Reservoir-shaft integrated prediction method for geothermal exploitation

Abstract

The invention discloses a reservoir-shaft integrated prediction method for geothermal exploitation, which belongs to the technical field of geothermal exploitation, and comprises the following steps: S1, constructing a cross-scale reservoir model, S2, establishing a reservoir-shaft coupling model, S3, establishing a heat-fluid-solid coupling control equation, S4, solving the coupling control equation, S5, simulating multiphase flow in a shaft, and S6, learning and optimizing a machine. S7, dynamically updating prediction; by constructing a cross-scale reservoir model and a wellbore-reservoir coupling model, accurate mapping from micro pores to macroscopic thermal storage and dynamic coupling of a reservoir and a wellbore are realized, and numerical solution and multiphase flow accurate simulation of a thermal-fluid-solid multi-field coupling control equation are combined. Rock heterogeneity, multi-physical field interaction and complex flow characteristics in the shaft are comprehensively considered, the prediction accuracy and comprehensiveness are greatly improved, and the defects that in a traditional method, reservoir and shaft analysis is split, and the model precision is limited are overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geothermal exploitation, more particularly to a reservoir-wellbore integrated prediction method for geothermal exploitation. BACKGROUND

[0002] Geothermal energy is a natural heat energy from the interior of the earth, which is a clean and renewable energy, has the characteristics of large reserves, wide distribution, stability and reliability, and its core utilization method is to obtain energy by exploiting geothermal water or hot dry rock, and is widely used in heating, power generation, hot spring and other fields.

[0003] The patent document with the authorization announcement number CN120069164A discloses a geothermal prediction method based on machine learning, which includes: S11, obtaining the temperature measurement data and boundary conditions of the historical exploitation data of the geothermal well exploitation influence area, processing and adding Gaussian noise to the missing points and abnormal values of the temperature measurement data and boundary conditions based on KNN interpolation method; S12, using a convolutional neural network to extract deep features of the boundary conditions, using a bidirectional long short-term memory network to capture the time sequence features of the temperature measurement data, and determining the target influence factors of the deep features of the boundary conditions and the time sequence features of the temperature measurement data; S13, constructing a geothermal prediction model, inputting the deep features of the boundary conditions, the time sequence features of the temperature measurement data and the target influence factors into the geothermal prediction model for training, and obtaining the trained geothermal prediction model; S14, inputting the actual geothermal well exploitation data into the geothermal prediction model to obtain the prediction of the geothermal reserves.

[0004] The patent document with the authorization announcement number CN116733460A discloses a method for determining the content of carbon dioxide in a geothermal reservoir, which includes the steps of: obtaining the field test results of wellbore profile temperature-pressure in static state and blowout state; establishing a two-phase phase change flow model of CO2-H2O system in high-temperature geothermal wellbore including flash evaporation process; giving a CO2 content analysis range and dividing it into segments, taking a parameter point in each segment to calculate the temperature-pressure prediction results through the two-phase flow model; comparing the prediction results with the field test results, drawing the error and parameter variation curve, and the point with the minimum error is the optimal solution of CO2 content.

[0005] Although the prior art can determine the carbon dioxide content in the geothermal reservoir by establishing a wellbore two-phase flow model, the model has a wide range of applications and small errors, does not require wellhead or bottom sampling, uses wellbore full-section temperature and pressure data to improve the representativeness of the results, can also provide a basis for productivity prediction and scale prevention, and can also handle data missing and outliers based on machine learning, extract features and optimize the model to improve the accuracy of geothermal reserve prediction, the prior art has problems such as the separation of reservoir and wellbore analysis, the lack of dynamic coupling of the two for integrated prediction, the lack of cross-scale modeling from micro pores to macro thermal reserves to reflect the influence of rock heterogeneity, the lack of comprehensive and accurate simulation of thermal-flow-solid multi-field coupling and multiphase flow in the wellbore, and the low degree of fusion of physical models and data-driven models, insufficient dynamic adaptability, and difficulty in real-time tracking of changes in reservoir and wellbore characteristics during the mining process and adjusting the prediction model. SUMMARY

[0006] The present application mainly provides a reservoir-wellbore integrated prediction method for geothermal exploitation, which can solve the problems raised in the above background art.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solution: a reservoir-wellbore integrated prediction method for geothermal exploitation, comprising:

[0008] S1, a cross-scale reservoir model is constructed, including a micro-pore network model and a macro-thermal reservoir model, and the connection between the two is realized through a scale up algorithm;

[0009]

[0009] S2, a wellbore-reservoir coupling model is established, and the wellbore and the reservoir are discretely processed by using the finite element method, and the interface coupling condition is set;

[0010] S3, a thermal-flow-solid multi-field coupling control equation is established, including a heat conduction equation, a fluid motion equation and a solid mechanics balance equation;

[0011] S4, the thermal-flow-solid multi-field coupling control equation is numerically solved by using the staggered grid technique and the Newton iteration method;

[0012] S5, a two-fluid model or a drift flux model based on the mixed physical theory is selected, and the SIMPLE algorithm and the phase interface tracking technology are combined to realize accurate simulation of multiphase flow in the wellbore;

[0013] S6, the historical data of geothermal exploitation are preprocessed, a machine learning model is constructed and trained, and the trained model is used for reservoir and wellbore parameter inversion;

[0014] S7, the parameters obtained by machine learning are fed back to the physical model to realize dynamic updating of the physical model and complete the reservoir-wellbore integrated prediction.

[0015] Further, in the S1, the micro-pore network model obtains the real pore structure by scanning the rock slice through the image processing technology, solves the fluid flow and heat transfer equation in the pore by using the computational fluid dynamics method, the macro-thermal reservoir model describes the fluid seepage by using the Darcy law based on the continuous medium assumption, and the scale up algorithm equivalent the pore flow characteristics in the micro-model to the permeability tensor and thermal conductivity of the macro-model based on the volume average theory.

[0016] Further, in the S2, when the finite element method is discretized and processed, the reservoir area adopts the tetrahedral unit, the wellbore area adopts the structured cylindrical unit, and the interface coupling conditions include the heat conduction coupling of the wellbore wall and the reservoir, the fluid mass exchange coupling, and the momentum exchange coupling.

[0017] Further, in the S3, the heat conduction equation considers the heat exchange between the rock skeleton and the fluid, and the formula is as follows:

[0018]

[0019] In the formula, ρ is the density, Cp is the specific heat capacity, k is the thermal conductivity, and Q is the heat source term. The fluid motion equation adopts the Darcy law for the reservoir and the Navier-Stokes equation for the wellbore, and the solid mechanics equilibrium equation is based on the linear elastic constitutive relation, and the formula is as follows:

[0020]

[0021] In the formula, σ is the stress tensor, and g is the gravitational acceleration.

[0022] Further, in the S4, the staggered grid technology stores the pressure and velocity variables at the cell center and the face center, respectively, to avoid numerical oscillation, and the Newton iteration method solves the nonlinear equation set by constructing the Jacobian matrix.

[0023] Further, in the S5, the two-fluid model describes the gas-liquid two-phase flow by the mass conservation equation and the momentum conservation equation of each phase, considers the interfacial drag force and the heat transfer coefficient, the drift flux model introduces the drift velocity to correct the interfacial velocity difference, and the phase interface tracking technology adopts the VOF method or the LevelSet method, and the SIMPLE algorithm realizes the coupled solution of velocity and pressure by the pressure correction equation.

[0024] Further, in the S6, the data preprocessing includes outlier rejection, missing value filling, and feature normalization, the machine learning model adopts the deep neural network, the training data set includes the historical production, the bottom hole pressure, the wellbore temperature, and the logging interpretation parameters, and the loss function adopts the mean square error.​​​​​

[0025] Further, in the S7, the parameters fed back to the physical model include reservoir permeability, porosity, wellbore roughness and scale layer thickness, wherein the dynamic updating frequency is once every 24 hours, the model parameter correction is triggered based on the real-time monitored wellhead pressure, liquid production and temperature data, and after the correction, the numerical solving process of S3-S5 is re-executed.

[0026] The reservoir-wellbore integrated prediction method for geothermal exploitation has the following beneficial effects:

[0027] By constructing the cross-scale reservoir model and the wellbore-reservoir coupling model, the accurate mapping from the micro-pore to the macro-thermal reservoir and the dynamic coupling of the reservoir and the wellbore are realized, the numerical solving of the heat-flow-solid multi-field coupling control equation and the accurate simulation of the multiphase flow are combined, the rock heterogeneity, the multi-physical field interaction and the complex flow characteristics in the wellbore are comprehensively considered, the prediction accuracy and comprehensiveness are greatly improved, and the defects of the traditional method that the reservoir and the wellbore analysis are split and the model precision is limited are overcome.

[0028] By deeply integrating the machine learning model and the physical model, the reservoir and wellbore parameters are inversed by using the historical data and the physical model is dynamically updated, so that the prediction method can track the changes of the reservoir characteristics, the wellbore conditions and the mining technology in the mining process in real time, the dynamic adaptability is significantly enhanced, a more reliable basis is provided for the production capacity evaluation, the risk warning and the production decision of the geothermal exploitation, and the problem of insufficient dynamic adjustment capability of the prior art is solved. BRIEF DESCRIPTION OF DRAWINGS

[0029] The present application will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0030] Figure 1 The present application is a method flowchart. DETAILED DESCRIPTION

[0031] In order to make the technical scheme of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] Embodiment 1

[0033] As shown in the following technical scheme: Figure 1 A reservoir-wellbore integrated prediction method for geothermal exploitation is provided, comprising:

[0034] Step 1, constructing a cross-scale reservoir model

[0035] The cross-scale reservoir model is constructed, including a micro-pore network model and a macro-thermal reservoir model, and the connection of the two is realized through a scale upscaling algorithm.

[0036] Specifically, the micro-pore network model obtains the real pore structure by scanning the rock slice through image processing technology, solves the fluid flow and heat transfer equations in the pore by computational fluid dynamics method, the macro-thermal reservoir model describes the fluid seepage by Darcy's law based on the continuous medium assumption, and the scale up algorithm equivalent the pore flow characteristics in the micro-model to the permeability tensor and thermal conductivity of the macro-model based on the volume average theory.

[0037] Among them, the image processing technology needs to scan and analyze the rock slice with high precision, extract the pore size, distribution, throat connection and other structure parameters, construct a network model consistent with the real pore morphology, and solve the Navier-Stokes equation of fluid flow and Fourier's law of heat transfer in the pore by computational fluid dynamics method to capture the fluid velocity distribution, pressure change and heat exchange process. Specifically, the computational fluid dynamics method divides the micro-pore network model into small grids, discretizes the Navier-Stokes equation into algebraic equations at the grid nodes, solves the fluid velocity and pressure values at each node, and then obtains the fluid velocity distribution and pressure change in the pore. At the same time, based on Fourier's law, the heat transfer equation is established, combined with the heat exchange conditions between the fluid and the pore wall, the temperature field distribution in the pore is obtained by numerical calculation to capture the heat exchange process. And the influence of physical properties such as fluid viscosity, density and thermal conductivity needs to be considered in the solving process.

[0038] In addition, when applying Darcy's law, the macro-thermal reservoir model needs to combine with geological data such as reservoir rock type and porosity distribution, divide the reservoir into multiple continuous medium units, and the seepage characteristics of each unit are characterized by equivalent parameters, while considering the enhancement effect of fractures on fluid flow.

[0039] And the scale up algorithm calculates the volume average of the flow and heat transfer parameters of different pore units in the micro-model, establishes the mapping relationship between the micro-pore structure parameters (such as porosity and specific surface area) and the macro-permeability and thermal conductivity, realizes the parameter transmission from micro to macro, and ensures that the macro-model can reflect the influence of micro-heterogeneity.

[0040] Finally, the micro-model and the macro-model need to be coupled and verified, by comparing the local flow characteristics calculated by the micro-model with the equivalent results of the macro-model, adjusting the parameter conversion coefficients in the upscaling process, to ensure the consistency and accuracy of the cross-scale model.

[0041] Step 2, establish a reservoir-wellbore coupling model

[0042] Establish a wellbore-reservoir coupling model, discretize the wellbore and reservoir by finite element method, and set the interface coupling conditions.

[0043] Specifically, when the finite element method is discretized, the reservoir area adopts tetrahedral elements, and the wellbore area adopts structured cylindrical elements. The interface coupling conditions include heat conduction coupling between the wellbore wall and the reservoir, fluid mass exchange coupling, and momentum exchange coupling.

[0044] Among them, the tetrahedral element can flexibly adapt to the complex geometry and heterogeneous distribution of the reservoir. By encrypting the grid in the fracture development area, the calculation accuracy is improved. The structured cylindrical element is consistent with the axisymmetric characteristics of the wellbore, which facilitates efficient solution of the parameter variation along the wellbore axis.

[0045] At the same time, the heat conduction coupling is realized by defining the heat flow continuity condition on the contact surface between the wellbore wall and the reservoir, that is, the heat flux density on both sides of the wall surface is equal. The fluid mass exchange coupling is based on the connection between Darcy's law and the flow equation in the wellbore, ensuring the conservation of fluid flow at the interface. The momentum exchange coupling considers the pressure loss when the fluid flows through the interface and introduces a local resistance coefficient to modify the momentum equation.

[0046] Finally, during the discretization process, the grid transition treatment is needed for the interface area between the wellbore and the reservoir to avoid calculation errors caused by sudden changes in element size. Node sharing or constraint equations are used to ensure the continuity of physical quantities at the interface. At the same time, the convergence of the coupling model needs to be verified through trial calculation, and the grid density and iteration accuracy need to be adjusted to ensure that the model can accurately reflect the dynamic interaction between the reservoir and the wellbore, laying a foundation for subsequent multi-field coupling solution.

[0047] Step 3, establish heat-flow-solid coupling control equation

[0048] The heat-flow-solid multi-field coupling control equation is established, including the heat conduction equation, the fluid motion equation, and the solid mechanics balance equation.

[0049] Specifically, the heat conduction equation considers the heat exchange between the rock skeleton and the fluid, and its formula is as follows:

[0050]

[0051] In the formula, is the density, is the specific heat capacity, is the thermal conductivity, is the heat source term, while the fluid motion equation adopts Darcy's law for the reservoir and Navier-Stokes equation for the wellbore. The solid mechanics balance equation is based on the linear elastic constitutive relation, and its formula is:

[0052]

[0053] In the formula, is the stress tensor, is the gravitational acceleration;

[0054] The heat conduction equation is based on the Fourier law, which describes not only the heat transfer of the rock matrix in the reservoir, but also the heat conduction of the fluid in the wellbore, while fully considering the influence of the pore structure, surface roughness and fluid-rock interaction on heat transfer.

[0055] In the fluid motion equation, the Darcy law is applicable to the seepage of geothermal fluid in the porous medium of the reservoir, which can reflect the flow law of the fluid in the complex pore and fracture, and the Navier-Stokes equation is used to describe the flow state of the fluid in the wellbore, which can accurately capture the velocity distribution and pressure change of the fluid.

[0056] In addition, the control equations are not independent, but are closely coupled, for example, the pressure change caused by the fluid flow in the thermal reservoir will cause the deformation of the rock matrix, and the rock deformation will change the permeability and porosity of the reservoir, thereby affecting the flow path and heat transfer efficiency of the fluid. The solution results of the solid mechanics equilibrium equation need to be fed back to the results of the heat conduction equation and the fluid motion equation for mutual feedback and collaborative calculation.

[0057] Step 4, solving the coupled control equations

[0058] The staggered grid technique and Newton iteration method are used to numerically solve the heat-flow-solid multi-field coupled control equations.

[0059] Specifically, the staggered grid technique stores the pressure and velocity variables at the cell center and face center respectively to avoid numerical oscillation, and the Newton iteration method solves the nonlinear equation set by constructing the Jacobian matrix;

[0060] Among them, when the staggered grid technique is used to discretize the control equations, the grid accuracy will be adjusted according to the characteristics of different physical fields, for example, the grid will be densified in the area where the fluid velocity changes sharply to improve the ability to capture the flow details, while ensuring that the discrete format of the pressure field and the velocity field matches, reducing the numerical error;

[0061] And in the process of solving, the Newton iteration method updates the variable values (such as temperature, flow rate, displacement) of each physical field (such as heat conduction, fluid flow and solid mechanics deformation) at each iteration step, and gradually approaches the convergent solution through continuous iteration, in order to handle the nonlinear coupling relationship between the physical fields;

[0062] At the same time, in the process of iterative solution, a reasonable convergence criterion needs to be set, when the change of each physical field variable is less than the pre-set threshold, the equation is considered to be converged, and the iteration is stopped, in order to ensure the balance between the accuracy and efficiency of the calculation results;

[0063] In addition, after the solution is completed, the results need to be verified by comparing with known analytical solutions or experimental data to evaluate the reliability of the numerical method. If there is a deviation, adjust the discrete parameters or iteration strategy to ensure that the solution results can truly reflect the multi-field coupling phenomena in the geothermal exploitation process.

[0064] Step 5, Multiphase flow simulation in wellbore

[0065] Choose the two-fluid model or drift flux model based on the mixing theory, combine the SIMPLE algorithm and the phase interface tracking technology to realize the accurate simulation of multiphase flow in the wellbore.

[0066] Specifically, the two-fluid model describes gas-liquid two-phase flow through the mass conservation equation and momentum conservation equation, considers the interfacial drag force and heat transfer coefficient, the drift flux model introduces the drift velocity to correct the interfacial velocity difference, and the phase interface tracking technology uses the VOF method or LevelSet method, and the SIMPLE algorithm realizes the coupled solution of velocity and pressure through the pressure correction equation;

[0067] In the two-fluid model, the mass conservation equation and momentum conservation equation are constructed strictly in accordance with the actual flow characteristics in the geothermal exploitation scene, which can accurately depict the mass transfer and momentum exchange of gas-liquid two-phase in the complex wellbore environment, and adapt the parameters such as interfacial drag force and heat transfer coefficient to the flow state of multiphase flow;

[0068] In the phase interface tracking technology, the VOF method tracks the fluid volume fraction to clearly capture the shape changes of the gas-liquid interface in the wellbore due to geothermal fluid flow, temperature changes, etc., such as bubble generation, coalescence and rupture process, and the LevelSet method uses the level set function to stably describe the evolution of the phase interface, which has obvious advantages in handling topological structure mutations of the interface;

[0069] At the same time, when coupling the velocity and pressure, the SIMPLE algorithm fully considers the strong correlation between the pressure field and the velocity field in the wellbore during geothermal exploitation, from the initial assumption of the pressure field to the velocity field obtained by solving the momentum equation, and then iteratively adjusts the pressure correction equation to make them adapt to the continuity equation, ensuring the accuracy of the calculation of flow parameters during the simulation process, and fitting the complex flow characteristics of geothermal multiphase flow;

[0070] Finally, compare the simulation results with the geothermal exploitation field monitoring data (such as pressure, temperature, phase content at different depths in the wellbore, etc.), and laboratory simulation experimental data, optimize the interfacial action coefficient and numerical solution parameters in the model according to the deviation, so that the multiphase flow simulation accurately reflects the geothermal exploitation working conditions, and lays a solid foundation for subsequent analysis of key issues such as wellbore heat transfer and flow resistance.

[0071] Step 6, Machine learning and optimization

[0072] The historical data of geothermal exploitation is preprocessed, a machine learning model is constructed and trained, and the trained model is used for reservoir and wellbore parameter inversion.

[0073] Specifically, data preprocessing includes outlier removal, missing value filling and feature normalization. The machine learning model uses a deep neural network, the training data set includes historical production, bottom hole pressure, wellbore temperature and well logging interpretation parameters, and the loss function uses mean square error.

[0074] Among them, the outlier removal can be identified by statistical analysis (such as Z-score method) and removed, the missing value filling uses interpolation method or estimation method based on similar samples, the feature normalization (such as min-max normalization) converts the data to [0, 1] interval, eliminates the interference of dimension difference on model training, and ensures that the influence weight of different features on the model is reasonable.

[0075] And the construction of deep neural network needs to design the number of network layers and nodes according to the data size and prediction target. The input layer receives the preprocessed feature variables (such as historical production, wellbore temperature, etc.), the hidden layer extracts the deep correlation of data through nonlinear activation function, and the output layer corresponds to the predicted value of reservoir and wellbore parameters. In the training process, the back propagation algorithm is used to calculate the prediction error, and the error is propagated back to each layer to update the weight and bias, and the model parameters are constantly optimized.

[0076] At the same time, the selection of training data set needs to cover different exploitation stages and geological conditions to ensure that the model has good generalization ability. For example, it contains data of initial high yield, stable and decaying stages, and also includes exploitation records of different lithology and porosity reservoirs, so that the model can adapt to complex and variable actual working conditions.

[0077] Finally, when performing parameter inversion, real-time monitored wellhead pressure, liquid production and other data are input into the trained model to quickly output the inversion results of key parameters such as reservoir permeability and wellbore roughness, providing more accurate initial parameters and boundary conditions for physical model. By comparing the calculation results of physical model, the inversion accuracy is verified. If the deviation exceeds the threshold, the model is retrained or the data preprocessing strategy is adjusted to ensure the reliability of the inversion parameters.

[0078] Step 7, dynamic update prediction

[0079] The parameters obtained by machine learning are fed back to the physical model to realize the dynamic update of the physical model and complete the reservoir-wellbore integrated prediction.

[0080] Specifically, the parameters fed back to the physical model include reservoir permeability, porosity, wellbore roughness and scale layer thickness, wherein the dynamic updating frequency is once every 24 hours, the model parameter correction is triggered based on the real-time monitored wellhead pressure, liquid production and temperature data, and after the correction, the numerical solution process of S3-S5 is re-executed;

[0081] Among them, the update of reservoir permeability and porosity needs to combine the multi-field coupling effects such as rock deformation and fluid-rock interaction (for example, when geothermal exploitation leads to a decrease in reservoir pressure, the compression of rock skeleton will reduce the porosity, and then reduce the permeability), and the update of wellbore roughness and scale layer thickness needs to be associated with the results of multiphase flow simulation. The increase of scale layer thickness will reduce the wellbore inner diameter and increase the roughness, resulting in an increase in flow resistance, which needs to be calibrated by comparing the machine learning inversion results with the calculated wellbore pressure loss;

[0082] At the same time, the triggering mechanism of dynamic update not only depends on the fixed 24-hour cycle, but also immediately starts the parameter correction process when the real-time monitoring data appears abnormal fluctuation (for example, sudden drop of wellhead pressure, sudden reduction of liquid production), and quickly responds to the sudden conditions in the exploitation process, such as wellbore blockage or reservoir fracture closure;

[0083] Finally, after the dynamic update is completed, the new prediction results need to be continuously compared with the actual exploitation data to calculate the error index (for example, the root mean square error of pressure and temperature prediction value and measured value), if the error exceeds the preset threshold, the parameter inversion accuracy of machine learning model or the solution parameter of physical model is further optimized, forming a closed loop of "monitoring-inversion-updating-verification", ensuring that the integrated prediction method is always consistent with the actual exploitation process, and providing a reliable basis for production decision.

[0084] In summary, through the above 7 steps, the accurate characterization of geothermal reservoir geological characteristics under multi-source data fusion is realized, a comprehensive and actual heat-flow-solid multi-field coupling physical model is constructed, and numerical simulation means is used for efficient solution. At the same time, the historical data value is mined by machine learning to realize the dynamic update and prediction of model parameters, and the dynamic prediction and optimization of reservoir-wellbore integration are achieved.

[0085] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it cannot be understood as limiting the scope of the present patent. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which belong to the protection scope of the present application. Therefore, the protection scope of the present patent should be subject to the appended claims.

Claims

1. A reservoir-wellbore integrated prediction method for geothermal production, characterized by: S1. Construct a cross-scale reservoir model, including a microscopic pore network model and a macroscopic heat storage model, and connect the two through upscaling. S2. Establish a wellbore-reservoir coupling model, discretize the wellbore and reservoir using the finite element method, and set interface coupling conditions; S3. Establish the heat-fluid-solid multi-field coupling control equations, including the heat conduction equation, fluid motion equation, and solid mechanics equilibrium equation; S4. Use staggered grid technology and Newton iteration method to numerically solve the thermal-fluid-solid multi-field coupling control equations; S5. Select a two-fluid model or drift flux model based on mixture theory, combined with the SIMPLE algorithm and phase interface tracking technology to achieve accurate simulation of multiphase flow in the wellbore; S6. Preprocess the historical data of geothermal production, build and train a machine learning model, and use the trained model to invert reservoir and wellbore parameters; S7. Feed the parameters obtained by machine learning back to the physical model to achieve dynamic update of the physical model and complete the reservoir-wellbore integrated prediction.

2. The reservoir-wellbore integrated prediction method for geothermal production according to claim 1, characterized in that: In S1, the microscopic pore network model obtains the real pore structure by scanning rock slices through image processing technology, and adopts computational fluid dynamics method to solve the fluid flow and heat transfer equations in the pores. The macroscopic heat storage model is based on the continuous medium assumption and adopts Darcy's law to describe the fluid seepage. The upscaling method is based on the volume average theory, which equates the pore flow characteristics in the microscopic model to the permeability tensor and thermal conductivity coefficient of the macroscopic model.

3. The reservoir-wellbore integrated prediction method for geothermal production according to claim 1, characterized in that: In S2, when the finite element method is used for discretization, tetrahedral units are used in the reservoir region and structured cylindrical units are used in the wellbore region. The interface coupling conditions include: thermal conduction coupling, fluid mass exchange coupling, and momentum exchange coupling between the wellbore wall and the reservoir.

4. The reservoir-wellbore integrated prediction method for geothermal production according to claim 1, characterized in that: In S3, the heat conduction equation considers the heat exchange between the rock skeleton and the fluid, and its formula is as follows: ， Where, is the density, is the specific heat capacity, is the thermal conductivity, is the heat source term. At the same time, the fluid motion equation uses Darcy's law for the reservoir and the Navier-Stokes equation for the wellbore. The solid mechanics equilibrium equation is based on the linear elastic constitutive relationship, and its formula is: ， Where, is the stress tensor, is the acceleration due to gravity.

5. The reservoir-wellbore integrated prediction method for geothermal production according to claim 1, characterized in that: In the S4, the staggered grid technology stores the pressure and velocity variables at the cell center and the surface center respectively to avoid numerical oscillation, and the Newton iteration method solves the nonlinear equations by constructing the Jacobian matrix.

6. The reservoir-wellbore integrated prediction method for geothermal production according to claim 1, characterized in that: In S5, the two-fluid model describes the gas-liquid two-phase flow through the mass conservation equation and momentum conservation equation of each phase, taking into account the interphase drag and heat transfer coefficient. The drift flux model introduces the drift velocity to correct the interphase velocity difference, and the phase interface tracking technology adopts the VOF method or the LevelSet method. The SIMPLE algorithm realizes the coupled solution of velocity and pressure through the pressure correction equation.

7. The reservoir-wellbore integrated prediction method for geothermal production according to claim 1, characterized in that: In S6, data preprocessing includes outlier removal, missing value filling and feature normalization. The machine learning model adopts a deep neural network. The training data set includes historical production, bottom hole pressure, wellbore temperature and logging interpretation parameters. The loss function adopts mean square error.

8. The reservoir-wellbore integrated prediction method for geothermal production according to claim 1, characterized in that: In S7, the parameters fed back to the physical model include reservoir permeability, porosity, wellbore roughness and scale layer thickness, with a dynamic update frequency of once every 24 hours. Model parameter correction is triggered based on real-time monitored wellhead pressure, liquid production and temperature data, and the numerical solution process of S3-S5 is re-executed after correction.

Images