group to be simulated as a whole.

[0089] In this embodiment, the method for obtaining the optimized mesh size includes:

[0090] The overall mesh scaling factor is calculated based on the deviation between the overall simulated heat transfer data and the predicted coupled heat transfer data. The regional mesh scaling factor is calculated based on the deviation between the local simulated heat transfer data and the overall simulated heat transfer data. The expression is as follows:

[0091]

[0092] Where S g ω is the overall mesh scaling factor, β<0 is the scaling sensitivity factor, and ω is the scaling factor. T ω v ω P These represent the weights for temperature deviation, flow velocity deviation, and pressure deviation, respectively; N1 is the number of segments in the geothermal pipe group; and ΔT... i Let Δv be the deviation between the simulated exchange fluid temperature and the predicted coupled exchange fluid temperature of the i-segment geothermal pipe group. Let sign(·) be the sign function, taking +1 for positive deviation and -1 for negative deviation. i Let ΔP be the velocity deviation of the exchange fluid in section i of the geothermal pipe group, max(v) be the maximum velocity, and ΔP be the velocity deviation. i σ represents the water pressure deviation of section i of the geothermal pipe group. P The standard deviation of water pressure distribution. Let α be the scaling factor for the local grid at point i in section j of the geothermal pipe group, where α < 0 is the deviation scaling factor, and Dij Let be the local simulation deviation factor at point i in section j of the geothermal pipe group, max(D) be the maximum local simulation deviation factor, and γ be the stress coupling coefficient. Δσ represents the element compliance sensitivity output from the finite element analysis of a local point i in section j of the geothermal pipe group. ij This represents the deviation between the local simulated stress and the overall simulated stress at the corresponding location;

[0093] The optimized mesh size is obtained by adaptive mesh optimization based on the overall mesh scaling factor and the region mesh scaling factor;

[0094] In the actual evaluation, taking the calculation of the regional grid scaling factor and the corresponding overall grid scaling factor at a local point a in section A of a medium-deep geothermal pipe group project as an example, the temperature deviation weight / flow velocity deviation weight / pressure deviation weight are taken as 0.5 / 0.3 / 0.2, the scaling sensitivity coefficient is -0.6, the deviation scaling factor is -0.4, and the deviation scaling factor is -0.15. The regional grid scaling factor at local point a in section A of the geothermal pipe group is calculated to be 0.705, and the overall grid scaling factor is 1.2. The product of the regional grid scaling factor and the overall grid scaling factor is taken as the comprehensive grid scaling factor (0.846) at local point a in section A of the geothermal pipe group. The HR-type adaptive meshing method is used to insert new nodes and merge adjacent elements. The new grid size is set according to the product of the original grid size and the comprehensive grid scaling factor (0.846) to obtain the optimized grid size. The optimized grid size is used to adjust the medium-deep buried pipe heat pump system model and output the medium-deep buried pipe multi-field coupling model.

[0095] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep learning-based multi-field coupling simulation method for medium-deep geothermal pipe groups, characterized in that, Includes the following steps: S1. Obtain measured heat exchange data of medium-deep geothermal pipe groups, construct a buried pipe heat exchanger model, and conduct numerical simulation cross-verification to obtain a medium-deep buried pipe heat pump system model. S2. Use the Monte Carlo method to configure the system and obtain the first parameter environment set. Then, filter the first parameter environment set according to the parameter environment function to obtain the second parameter environment set. S3. Obtain simulation data of medium-deep geothermal pipe group by performing data augmentation based on the second parameter environment set and the medium-deep buried pipe heat pump system model, and construct a multi-field coupled deep learning model of medium-deep geothermal pipe group based on the simulation data of medium-deep geothermal pipe group. S4. Input the basic parameters of the medium-deep geothermal pipe group project to be simulated into the medium-deep buried pipe heat pump system model and the medium-deep geothermal pipe group multi-field coupled deep learning model respectively to obtain simulated heat transfer data and predicted coupled heat transfer data. S5. Based on the simulated heat transfer data and the predicted coupled heat transfer data, perform adaptive mesh optimization to obtain the optimized mesh size, and use the optimized mesh size to adjust the medium-deep buried pipe heat pump system model to output the medium-deep buried pipe multi-field coupled model.

2. The deep learning-based multi-field coupling simulation method for medium-deep geothermal pipe groups according to claim 1, characterized in that, The method for obtaining a model of a medium-deep buried pipe heat pump system includes: Acquire measured heat exchange data for a medium-deep geothermal pipe network project. This measured heat exchange data includes user-side data, heat source-side data, geothermal environment data, and buried pipe network design data. User-side data includes user-side circulation flow rate, supply and return water temperatures, user-side water system pressure distribution, and corresponding equipment inlet and outlet water temperatures. Heat source-side data includes heat source-side circulation flow rate, inlet and outlet water temperatures, heat source-side water system pressure distribution, and corresponding equipment inlet and outlet water temperatures. The geothermal environment includes soil and rock temperature distribution, soil and rock thermal properties, and outdoor air temperature. The buried pipe network design includes buried pipe network configuration and power. The buried pipe network configuration includes buried pipe specifications, number of buried pipes, buried pipe depth, buried pipe spacing, and operating parameters. The buried pipe network power includes heat pump units and delivery water pumps. Based on the design of the underground pipe group, a model of the underground pipe heat exchanger was constructed, and the parameters of the background module of the energy system dynamic simulation software TRNSYS were set. The background module parameters include heat pump unit module, water pump module, building load module, hourly electricity price input module, outdoor meteorological parameters module for typical meteorological year, and optimization calculation module. The buried pipe heat exchanger model is embedded into the energy system dynamic simulation software TRNSYS, and the background module is called to perform heat exchange simulation to obtain simulated heat exchange data. The element type, mesh type and TRNSYS boundary conditions of the buried pipe heat exchanger model are adjusted according to the deviation between the measured heat exchange data and the simulated heat exchange data until the deviation is minimized, and then the medium-deep buried pipe heat pump system model is output.

3. The deep learning-based multi-field coupling simulation method for medium-deep geothermal pipe groups according to claim 1, characterized in that, The method for obtaining the second parameter environment set includes: Based on the measured heat exchange data of deep geothermal pipe group projects in history, the design range of the buried pipe group and the range of geothermal properties of the soil and rock are determined. The first parameter environment is obtained by random and uniform sampling within the design range of the buried pipe group and the range of geothermal properties of the soil and rock using the Monte Carlo method. The first parameter environment set is composed of the first parameter environment. The statistical indicators for the configuration of the buried pipe group in the first parameter environment are determined based on the measured heat exchange data of historical deep geothermal pipe group projects; the statistical indicators include historical frequency, equal division distance, data confidence level, and mode; Substitute statistical indicators into the parameter environment function to calculate the parameter environment score, and select the first parameter environment whose parameter environment score is less than the parameter environment score threshold as the second parameter environment to form the second parameter environment set. The parameter environment function expression is: Where Env is the parameter environment score, N1 is the number of segments in the geothermal pipe group, and w i For segmented weights, w v For flow rate weighting, Configure parameter j to take the value x for section i of the geothermal pipe group ij Historical frequency, x j,l The parameters for configuring the type j geothermal pipe group are taken at the equally divided points l within the design range, d j,l Configure the geothermal pipe group with equal intervals of parameter j, P ij For x ij confidence level, x j,most The mode of the parameters configured for type j geothermal pipe groups. Let v be the flow rate of the exchange fluid in section i. i Historical frequency, v l Let d be the flow velocity of the exchange fluid at point l, which is an equal division point within the design range. v,l The distance that divides the exchange fluid flow rate equally. For v i confidence level, v most It is the mode of the exchange fluid flow rate.

4. The deep learning-based multi-field coupling simulation method for medium-deep geothermal pipe groups according to claim 1, characterized in that, The method for constructing a multi-field coupled deep learning model for medium-deep geothermal pipe groups includes: Based on the second parameter environment, the buried pipe heat exchanger model of the medium-deep buried pipe heat pump system model is adjusted according to the design of the centralized buried pipe group. The background module of the medium-deep buried pipe heat pump system model is adjusted according to the corresponding geothermal properties to perform heat exchange simulation and obtain simulation data of the medium-deep geothermal pipe group. The simulation data of the medium-deep geothermal pipe group and the second parameter environment are combined into a comprehensive training set. The comprehensive training set is randomly divided into a training set and a test set in a ratio of 6:

4. The training set is used to train the deep learning model of the medium-deep geothermal pipe group with multiple fields coupled, and the test set is used to evaluate the performance of the deep learning model of the medium-deep geothermal pipe group with multiple fields coupled. The multi-field coupling deep learning model for the medium-deep geothermal pipe group includes an input layer, a feature extraction layer, a physical constraint coupling layer, a parallel prediction layer, a policy fusion layer, and an output layer. The feature extraction layer is used to extract the second parameter environmental features, including the ANN path, CNN path, RNN path and CNN-LSTM path; the ANN path uses a fully connected network to extract global features, the CNN path uses a convolutional network to extract spatial features, the RNN path uses a recurrent network to extract temporal features, and the CNN-LSTM path uses a convolutional-long short-term memory network to extract spatiotemporal features. The physical constraint coupling layer updates the interaction between the temperature field, seepage field, and stress field through coupling equations, and outputs the coupled eigenvectors. The coupling equation expression is as follows: Where T' is the update of the temperature field characteristic T, P' is the update of the seepage field characteristic P, σ' is the update of the stress field characteristic σ, and α1, α2, β1, β2, γ, and δ are learnable coupling coefficients, determined by the range method of orthogonal experiments. For the partial derivative of the stress field characteristic σ with respect to time t, Let v be the convection term of the fluid velocity v and the temperature field characteristic T, representing the flow of the fluid carrying heat. Let ∈ be the partial derivative of the soil strain field characteristic with respect to time t, and k(T) be the temperature-dependent permeability. For the pressure gradient, E stone Here, ΔT represents the elastic modulus of the rock, and ΔT represents the temperature change. The parallel prediction layer uses parallel prediction channels to predict the feature vectors coupled with physical constraints to obtain multi-field prediction results, including an ANN prediction layer, a CNN prediction layer, an RNN prediction layer, and a CNN-LSTM prediction layer. The ANN prediction layer uses a fully connected network to predict the distribution of temperature, pressure, and stress fields. The CNN prediction layer uses a convolutional network to predict the field distribution with spatial feature enhancement. The RNN prediction layer uses a recurrent network to predict the field distribution with temporal evolution. The CNN-LSTM prediction layer uses a convolutional-long short-term memory network to predict the field distribution with spatiotemporal coupling. The strategy fusion layer performs weighted fusion of the multi-field prediction results from the parallel prediction layer to output a simulation coupled heat transfer feature vector. The weighted fusion weight expression is as follows: Where w i Let η be the weighted fusion weight of the i-th prediction layer, η be the positive activation coefficient, and θ be the negative penalty coefficient. MAPE is the ratio coefficient between the predicted value and the actual value of the i-th prediction layer. i Let RMSE be the mean absolute percentage error of the i-th prediction layer. i Let be the root mean square error of the i-th prediction layer.

5. The deep learning-based multi-field coupling simulation method for medium-deep geothermal pipe groups according to claim 1, characterized in that, The method for obtaining simulated heat transfer data and predicted coupled heat transfer data includes: The basic parameters of the medium-deep geothermal pipe group project to be simulated are input into the medium-deep buried pipe heat pump system model to obtain simulated heat exchange data; the basic parameters include geothermal environment and buried pipe group design; the simulated heat exchange data includes overall simulated heat exchange data and local simulated heat exchange data; The basic parameters of the deep geothermal pipe group project to be simulated are input into the deep geothermal pipe group multi-field coupling deep learning model to obtain the predicted coupled heat transfer data; the predicted coupled heat transfer data is related to the deep geothermal pipe group to be simulated as a whole.

6. The deep learning-based multi-field coupling simulation method for medium-deep geothermal pipe groups according to claim 1, characterized in that, The method for obtaining the optimized mesh size includes: The overall mesh scaling factor is calculated based on the deviation between the overall simulated heat transfer data and the predicted coupled heat transfer data. The regional mesh scaling factor is calculated based on the deviation between the local simulated heat transfer data and the overall simulated heat transfer data. The expression is as follows: Where S g ω is the overall mesh scaling factor, β<0 is the scaling sensitivity factor, and ω is the scaling factor. T ω v ω P These represent the weights for temperature deviation, flow velocity deviation, and pressure deviation, respectively; N1 is the number of segments in the geothermal pipe group; and ΔT... i Let Δv be the deviation between the simulated exchange fluid temperature and the predicted coupled exchange fluid temperature of the i-segment geothermal pipe group. Let sign(·) be the sign function, taking +1 for positive deviation and -1 for negative deviation. i Let ΔP be the velocity deviation of the exchange fluid in section i of the geothermal pipe group, max(v) be the maximum velocity, and ΔP be the velocity deviation. i σ represents the water pressure deviation of section i of the geothermal pipe group. P The standard deviation of water pressure distribution. Let α be the scaling factor for the local grid at point i in section j of the geothermal pipe group, where α < 0 is the deviation scaling factor, and D ij Let be the local simulation deviation factor at point i in section j of the geothermal pipe group, max(D) be the maximum local simulation deviation factor, and γ be the stress coupling coefficient. Δσ represents the element compliance sensitivity output from the finite element analysis of a local point i in section j of the geothermal pipe group. ij This represents the deviation between the local simulated stress and the overall simulated stress at the corresponding location; The optimized mesh size is obtained by adaptive mesh optimization based on the overall mesh scaling factor and the regional mesh scaling factor.

Images