Prediction and Control Method for Wellbore Instability in Deep Altered Tuff Coupled with Thermal-Fluid Solidification

Abstract

This application belongs to the field of drilling engineering, specifically disclosing a method for predicting and controlling wellbore instability in deep altered tuff using a heat-fluid-solidification coupling approach. The method includes: acquiring formation mechanical parameters, mineral composition, thermodynamics, and hydrochemical parameters to construct a microscopic heat-fluid-solidification coupling model of deep altered tuff; simulating the wellbore environment after actual drilling to analyze the impact of drilling fluid circulation on the wellbore tuff; calculating collapse, leakage, and fracturing pressures to obtain a safe density window; training a high-fidelity proxy model using simulated data and on-site monitoring data during drilling; and during drilling, using this proxy model to recommend and control drilling fluid density in real time, while continuously feeding back on-site monitoring data to the proxy model for iterative updates, forming a closed loop of "monitoring → prediction → control → re-monitoring" to achieve dynamic real-time control of wellbore stability. This application balances microscopic accuracy with on-site real-time performance, significantly improving drilling safety and efficiency.

Description

Technical Field

[0001] This application belongs to the field of drilling engineering, and more specifically, relates to a method for predicting and controlling the instability of deep altered tuff wellbore through thermal flow-solidification coupling. Background Technology

[0002] Wellbore instability is a frequent and complex downhole accident in drilling operations, mainly manifested as wellbore narrowing, wellbore collapse, borehole enlargement, and wellbore fracturing. Deep and ultra-deep (≥4000m) altered tuff is particularly prone to wellbore instability during drilling due to its high content of volcanic glass alteration products, high clay mineral content (especially montmorillonite), well-developed micropores and fractures, and intense physical-chemical-mechanical coupling upon contact with water. This is especially pronounced in highly deviated and horizontal wells. The expansion mechanism of altered tuff is more complex than that of ordinary mudstone: on the one hand, minerals such as montmorillonite absorb water, causing lattice expansion; on the other hand, residual volcanic glass from the alteration process undergoes ion exchange with the drilling fluid, leading to particle dispersion and structural disintegration. Wellbore instability not only significantly increases drilling costs and prolongs the drilling cycle, but in severe cases, it can even lead to the abandonment of the entire well. Therefore, accurately predicting the wellbore stability of altered tuff formations and rationally designing drilling fluid density windows are key prerequisites for achieving safe and efficient drilling in deep and ultra-deep formations.

[0003] Extensive research has been conducted by scholars both domestically and internationally on wellbore stability. Early studies primarily relied on continuum mechanics, employing strength theories such as the Mohr-Coulomb criterion and the Drucker-Prag criterion, combined with geostress distribution and pore pressure, to establish analytical models of wellbore stress and determine the safe density window for drilling fluid. These methods treat the rock as a homogeneous continuum, neglecting the rock's microscopic heterogeneous structure (such as particle size distribution, pore size, microfractures, and bedding planes) and the localized hydration and expansion effects caused by drilling fluid intrusion. In recent years, some studies have focused on the hydration of altered tuff, establishing macroscopic force-chemistry multi-field coupled equations by introducing chemical potential difference and semi-permeable membrane models to analyze the impact of hydration on wellbore stress. However, these models still operate within a macroscopic continuum framework, simplifying the hydration effect to an additional stress term. They fail to accurately reflect the microscopic causal chains of drilling fluid ion diffusion, changes in the electric double layer on particle surfaces, and the generation and transmission of expansion forces. This results in an inaccurate characterization of the wellbore instability mechanism in altered tuff, leading to significant discrepancies between predicted and actual conditions.

[0004] Therefore, how to accurately characterize the microscopic instability mechanism and improve the accuracy of wellbore instability prediction is an urgent problem to be solved. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this application is to provide a method for predicting and controlling wellbore instability in deep altered tuff using a thermal-fluid-solidification coupling approach. This method can accurately characterize the microscopic instability mechanism and improve the accuracy of wellbore instability prediction.

[0006] To achieve the above objectives, in a first aspect, this application provides a method for predicting wellbore instability in deep altered tuff using a thermal-fluid-solidification coupling approach, comprising the following steps: S10. Based on the stratigraphic mechanical parameters, mineral composition, thermodynamic and hydrochemical parameters of the deep altered tuff strata, a microscopic thermal-fluid solidification coupling model is constructed using the computational fluid dynamics-discrete element method. S20, apply initial geostress, initial formation temperature field, pore pressure field and initial fluid chemical environment conditions to the microscopic thermal flow solidification coupling model, and delete rock particles within the preset wellbore diameter range in the model to simulate wellbore formation; S30, Traverse all particles and their contact information in the model, define the closed area formed by the particles as a pore, define the flow channel between the contacting particles, the opening of the flow channel dynamically changes with the normal contact force between the two particles, and correct the opening of the channel according to the degree of mineral hydration and solidification reaction, and establish a well wall stability analysis model under the combined action of thermal flow solidification. S40, set drilling fluid parameters, use the wellbore stability analysis model to calculate the flow, heat transfer and reaction process of drilling fluid in the flow channel, and calculate the water content in each pore in real time. Control the radial expansion of altered tuff particles around the pore according to the water content, and dynamically update the mechanical strength parameters between particles according to the degree of hydrochemical reaction and solidification reaction to obtain the wellbore instability and failure under different drilling fluid densities. Obtain the drilling fluid safe density window of the altered tuff formation according to the preset allowable well diameter enlargement rate.

[0007] As a further preferred embodiment, in step S30, the specific way in which the pipe opening dynamically changes with the normal contact force between the two particles is as follows: when the normal contact force between the two particles is compressive stress, the pipe opening satisfies the formula ,in This represents the initial opening of the pipe. F 0 represents the compressive force when the pipe opening is reduced to half of its initial opening. F The current compressive force; when the two particles are in a tensile state or the cementation between the particles fails, the pipe opening satisfies the formula. ,in l It is a dimensionless multiplier. d The distance between the centers of the two balls. R 1. R 2 represents the radii of the two particles; The specific method for correcting the pipe opening based on the degree of mineral hydration and solidification reaction is as follows: using the formula... Make corrections, among which k w This is the hydration correction factor. oh Moisture content, k c This is the curing correction factor. This indicates the degree of curing reaction.

[0008] As a further preferred embodiment, in step S40, the wellbore stability analysis model includes: Flow rate calculation formula in a flow pipe ,in k Permeability coefficient, For pipe opening, p The pressure difference between the two pores. L This refers to the length of the pipe. Formula for the change of fluid pressure in pores ,in For fluid bulk modulus, For pore volume, This represents the total flow rate of fluid entering the pore. This represents the change in pore volume. Water content in pores ; The wellbore stability analysis model also includes a temperature field evolution equation, where temperature changes in the pore and particle system satisfy the energy conservation equation: ,in r For the density of the medium, c For specific heat capacity, T For temperature, t For time, l Thermal conductivity, Q For heat source items; The specific way in which the temperature field affects the particle contact state and the flow channel opening is as follows: when the particle temperature changes, the particle radius increment satisfies... ,in The coefficient of thermal expansion of the particles. The initial radius of the particle, ΔT This represents the change in temperature; the thermal expansion of particles causes changes in the normal contact force between particles, which in turn drives the dynamic change in the opening of the flow channel.

[0009] As a further preferred embodiment, in step S40, the specific method for controlling the radial expansion of the altered tuff particles around the pores based on the water content is as follows: when the water content... At that time, a radial expansion force is applied to all particles surrounding the pore, away from the pore center. The magnitude of the expansion force is...F swell = k swell · oh ,in k swell The coefficient of expansion is determined through indoor hydration expansion experiments.

[0010] As a further preferred embodiment, in step S40, the degree of curing reaction is characterized by a reaction kinetic equation: the degree of curing reaction satisfies , x To indicate the degree of curing reaction, k(T) This is the temperature-dependent reaction rate constant. f(C) A reaction function related to the concentration of fluid chemical components; The specific way in which the degree of curing reaction affects the particle mechanical strength parameters is as follows: the cohesion or strength parameters of the particle system change dynamically with the degree of curing reaction, satisfying the formula... s=s 0 +vx ,in s These are the strength parameters after curing. s 0 represents the initial strength parameter. β The curing enhancement factor is denoted as .

[0011] As a further preferred embodiment, before step S10, a step of obtaining micromechanical parameters through nanoindentation testing is also included: performing nanoindentation testing on the altered tuff sample, extracting the load-displacement curve, calculating the contact stiffness, contact depth, contact area, and reduced elastic modulus based on the Oliver-Pharr method, and then solving for the microelastic modulus and microhardness of different micromineral phases, which serve as prior input parameters for particle contact modulus and bonding strength in the microthermal-fluid-solidification coupling model.

[0012] As a further preferred embodiment, after step S10, a model verification step is also included: using the calibrated microscopic parameters to conduct virtual uniaxial compression tests and virtual Brazilian splitting tests on the microscopic thermal-fluid-curing coupling model, comparing the simulated macroscopic mechanical parameters with the measured values ​​from indoor tests, and determining that the model passes verification when the relative errors of the four indicators—elastic modulus, Poisson's ratio, uniaxial compressive strength, and tensile strength—are all less than 10%.

[0013] As a further preferred embodiment, after step S40, an offline modeling stage is also included, which is used to construct a physical information graph neural network surrogate model to achieve rapid prediction; The offline modeling phase includes the following steps: Within the preset range of drilling fluid density, drilling fluid pressure, and geostress parameters, multiple sets of input parameter combinations are generated using Latin hypercube sampling. The microscopic heat flow solidification coupling model described in steps S10 to S40 is run in batches to obtain N1 sets of high-fidelity training data. At the same time, N2 sets of low-fidelity training data are generated using a reduced-order analytical model, where N2 ≥ 5 × N1. A physical information graph neural network is constructed as a surrogate model. Its inputs are the adjacency matrix, node features, and global features of the discrete element particle graph, and its outputs are the collapse pressure equivalent density, the fracture pressure equivalent density, and the well diameter enlargement rate. The loss function of the physical information graph neural network includes data fitting loss, physical constraint loss, and contact force-pipe opening consistency loss: where, data fitting loss is the weighted mean square error between the surrogate model output and the high-fidelity and low-fidelity data; physical constraint loss is that the stress field output by the surrogate model satisfies the discrete element force balance equation and the output pressure field satisfies the equivalent form of Darcy's law for pore flow; contact force-pipe opening consistency loss is that the relationship between the contact force output by the intermediate layer of the surrogate model and the pipe opening is consistent with the dynamic change of the pipe opening with the normal contact force between the two particles in step S30; A residual multi-fidelity training strategy is adopted: first, the physical information graph neural network is pre-trained using low-fidelity data, and then the parameters of the last two layers are fine-tuned using high-fidelity data.

[0014] As a further preferred embodiment, the drilling process also includes a closed-loop self-correction step: comparing the actual wellbore data measured while drilling with the wellbore change rate predicted by the physical information graph neural network surrogate model; if the relative deviation between the two exceeds a preset value, the microscopic heat flow solidification coupling model is triggered to perform supplementary calculations using the latest measured data, and the physical information graph neural network surrogate model is updated by transfer learning using the supplementary calculation results.

[0015] Secondly, this application provides a method for real-time control of wellbore instability, comprising the following steps: Using the thermal-fluid-solidification coupling method for predicting wellbore instability in deep altered tuff formations as described above, the drilling fluid safety density window for deep altered tuff formations is obtained. Obtain the actual drilling fluid density at the drilling site and compare the actual drilling fluid density with the drilling fluid safety density window; The drilling fluid density is adjusted in real time based on the comparison results: if the actual drilling fluid density is higher than the upper limit of the safe density window, the drilling fluid density is reduced; if the actual drilling fluid density is lower than the lower limit of the safe density window, the drilling fluid density is increased.

[0016] The beneficial effects of this application are as follows: (1) Accurately depict the micro-instability mechanism: The CFD-DEM discrete element method is used to construct a micro-thermal flow solidification coupling model, which breaks through the limitations of the traditional macro-continuous medium model. It can truly reflect the micro-heterogeneous structure of altered tuff and the dynamic processes such as hydration expansion and solidification reaction after drilling fluid intrusion, which significantly improves the prediction accuracy.

[0017] (2) Achieve multi-field coupling drive: For the first time, stress field, seepage field, temperature field and chemical / solidification reaction field are coupled at the microscale, and quantitative relationships such as water content-expansion force, degree of solidification-bonding strength, temperature change-pipe opening are established, completely replicating the evolution process of wellbore instability, and providing a reliable basis for determining the safe density window.

[0018] (3) Balancing high accuracy and real-time performance: By introducing the PI-GNN dual-fidelity proxy model, high-fidelity offline simulation is combined with lightweight online prediction. The time taken for a single prediction is in the millisecond range, which solves the contradiction that the micro-model has a large amount of computation and cannot meet the requirements of real-time decision-making on site.

[0019] (4) It has closed-loop self-correction capability: it compares the measured well diameter data while drilling with the predicted value of the proxy model in real time. When the deviation exceeds the threshold, it automatically triggers the high-fidelity model to supplement calculation and transfer learning update, so that the system can continuously adapt to formation changes and avoid model inaccuracy.

[0020] (5) Provide engineering control basis and reduce risk costs: output accurate drilling fluid safety density window and clear control rules, combined with millisecond-level prediction capability, guide the real-time adjustment of drilling fluid performance on site, effectively prevent wellbore instability accidents, reduce non-production time, and significantly reduce drilling costs and operational risks.

[0021] (6) Strong applicability and scalability: The technical framework of this application can be extended to wellbore stability prediction and control in complex and easily unstable formations such as shale, mudstone, and fractured formations, and has broad market application prospects. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the method flow provided in this application; Figure 2 This is a schematic diagram of drilling fluid permeation in the altered tuff wellbore provided in this application; Figure 3 This is a schematic diagram of the hydration expansion of the well-perimeter altered tuff after being soaked in drilling fluid, as provided in this application. Figure 4 This is a comparative diagram of well wall collapse modes with and without considering tuff hydration expansion; where (a) is the well wall collapse mode with tuff hydration expansion considered, and (b) is the well wall collapse mode without considering tuff hydration expansion. Figure 5This is a schematic diagram comparing wellbore instability with and without consideration of tuff hydration expansion when the drilling fluid pressure is greater than the formation fracture pressure; where (a) is the wellbore state with consideration of tuff hydration expansion, and (b) is the state with expansion of cracks and collapse range without consideration of tuff hydration expansion. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0024] like Figure 1 As shown, this application provides a method for predicting and controlling wellbore instability in deep altered tuff based on microscopic thermal-fluid-solidification coupling, comprising the following steps: S1. Data Acquisition: Calculate macroscopic geomechanical parameters of deep altered tuff formations using well logging data; obtain particle size distribution, mineral composition, and fracture network of altered tuff through drilling fluid cuttings; perform nanoindentation tests on altered tuff samples, extract load-displacement curves of different mineral components, and calculate the elastic modulus and microhardness of each microphase as prior mechanical parameters for the discrete element micro-thermal-fluid-solidification coupling model; obtain thermodynamic parameters, hydrochemical and solidification reaction parameters, and fluid phase parameters of deep altered tuff through laboratory experiments.

[0025] It should be noted that this application uses well logging data to calculate the geomechanical parameters of deep altered tuff formations, including overlying formation pressure, maximum horizontal principal stress, and minimum horizontal principal stress. Rock mechanical parameters are obtained through laboratory tests. The compressive strength, elastic modulus, and Poisson's ratio of the altered tuff are obtained through uniaxial compression tests, and the tensile strength is obtained through Brazilian fracturing tests. Alternatively, the compressive strength, tensile strength, elastic modulus, and Poisson's ratio can be directly calculated from well logging data.

[0026] Geomechanical parameters are obtained by inverting the macroscopic mechanical parameters and stress profile of the formation using conventional well logging interpretation models in this field (such as using sonic / density logging curves). As an example, the geomechanical and rock mechanics parameters calculated from well logging data include: overlying formation pressure, maximum horizontal principal stress, minimum horizontal principal stress, rock compressive strength, tensile strength, rock elastic modulus, and Poisson's ratio. Specific parameter values ​​are shown in Table 1.

[0027] Table 1 Mechanical parameters

[0028] As one example, the grain size distribution of altered tuff is obtained through drilling fluid cuttings return: D 10 =0.8μm, D 50=3.2μm, D 90 =18μm, porosity 12.5%. Indoor hydration expansion experiment to calibrate the expansion coefficient k. swell =1.15MPa (the linear model is applicable to a moisture content range of 0~25%).

[0029] As one embodiment, in order to obtain accurate micromechanical parameters as prior inputs for the discrete element microstructure interaction model, this application innovatively employs nanoindentation testing to calculate the mechanical properties of different microphases (such as montmorillonite infill, tuffaceous matrix, quartz remnants, etc.) in altered tuff. The specific calculation process is analytically derived based on the Oliver-Pharr method: First, based on the slope of the initial stage (at the maximum load) of the load-displacement unloading curve obtained from the test, the contact stiffness is calculated: (1) Because the material around the indenter elastically recovers under maximum load, the actual contact depth is less than the maximum indentation depth. Therefore, the actual contact depth is calculated as follows: (2) The actual contact area is calculated from the contact depth based on the pre-calibrated diamond indenter area function. For the Berkovich indenter used in this experiment, the calculation formula is as follows: (3) Furthermore, by introducing an indenter shape correction factor, the reduced elastic modulus affected by both the test sample and the indenter is calculated: (4) Subsequently, by removing the influence of the elastic deformation of the diamond indenter itself, the true microelastic modulus of different micro-mineral phases was derived: (5) Finally, the microhardness of different micromineral phases is calculated: (6) In equations (1) to (6), P For the applied load, h Indentation depth For the maximum test load, This represents the maximum indentation depth. A constant related to the geometry of the indenter is taken as 0.75; This is the indenter shape correction factor, which is 1.034 for the Berkovich indenter; To convert the elastic modulus, To test the Poisson's ratio of the micro-mineral phases in the sample, This is the desired true microscopic elastic modulus; and These are the elastic modulus and Poisson's ratio of the diamond indenter (typically...). =1141 GPa, =0.07); This refers to microscopic hardness.

[0030] The true microscopic elastic modulus of each microphase obtained through the above steps This will be directly used as the initial input value for the corresponding particle contact modulus in the discrete element model in step S2; the calculated microhardness This is equivalent to the normal and tangential bonding strength between discrete element particles, thus significantly improving the initial accuracy of the microscopic fluid-structure interaction model.

[0031] S2. Construct and validate the microscopic thermal-fluid-solidification coupling model: Based on particle size distribution, porosity, mineral composition, fracture network, and formation temperature and pressure, a microscopic thermal-fluid-solidification coupling model of the wellbore in deep altered tuff is constructed using computational fluid dynamics-discrete element method. The correctness of the microscopic thermal-fluid-solidification coupling model is validated using mechanical parameters, thermodynamic parameters, hydrochemical and solidification reaction parameters, and fluid phase parameters. As one embodiment, this application employs the discrete element method (DEM) to construct a microscopic thermal-fluid-solidification coupling model of altered tuff. Unlike traditional continuous medium models, the DEM treats the rock as an aggregate composed of a large number of rigid particles (particle units), with forces and torques transmitted between particles through contact models (such as linear contact models and parallel bonding models). By rationally setting the particle size distribution, spatial arrangement, and contact parameters, the discontinuous and heterogeneous mechanical behavior of altered tuff can be accurately simulated, especially exhibiting unique advantages in localized failure processes such as hydration expansion and fracture propagation.

[0032] The specific steps for building are as follows: Step S21: Generate particle aggregates S211. Determine particle size distribution: Based on the particle size distribution curve of the altered tuff obtained in step S1, the particle size range and mass fraction of the particles in the model are determined. To balance computational accuracy and efficiency, the particle size is usually scaled down: retaining the original particle size distribution D... 10 D 50 D 90 The three characteristic particle sizes will be smaller than D. 10 The fine particles merge into D 10 Particles of size D will be larger than D. 90 The coarse particles are proportionally reduced to D 90 The scaling principle is to keep the mass fraction of each particle size segment constant.

[0033] Example: Core results of a certain altered tuff show: D 10 =0.5μm, D 50 =2.5μm, D 90 =15μm. The model uses three particle sizes: 0.5μm (15% by mass), 2.5μm (60% by mass), and 15μm (25% by mass).

[0034] S212, Generate initial particle packing: Within the defined model boundaries (typically a cube with a side length of 1-5 mm or a cylinder with a radius of 1-3 mm, the size depending on computational resources), particle aggregates are generated using either the "particle size expansion method" or the "gravity deposition method": Particle size expansion method: Randomly generate location points within the boundary, assigning each point an initial radius much smaller than the target particle size. Then, gradually increase the particle radius to the target value, while detecting and adjusting the positions of overlapping particles until the contact overlap of all particles is less than a preset threshold (typically 1% of the minimum particle size). This method is suitable for generating high-density particle packing.

[0035] Gravity deposition method: Particles of the target size are generated within the boundary and allowed to settle naturally under gravity until the system reaches equilibrium. This method is closer to the natural deposition process, but it is computationally time-consuming.

[0036] S213. Adjusting porosity: After generating the initial particle aggregate, calculate the porosity of the current model. ,in The sum of the volumes of all particles. This represents the total volume of the model. Compared with the true porosity measured by CT scan in step S1 Compare. If Then, isobaric compression is applied to the model (by moving the boundary particles inward to gradually increase the particle density); if If necessary, expand the boundary or remove some severely overlapping particles. Repeat this process until... .

[0037] Step S22: Set the contact model and micro parameters S221. Contact Model Selection: Altered tuff, as a cemented sedimentary rock, exhibits a certain degree of cementation strength between its grains. This application preferably uses a parallel bond model to describe the mechanical behavior between the grains. This model incorporates a "cementing disk" with a certain cross-sectional area at the grain contact point, which can transmit forces, moments, and bending moments, and can simulate the elastoplastic deformation and brittle fracture of the rock.

[0038] S222, Calibration of microscopic parameters: The microscopic parameters of the discrete element model (such as particle friction coefficient, bond strength, and bond stiffness) cannot be directly obtained from laboratory tests and need to be determined through a "parameter calibration" method. The calibration process is as follows: S2221. Initial Value Estimation: Based on empirical formulas or reference values ​​in literature, set an initial range for each microscopic parameter. For example: particle contact modulus. Typically 0.5 to 2 times the macroscopic elastic modulus, initially set at 30 GPa. Bond strength. The compressive strength of the rock is on the same order of magnitude, initially set at 60 MPa. (Particle friction coefficient) For altered tuff, the value is usually taken as 0.3 to 0.6, and initially set as 0.5.

[0039] S2222 Numerical simulation test: Using the current microscopic parameters, simulate the model with virtual uniaxial compression test and Brazilian splitting test, and calculate the macroscopic mechanical response of the model (elastic modulus, Poisson's ratio, compressive strength, tensile strength).

[0040] S2223. Compare with measured values: Compare the simulation results with the actual mechanical parameters obtained from the indoor test in step S1. Calculate the relative error. Parameter adjustment: Use a trial-and-error method or a more efficient response surface methodology / particle swarm optimization to adjust the microscopic parameters. Repeat steps S2222-S2223 until the error between the simulation results and the measured values ​​is controlled within 10%.

[0041] Example calibration results are shown in Table 2 (for a certain deep altered tuff): Table 2 Example Results

[0042] Step S23: Construct a wellbore stability analysis model (see step S4 for details) Based on the particle aggregate, by traversing all particles and their contact information, the closed regions enclosed by the particles are identified as pores (fluid domains), and flow channels are established between adjacent pores and between pores and model boundaries. The initial opening of the flow channels is determined. The calculation is based on the spacing between adjacent particles and the amount of contact overlap. See steps S4 and S5 for the specific implementation method.

[0043] The purpose of model validation is to ensure that the constructed microscopic thermal-fluid solidification coupling model can accurately reflect the mechanical behavior of real altered tuff, providing a reliable basis for subsequent wellbore instability prediction.

[0044] As a specific example, this application includes the following verification methods: Verification Project 1: Verification of Macroscopic Mechanical Parameters Using the calibrated microscopic parameters, the virtual uniaxial compression test and the virtual Brazilian splitting test were rerun, and the following indicators were recorded and compared with the results of the laboratory test, as shown in Table 3: Table 3 Validation Results

[0045] The criterion is that if the relative errors of all four indicators are less than 10%, the model is considered to have passed the verification in terms of static properties.

[0046] Validation Item 2: Validation of Stress-Strain Curve Morphology The axial stress-strain curves obtained from the virtual uniaxial compression test were compared with those from the laboratory test. Key points to consider included: whether the slopes of the linear elastic segment were consistent; whether the peak strengths were similar; and whether the trends of the post-peak softening segment were similar (qualitative comparison).

[0047] Example: The simulated stress-strain curve basically coincides with the measured curve in the elastic segment, with a peak strength deviation of less than 5%. After the peak, it exhibits brittle drop, which is consistent with the brittle failure characteristics of the measured altered tuff. The verification is successful.

[0048] Validation Project 3: Penetration Characteristics Validation If conditions permit, a virtual permeability test can be conducted: a constant water pressure is applied to one side of the model to simulate fluid flow in the wellbore stability analysis model, and the equivalent permeability of the model is calculated. And compared with the permeability obtained from CT scan or indoor test in step S1. Compare. If If the wellbore stability analysis model is deemed reasonable (permeability is greatly affected by pore connectivity, and the allowable error is relaxed to 30%).

[0049] S3. Simulating the Real Drilling Environment: Initial horizontal in-situ stress, initial formation temperature field, pore pressure field, and initial fluid chemical environment conditions are applied to the microscopic thermal-fluid-solidification coupling model. Rock particles within a preset wellbore diameter centered on the wellbore center are removed from the model to simulate wellbore formation. As an example, based on the microscopic thermal-fluid-solidification coupling model verified in step S2, initial conditions are applied to simulate the in-situ formation state. First, horizontal in-situ stress is applied to the model boundary through a servo control mechanism: a maximum horizontal principal stress of 70 MPa is applied in the X direction, a minimum horizontal principal stress of 50 MPa is applied in the Y direction, and a vertical stress of 60 MPa (overburden pressure) is applied in the Z direction. Simultaneously, the initial formation temperature field is set to a uniform temperature of 135℃ (corresponding to a geothermal gradient of 2.7℃ / 100m at 4850m), the pore pressure field is set to a hydrostatic pressure of 48.5 MPa, and the initial fluid chemical environment condition is a drilling fluid filtrate ion activity of 0.85. Then, rock particles within a preset wellbore diameter (assumed to be 215.9 mm, corresponding to a 0.5 mm radius area after scaling down the model) centered on the wellbore are removed to simulate the process of drill bit forming the wellbore. After removing the particles, the particles around the wellbore are exposed, the original force balance is broken, and the model automatically enters the stress redistribution calculation stage.

[0050] S4. Construct a wellbore stability analysis model: Traverse all particles and their contact information in the model, define the closed region formed by altered tuff particles as pores, define the flow channels between contacting particles, and characterize the fluid conduction capacity of the flow channels by the channel opening. The thermal stress caused by temperature field changes, the seepage force caused by fluid flow, and the force adjustment of particles caused by reservoir pressure changes all act on the normal contact force between particles. The channel opening changes dynamically with the normal contact force between two particles. At the same time, the channel opening is corrected according to the degree of mineral hydration and solidification reaction, and a wellbore stability analysis model under the combined action of thermal flow solidification is established.

[0051] like Figure 2 As shown, drilling fluid enters the formation through the wellbore mud cake under wellbore pressure. In the microscopic thermal-fluid-solidification coupling model of this application, by traversing altered tuff particles and particle contact information, the relationship between pores and the altered tuff particles surrounding the pores is established, binding the altered tuff particles to the corresponding pores. The pores at the center of the closed region formed by the altered tuff particles become the fluid domain. Flow channels are set between the fluid domains to realize fluid flow, and additional channel properties are assigned to the contacts, thus defining the flow channels. The flow channels and the fluid domains, i.e., the interparticle pores, together constitute the entire wellbore stability analysis model.

[0052] The specific way in which the pipe opening dynamically changes with the normal contact force between the two particles in step S4 is as follows: When the normal contact force between two particles is compressive stress, the pipe opening is: (7) in This represents the initial opening of the pipe. This refers to the compressive force when the pipe opening is reduced to half of its initial opening. The current compressive force; When two particles are under tension or the cementation between particles breaks down, the pipe opening... (8) in It is a dimensionless multiplier. The distance between the centers of the two balls. Let be the radii of the two particles. As one example, the algorithm iterates through all particles and their contact information in the model, executing a pore-channel identification algorithm: closed regions enclosed by three or more altered tuff particles are defined as pores (fluid domains), identifying approximately 3800 pores; flow channels are defined between contacting particles, with each contact pair assigned an initial channel opening. a 0 = 0.5 μm. Thermal stress caused by temperature field changes (particle thermal expansion coefficient) α T =1.2×10 -5 K - ¹) The seepage force caused by fluid flow and the force adjustment of particles caused by changes in reservoir pressure work together to exert normal contact force between particles. F The pipe opening dynamically changes with the normal contact force: when there is compressive stress between the two particles, the pipe opening... a = a 0 / (1+ F / F 0), of which F 0 = 10N (compressive force that halves the opening); when the two particles are in a tensile state or the cementation fails, a = a 0+ l ( d -( R 1+ R 2)), l Take 0.8. At the same time, based on the degree of mineral hydration (water content)... oh ) and degree of curing reaction Correct the pipe opening: ,in k w =0.3, k c =0.2. The above together establish a wellbore stability analysis model under the combined effects of heat, fluid, solids, and chemistry.

[0053] S5. Simulation of drilling fluid entering deep altered tuff: Set the drilling fluid pressure, temperature, flow rate, and chemical composition. Use a wellbore stability analysis model to calculate the flow, heat transfer, and reaction processes of the drilling fluid in the flow channels and pores. Calculate the fluid pressure, temperature, water content, and reaction degree in each pore in real time. Control the radial expansion of altered tuff particles around the pore based on the water content, where the expansion force is proportional to the water content. Calculate the thermal expansion or thermal stress of the particles based on local temperature changes. Dynamically update the interparticle bonding strength, contact stiffness, friction parameters, and mechanical strength parameters based on the degree of hydrochemical reaction and solidification reaction.

[0054] The wellbore stability analysis model in step S5 includes: Formula for calculating flow rate in a flow pipe: (9) in Permeability coefficient, For pipe opening, The pressure difference between the two pores. This refers to the length of the pipe. Formula for the change in fluid pressure in pores: (10) in For fluid bulk modulus, For pore volume, This represents the total flow rate of fluid entering the pore. This represents the change in pore volume. Water content in pores: (11) like Figure 3 As shown, when drilling fluid enters the pores of altered tuff through the wellbore stability analysis model, the water content in the pores gradually increases. Based on the water content-swelling force proportional relationship established in this application (…),… The altered tuff particles around the pores undergo radial expansion. The expanded particles compress each other, leading to a reduction in pore volume and a redistribution of stress around the well, which may ultimately cause wellbore instability and failure. Figure 3 It visually demonstrates the process of altered tuff around the well expanding layer by layer from the inside out after being soaked in drilling fluid.

[0055] During drilling in deep and ultra-deep altered tuff, after water-based drilling fluid filtrate intrudes into the surrounding formation, clay minerals (mainly montmorillonite and illite) and residual volcanic glass in the altered tuff absorb water, undergoing hydration expansion and dispersion. The magnitude of the expansion force directly affects the wellbore stress distribution and wellbore stability. To accurately simulate this process in a microscopic thermal-fluid-solidification coupling model, this application establishes a quantitative relationship between water cut and expansion force. The expansion coefficient k... swell The specific altered tuff core needs to be calibrated through indoor hydration expansion experiments. The calibration method is the same as for mudstone, but the typical value range may be wider (e.g., 0.5~2.0 MPa).

[0056] Water content ω is defined as: the volume of fluid within a certain pore. With the total volume of the pores The ratio, i.e. When the water content is greater than 0, the altered tuff particles begin to absorb water and swell. (Swelling force) This refers to the additional force exerted by a single altered tuff grain on surrounding contacting grains due to hydration, with the direction away from the pore center.

[0057] This application proposes the following linear model: (12) in, The expansion coefficient (unit: N or MPa·m², depending on the force unit in the model) needs to be calibrated through indoor hydration expansion experiments. The model assumes that the expansion force is approximately linearly related to the water content when the water content is low (<30%); when the water content is high, piecewise linear or nonlinear corrections can be used, but for the sake of simplifying the calculation, this application prefers the linear model.

[0058] In addition, this application conducted calibration experiments on the coefficient of thermal expansion. The specific steps are as follows: 1. Drill altered tuff cores from the target strata and process them into standard cylindrical specimens with a diameter of 25 mm and a height of 50 mm.

[0059] 2. Place the specimen in a triaxial dilatometer, apply an axial load (e.g., 5 MPa) simulating overburden pressure, and maintain lateral constraints.

[0060] 3. Introduce a solution with the same composition as the drilling fluid in the field into the sample and record the axial expansion force and axial strain during the water absorption process of the sample.

[0061] 4. At the same time, the moisture content ω of the sample at different times was determined by nuclear magnetic resonance or gravimetric method.

[0062] 5. Plot the "expansion force-moisture content" curve. Perform a linear fit within the moisture content range of 0-15%, and the slope is the curve. .

[0063] For example, for a deep altered tuff in a certain sea area (well depth 4850m, montmorillonite content 32%), the calibration yielded... (For particle-scale models, normalization based on particle volume is required.) Alternatively, a dimensionless form can be used: expansion pressure. ,in .

[0064] In step S5, after each time step, an expansion force is applied to all altered tuff particles bound to that pore based on the current pore water content ω. The specific algorithm is as follows: For each particle surrounding the pore, calculate the expansion force vector: its direction is from the pore center to the particle center, and its magnitude is... .

[0065] The expansion force is superimposed on the resultant force of the particles and used to solve the discrete element motion equations in the next time step.

[0066] In this way, the higher the water content of the pores, the greater the expansion force on the surrounding particles, which pushes the particles outward, resulting in an increase in pore volume and a change in permeability, thereby simulating the gradual destruction process of the well wall caused by hydration expansion.

[0067] It should be noted that the wellbore stability analysis model in step S5 also includes a temperature field evolution equation, and the temperature changes in the pore and particle system satisfy the energy conservation equation:

[0068] in, r For the density of the medium, c For specific heat capacity, T For temperature, t For time, l Thermal conductivity, Q The heat source term; the temperature field is used to characterize the heat exchange process between the drilling fluid and the surrounding rock of the wellbore, and further affects the opening of the flow channel and the stability of the wellbore by influencing the thermal expansion of particles and the contact force between particles. It should be noted that the curing reaction process in step S5 is characterized by a reaction kinetic equation, specifically: the degree of curing reaction satisfies:

[0069] in, x To indicate the degree of curing reaction, k(T) This is the temperature-dependent reaction rate constant. f(C) As a reaction function related to the concentration of fluid chemical components, the degree of solidification reaction is used to characterize the cementation and reinforcement between particles in the wellbore surrounding rock, the precipitation and sealing of reaction products, and the local structural reinforcement process.

[0070] It should be noted that the effect of the curing reaction on the particle mechanical strength parameters in step S5 is as follows: the cohesion or strength parameters of the particle system change dynamically with the degree of curing reaction, satisfying... s=s 0 +vx ,in, s These are the strength parameters after curing. s 0 represents the initial strength parameter. β The curing reinforcement coefficient, x The degree of curing reaction is considered; dynamically changing mechanical parameters are used to update the interparticle bonding strength, contact stiffness, and local wellbore instability resistance.

[0071] As one example, Figure 4 This demonstrates a comparison of wellbore collapse morphologies considering and not considering tuff hydration and expansion. For example... Figure 4 As shown in section (b), when the hydration and expansion of tuff are not considered, the collapse range of the well wall is relatively small; as Figure 4 As shown in section (a), when the tuff around the well wall undergoes hydration and expansion, the strength of the tuff decreases, and the tuff around the well is prone to collapse. Moreover, the collapse range of the well wall becomes significantly larger, and it collapses in an elliptical shape along the direction of the minimum principal stress. It can be seen that the hydration and expansion of tuff makes the well wall more prone to instability and collapse.

[0072] Figure 5 This demonstrates a comparison of wellbore instability considering and not considering tuff hydration expansion when drilling fluid pressure exceeds formation fracturing pressure. Figure 5 As shown in section (b), without considering the hydration and expansion of tuff, only localized fractures occur in the wellbore; as Figure 5 As shown in section (a), when the drilling fluid pressure is greater than the formation fracture pressure and the tuff undergoes hydration and expansion, it is prone to collapse in the direction of minimum principal stress, and then forms cracks extending along the direction of maximum principal stress, making it easier for the drilling fluid to penetrate the formation and form collapses and fractures over a larger area.

[0073] S6. Change the drilling fluid pressure and repeat step S5 to obtain the wellbore instability and failure under different drilling fluid densities. Based on the preset allowable wellbore enlargement rate, determine the drilling fluid safe density window for the altered tuff formation.

[0074] S7. Obtain the actual drilling fluid density at the drilling site and compare it with the drilling fluid safety density window; adjust the drilling fluid density in real time according to the comparison results: if the actual drilling fluid density is higher than the upper limit of the safety density window, reduce the drilling fluid density; if it is lower than the lower limit of the safety density window, increase the drilling fluid density.

[0075] It should be noted that although the microscopic heat-fluid-solidification coupling model constructed in steps S2 to S6 (referred to as the "high-fidelity model") has high accuracy, its single calculation takes an extremely long time (several hours to several days), which cannot meet the real-time decision-making needs of drilling sites. To resolve this contradiction, this application proposes a two-layer architecture of offline preparation + online prediction: Offline preparation phase: On a high-performance computing cluster, a high-fidelity model is used to perform batch simulations of a large number of parameter combinations to generate a training dataset covering various working conditions.

[0076] Online implementation phase: A lightweight PI-GNN surrogate model is trained, with inputs consisting of parameters easily obtainable in the field (such as geostress, drilling fluid density, etc.), and outputs including a safe density window and wellbore enlargement rate. The surrogate model's prediction time is in the millisecond range, enabling real-time response.

[0077] Therefore, the method also includes an offline modeling phase, which includes: A1. Within the preset range of drilling fluid density, drilling fluid pressure, and geostress parameters, multiple sets of input parameter combinations are generated using Latin hypercube sampling. Steps S2 to S6 of the microscopic heat flow solidification coupling model are run in batches to obtain N1 sets of high-fidelity training data. At the same time, a reduced-order analytical model is used to quickly generate N2 sets of low-fidelity training data, where N2 ≥ 5 × N1. A2. Construct a Physical Information Graph Neural Network (PI-GNN) as a surrogate model. Its inputs are: the adjacency matrix, node features, and global features of the discrete element particle graph. Its outputs are: the collapse pressure equivalent density, the fracture pressure equivalent density, and the wellbore enlargement rate. The loss function of PI-GNN consists of three parts: Data fitting loss: the weighted mean square error between the surrogate model output and the high-fidelity / low-fidelity data; Physical constraint loss: The stress field output by the surrogate model should satisfy the discrete element force balance equation, and the output pressure field should satisfy the equivalent form of Darcy's law for pore flow; Contact force-pipe opening consistency loss: The relationship between the contact force output by the intermediate layer of the proxy model and the pipe opening should be consistent with the dynamic change of the pipe opening with the normal contact force between the two particles in step S4. A residual multifidelity training strategy is adopted: first, PI-GNN is pre-trained using low-fidelity data, and then the parameters of the last two network layers are fine-tuned using high-fidelity data. Furthermore, in steps S6 and S7, the comparison of actual drilling fluid density and the acquisition of the safe density window are completed in real time by calling the trained PI-GNN proxy model, with a single prediction taking milliseconds.

[0078] The following example illustrates a dual-fidelity proxy model based on a Physical Information Graph Neural Network (PI-GNN).

[0079] To reduce the time required for a single wellbore stability prediction from several hours in high-fidelity discrete element simulation to milliseconds while ensuring prediction accuracy, this application constructs a Physics-Informed Graph Neural Network (PI-GNN) surrogate model and adopts a multi-fidelity training strategy.

[0080] I. Generation of Training Data Taking a deep and ultra-deep altered tuff formation (well depth 4850m, montmorillonite content 32%, porosity 12.5%) as an example: High-fidelity data: 200 sets of input parameter combinations were generated using Latin hypercube sampling. For each set of parameters, the high-fidelity discrete element fluid-structure interaction model in steps S2-S6 was used for calculation to obtain the corresponding collapse pressure equivalent density. r c Rupture pressure equivalent density r f and wellbore enlargement rate d The high-fidelity data computation took approximately 200 hours.

[0081] Low-fidelity data: Based on the same parameter sampling space, a reduced-order analytical model of porosimetry (ignoring the non-uniformity of microparticle structure and hydration expansion) is used to quickly generate 2000 sets of input-output data, with a single set of calculations taking less than 1 second. Although the accuracy of low-fidelity data is lower (the average relative error compared to the high-fidelity model is about 15%), it can cover a wider parameter space and has extremely low generation cost.

[0082] The specific structure of the PI-GNN network is as follows: The discrete element-level aggregate is abstracted as a graph structure G=(V,E), where: Node V: Each particle is a node, and node characteristics include particle radius. R i Spatial coordinates ( x i , y i ), the combined external forces currently affecting F i ; Edge E: Each pair of contacting particles is connected by an edge, the edge characteristics of which include normal contact force. F n Tangential contact force F s Pipe opening ; Global characteristics: Horizontal maximum principal stress s H Horizontal minimum principal stress s h Overlying strata pressure s v Drilling fluid density ρ m Drilling fluid ion activity a w .

[0083] PI-GNN employs a three-layer graph convolutional network (GCN): First layer: Input node feature dimension 8, output hidden dimension 64; Second layer: Hidden dimension 64, output hidden dimension 128; Third layer: Hidden dimension 128, output graph-level feature vector of 256 dimensions; Global pooling: Average the 256-dimensional features to obtain the global feature vector; Fully connected output layer: Three fully connected layers (256→128→64→3), output... r c , r f , d .

[0084] The activation function used is ReLU, and the graph convolutional layer uses the Chebyshev polynomial approximation to reduce computation.

[0085] Traditional data-driven surrogate models only minimize the mean squared error between the predicted value and the training data, which is prone to overfitting and has poor extrapolation ability. This application embeds three physical constraints into the loss function: Force balance constraints Loss phys1 For each particle node i, the sum of the contact forces predicted by the surrogate model should be approximately equal to the external load (generated by the geostress gradient) acting on that particle. Definition: (14) in The contact force in the edge features output by the surrogate model. Calculated from geostress.

[0086] Flow continuity constraints Loss phys2 For each pore (corresponding to a closed loop in the diagram), the algebraic sum of the inflow and outflow rates should equal the pore volume change rate. The surrogate model can output the pore pressure field p_f, from which the flow residual can be calculated. (15) in The flow rate in the pipeline is calculated using formula (10). This represents the change in pore volume.

[0087] Pipe opening-contact force consistency constraint Loss phys3 Formulas (8) and (9) in step S4 establish the pipe opening. a Contact force with normal direction F The physical relationship between them. The surrogate model predicts an intermediate variable from the edge features. â ,Require: (16) in Calculated according to formula (8) or (9).

[0088] The total loss function is: Through validation set optimization, select... .

[0089] To avoid overfitting due to insufficient high-fidelity data, the following two-stage training method is adopted: Pre-training phase: Using only 2000 sets of low-fidelity data, PI-GNN was trained for 200 rounds with the loss loss of pure data fitting, and the learning rate was 0.001.

[0090] Fine-tuning phase: Freeze the parameters of the first two layers of the graph convolutional network, use 200 sets of high-fidelity data, and continue training for 50 rounds with the full loss function (including physical constraints) and a learning rate of 0.0001.

[0091] This strategy enabled PI-GNN to achieve a determination coefficient R² of 0.96 (collapse pressure), 0.94 (rupture pressure), and 0.91 (wellbore enlargement rate) on the test set, while reducing the amount of high-fidelity data by 80%, and a mean absolute percentage error (MAPE) of less than 5%.

[0092] The trained PI-GNN proxy model was deployed on an on-site edge computing device (NVIDIA Jetson AGXOrin). The average time for a single prediction (input parameters → output safety window + wellbore enlargement rate) was 8.3 milliseconds, meeting real-time control requirements. Compared to traditional DNN models, the PI-GNN proposed in this application reduced prediction errors by approximately 40% under extrapolation conditions (such as drilling fluid density exceeding the training range by 15%).

[0093] It should be noted that during field drilling, formation conditions may deviate from the assumptions made during offline modeling, causing discrepancies between the wellbore enlargement rate predicted by the surrogate model and the actual measured value. To maintain model accuracy, this application designs a closed-loop self-correction mechanism: comparing the actual wellbore data measured while drilling with the predicted value from the surrogate model, and automatically triggering a model update when the deviation exceeds a threshold.

[0094] Specifically, the method also includes a closed-loop self-calibration step: during drilling, the actual well diameter data measured while drilling is compared with the well diameter enlargement rate predicted by the PI-GNN surrogate model. If the relative deviation between the two exceeds the preset value, the high-fidelity micro-thermal-fluid solidification coupling model is triggered to perform a supplementary calculation using the latest measured data, and the PI-GNN surrogate model is updated by transfer learning using the supplementary calculation results.

[0095] As one example, the deviation triggering condition is as follows: Logging while drilling (LWD) measures the wellbore diameter in real time, typically recording a data point every 1 meter. Relative deviation is defined as follows: when When this happens, the model is updated.

[0096] The model update process is as follows: 1. Incremental data acquisition: Collect the measured data (well diameter, drilling fluid density, geostress inversion value, etc.) of the nearest 100-meter well section.

[0097] 2. High-fidelity model recalibration: Using these measured data as new boundary conditions, a supplementary calculation is performed on the high-fidelity model from steps S2-S6. Since only local parameters (such as mechanical parameters in the near-wellbore zone) are modified, the computational workload is relatively small and can usually be completed within a few hours.

[0098] 3. Generate a supplementary training set: Using the recalibrated high-fidelity model, generate 500 sets of data again by Latin hypercube sampling around the parameter space (±10%) near the current operating condition.

[0099] 4. Transfer learning to update the surrogate model: Merge the supplementary training set with the original training set, but use a fine-tuning strategy: keep the parameters of the first two layers of the surrogate model unchanged, and only retrain the last 1-2 layers. This can avoid catastrophic forgetting and quickly adapt to new working conditions.

[0100] 5. Verification and Deployment: Verify the accuracy of the updated agent model using the latest test data. If δ < 5%, replace the online model.

[0101] For example: A well was drilled to the 4850m altered tuff section. The surrogate model predicted a wellbore enlargement rate of 12%, while the actual LWD measurement was 18%. Since δ=50%>10%, an update was triggered.

[0102] Field engineers collected measured data from the 2800-3100m well section, including well diameter, drilling fluid density, and drilling speed. Using this data as constraints, a high-fidelity model was run for inversion, revealing that the actual formation horizontal principal stress was 8 MPa higher than the preset value, and the clay mineral content was higher (with a larger expansion coefficient). The expansion coefficient was then recalibrated. The pressure was increased from 0.8 MPa to 1.2 MPa. 500 sets of data were generated in the parameter space (geological stress ±10%, drilling fluid density 1.3-1.7) using the updated high-fidelity model. The original deep neural network surrogate model was fine-tuned (freezing the first two layers and retraining the output layer and the penultimate layer). After 20 training cycles, the deviation between the predicted wellbore enlargement rate and the measured rate decreased to 4%. The updated model was deployed online to continue guiding subsequent drilling.

[0103] Through the aforementioned closed-loop self-correction, the prediction system of this application can continuously adapt to formation changes, maintain high accuracy, and greatly reduce drilling risks.

[0104] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting wellbore instability in deep altered tuff using thermal-fluid-solidification coupling, characterized in that, The steps include the following: S10. Based on the stratigraphic mechanical parameters, mineral composition, thermodynamic and hydrochemical parameters of the deep altered tuff strata, a microscopic thermal-fluid solidification coupling model is constructed using the computational fluid dynamics-discrete element method. S20, apply initial geostress, initial formation temperature field, pore pressure field and initial fluid chemical environment conditions to the microscopic thermal flow solidification coupling model, and delete rock particles within the preset wellbore diameter range in the model to simulate wellbore formation; S30, Traverse all particles and their contact information in the model, define the closed area formed by the particles as a pore, define the flow channel between the contacting particles, the opening of the flow channel dynamically changes with the normal contact force between the two particles, and correct the opening of the channel according to the degree of mineral hydration and solidification reaction, and establish a well wall stability analysis model under the combined action of thermal flow solidification. S40, set drilling fluid parameters, use the wellbore stability analysis model to calculate the flow, heat transfer and reaction process of drilling fluid in the flow channel, and calculate the water content in each pore in real time. Control the radial expansion of altered tuff particles around the pore according to the water content, and dynamically update the mechanical strength parameters between particles according to the degree of hydrochemical reaction and solidification reaction to obtain the wellbore instability and failure under different drilling fluid densities. Obtain the drilling fluid safe density window of the altered tuff formation according to the preset allowable well diameter enlargement rate.

2. The method for predicting wellbore instability in deep altered tuff using thermal-fluid solidification coupling as described in claim 1, characterized in that, In step S30, the specific way in which the pipe opening dynamically changes with the normal contact force between the two particles is as follows: when the normal contact force between the two particles is compressive stress, the pipe opening satisfies the formula ,in This represents the initial opening of the pipe. F 0 represents the compressive force when the pipe opening is reduced to half of its initial opening. F The current compressive force; when the two particles are in a tensile state or the cementation between the particles fails, the pipe opening satisfies the formula. ,in λ It is a dimensionless multiplier. d The distance between the centers of the two balls. R 1. R 2 represents the radii of the two particles; The specific method for correcting the pipe opening based on the degree of mineral hydration and solidification reaction is as follows: using the formula... Make corrections, among which k w This is the hydration correction factor. ω Moisture content, k c This is the curing correction factor. This indicates the degree of curing reaction.

3. The method for predicting wellbore instability in deep altered tuff using thermal-fluid solidification coupling as described in claim 1, characterized in that, In step S40, the wellbore stability analysis model includes: Flow rate calculation formula in a flow pipe ,in k Permeability coefficient, For pipe opening, p The pressure difference between the two pores. L This refers to the length of the pipe. Formula for the change of fluid pressure in pores ,in For fluid bulk modulus, For pore volume, This represents the total flow rate of fluid entering the pore. This represents the change in pore volume. Water content in pores ; The wellbore stability analysis model also includes a temperature field evolution equation, where temperature changes in the pore and particle system satisfy the energy conservation equation: ,in ρ For the density of the medium, c For specific heat capacity, T For temperature, t For time, λ Thermal conductivity, Q For heat source items; The specific way in which the temperature field affects the particle contact state and the flow channel opening is as follows: when the particle temperature changes, the particle radius increment satisfies... ,in The coefficient of thermal expansion of the particles. The initial radius of the particle, Δ T This represents the change in temperature; the thermal expansion of particles causes changes in the normal contact force between particles, which in turn drives the dynamic change in the opening of the flow channel.

4. The method for predicting wellbore instability in deep altered tuff using thermal-fluid solidification coupling as described in claim 1, characterized in that, In step S40, the specific method for controlling the radial expansion of altered tuff particles around the pores based on the water content is as follows: when the water content... At that time, a radial expansion force is applied to all particles surrounding the pore, away from the pore center. The magnitude of the expansion force is... F swell = k swell · ω ,in k swell The coefficient of expansion is determined through indoor hydration expansion experiments.

5. The method for predicting wellbore instability in deep altered tuff using thermal-fluid solidification coupling as described in claim 1, characterized in that, In step S40, the degree of curing reaction is characterized by a reaction kinetic equation: the degree of curing reaction satisfies , ξ To indicate the degree of curing reaction, k(T) This is the temperature-dependent reaction rate constant. f(C) A reaction function related to the concentration of fluid chemical components; The specific way in which the degree of curing reaction affects the particle mechanical strength parameters is as follows: the cohesion or strength parameters of the particle system change dynamically with the degree of curing reaction, satisfying the formula... σ=σ 0 +βξ ,in σ These are the strength parameters after curing. σ 0 represents the initial strength parameter. β This represents the curing enhancement factor.

6. The method for predicting wellbore instability in deep altered tuff using thermal-fluid solidification coupling as described in claim 1, characterized in that, Before step S10, the method further includes obtaining micromechanical parameters through nanoindentation testing: performing nanoindentation testing on the altered tuff sample, extracting the load-displacement curve, calculating the contact stiffness, contact depth, contact area, and reduced elastic modulus based on the Oliver-Pharr method, and then solving for the microelastic modulus and microhardness of different micromineral phases, which serve as prior input parameters for particle contact modulus and bond strength in the microthermal-fluid-solidification coupling model.

7. The method for predicting wellbore instability in deep altered tuff using thermal-fluid solidification coupling as described in claim 1, characterized in that, Following step S10, a model verification step is also included: using the calibrated microscopic parameters, a virtual uniaxial compression test and a virtual Brazilian splitting test are performed on the microscopic thermal-fluid-curing coupling model. The simulated macroscopic mechanical parameters are compared with the measured values ​​from the indoor test. When the relative errors of the four indicators—elastic modulus, Poisson's ratio, uniaxial compressive strength, and tensile strength—are all less than 10%, the model is deemed to have passed verification.

8. The method for predicting wellbore instability in deep altered tuff using thermal-fluid solidification coupling as described in claim 1, characterized in that, Following step S40, an offline modeling stage is also included, which is used to build a physical information graph neural network surrogate model to achieve rapid prediction; The offline modeling phase includes the following steps: Within the preset range of drilling fluid density, drilling fluid pressure, and geostress parameters, multiple sets of input parameter combinations are generated using Latin hypercube sampling. The microscopic heat flow solidification coupling model described in steps S10 to S40 is run in batches to obtain N1 sets of high-fidelity training data. At the same time, N2 sets of low-fidelity training data are generated using a reduced-order analytical model, where N2 ≥ 5 × N1. A physical information graph neural network is constructed as a surrogate model. Its inputs are the adjacency matrix, node features, and global features of the discrete element particle graph, and its outputs are the collapse pressure equivalent density, the fracture pressure equivalent density, and the well diameter enlargement rate. The loss function of the physical information graph neural network includes data fitting loss, physical constraint loss, and contact force-pipe opening consistency loss: where, data fitting loss is the weighted mean square error between the surrogate model output and the high-fidelity and low-fidelity data; physical constraint loss is that the stress field output by the surrogate model satisfies the discrete element force balance equation and the output pressure field satisfies the equivalent form of Darcy's law for pore flow; contact force-pipe opening consistency loss is that the relationship between the contact force output by the intermediate layer of the surrogate model and the pipe opening is consistent with the dynamic change of the pipe opening with the normal contact force between the two particles in step S30; A residual multi-fidelity training strategy is adopted: first, the physical information graph neural network is pre-trained using low-fidelity data, and then the parameters of the last two layers are fine-tuned using high-fidelity data.

9. The method for predicting wellbore instability in deep altered tuff using thermal-fluid solidification coupling as described in claim 8, characterized in that, During the drilling process, a closed-loop self-correction step is also included: the actual well diameter data measured while drilling is compared with the well diameter change rate predicted by the physical information graph neural network surrogate model. If the relative deviation between the two exceeds a preset value, the microscopic heat flow solidification coupling model is triggered to perform supplementary calculations using the latest measured data, and the physical information graph neural network surrogate model is updated by transfer learning using the supplementary calculation results.

10. A method for real-time control of wellbore instability, characterized in that, The steps include the following: Using the thermal-fluid-solidification coupling method for predicting wellbore instability in deep altered tuff formations as described in any one of claims 1 to 9, the drilling fluid safety density window for deep altered tuff formations is obtained. Obtain the actual drilling fluid density at the drilling site and compare the actual drilling fluid density with the drilling fluid safety density window; The drilling fluid density is adjusted in real time based on the comparison results: if the actual drilling fluid density is higher than the upper limit of the safe density window, the drilling fluid density is reduced; if the actual drilling fluid density is lower than the lower limit of the safe density window, the drilling fluid density is increased.

Images