Prediction method and system for casing strength under temperature-pressure alternating working condition

Abstract

The invention belongs to the technical field of petroleum pipe engineering structure safety, and discloses a method and system for predicting the strength of a casing under the temperature-pressure alternating working condition, and the method comprises the steps: building a casing physical parameter calculation model, calculating coupling alternating parameters, carrying out a thermal-mechanical coupling alternating test, and obtaining casing yield strength test data. Obtaining microstructure characteristic experimental data influencing the strength of the casing, and obtaining a training sample; constructing a casing yield strength prediction model fusing physical constraint and data driving, and training the prediction model through the training sample; and predicting the yield strength of the casing through the prediction model. According to the method, the coupling alternating parameter load is used as the load condition of the thermal coupling alternating test to obtain the microstructure characteristic experimental data, the microstructure characteristic experimental data is used as the sample data, and the sample data has load, temperature and microstructure evolution characteristics at the same time, so that the prediction model obtained through training can accurately predict the yield strength of the casing.

Description

Technical Field

[0001] This invention belongs to the field of structural safety technology for oil pipeline engineering, and specifically relates to a method and system for predicting the strength of casing under alternating temperature and pressure conditions. Background Technology

[0002] With the development of unconventional oil and gas fields such as shale oil / gas, tight oil / gas, and deep oil and gas, multi-stage volumetric fracturing technology for horizontal or deep wells is being used more and more widely. Large-volume fracturing fluid is injected into the target reservoir in a short period of time, and production is sought after maintaining pressure. Subsequently, a multi-stage fracturing production cycle is required, and the wellbore experiences large-scale periodic temperature and pressure fluctuations.

[0003] The static strength formulas specified in standards such as API 5C3, which are commonly used in the current technology for casing design evaluation, do not take into account the temperature and pressure coupling cycle effect. Indoor simulation of casing strength evaluation under alternating temperature and pressure conditions is costly and time-consuming, and the test results have great limitations. Although finite element analysis can simulate the thermo-mechanical coupling field, the calculation of a single full-size model is time-consuming and cannot meet the requirements for real-time casing strength prediction.

[0004] In the prior art, for example, the patent with publication number CN112749514A proposes a method for predicting the extrusion resistance strength of casing based on neural networks, but it does not solve the problem of cumulative damage caused by temperature and pressure coupling cycles.

[0005] For example, the journal's method for predicting the stress of thermal recovery tubing based on an improved PSO-BP neural network (Cui Lu, Li Mingfeng, Chang Wenquan, Pipeline Technology and Equipment, CAS 2024, No. 2) improved the accuracy of stress prediction for thermal recovery tubing, but did not consider the influence mechanism of the differences in the microstructural characteristics of the same steel grade casing on the strength variation.

[0006] In summary, existing technologies for casing string design and safety evaluation are mostly based on the static yield strength of the casing material, without considering the attenuation of the casing yield strength under thermo-mechanical coupling alternating conditions, resulting in a small safety factor. Full-scale indoor tests simulating temperature and pressure alternating conditions are costly and time-consuming, and the test results have limited applicability. Calculations for casing strength prediction using finite element thermo-mechanical coupling models are time-consuming and cannot achieve real-time prediction. Furthermore, none of the above methods consider the micro-damage evolution mechanism of the casing material, and the strength attenuation prediction based solely on the casing material steel grade has low accuracy. Summary of the Invention

[0007] To address the above problems, this invention provides a method and system for predicting the strength of bushings under alternating temperature and pressure conditions, employing the following technical solution: A method for predicting the strength of bushings under alternating temperature and pressure conditions includes the following steps: Based on the actual working parameters of the casing, a calculation model for the physical parameters of the casing is established, and based on the calculation model for the physical parameters of the casing, the coupled alternating parameters of the casing service temperature and the equivalent axial load are calculated. The coupled alternating parameters of the casing service temperature and the equivalent axial load are used as load conditions to conduct a thermo-mechanical coupled alternating test to obtain the casing yield strength test data. The microstructure characteristics of the original casing and the casing after thermo-coupling alternating test were characterized to obtain experimental data on the microstructure properties that affect the casing strength. Preprocessing of casing material parameters, coupling alternating parameters, casing yield strength test data, and microstructure experimental data is performed to obtain training samples. A casing yield strength prediction model integrating physical constraints and data-driven approaches was constructed, and the model was trained using training samples. The actual working parameters of the casing, which are acquired in real time, and the experimental data of the microstructure characteristics related to the casing strength are input into the trained casing yield strength prediction model, and the yield strength of the casing is output.

[0008] Furthermore, the casing physical parameter calculation model includes the casing axial load calculation model and the wellbore temperature field calculation model during hydraulic fracturing.

[0009] Furthermore, based on the actual operating parameters of the casing, a calculation model for the axial load of the casing during hydraulic fracturing is established, including the following steps: Based on the experimental simulation results, the axial stress of the casing during hydraulic fracturing is determined to be the sum of the compressive stress generated by the buoyancy of the casing in the cementing cement, the tensile stress caused by the casing's own weight, the axial stress caused by the casing bending, and the additional axial stress generated by the casing during the hydraulic fracturing process.

[0010] Furthermore, the wellbore temperature field calculation model during hydraulic fracturing includes the fluid temperature change equation inside the casing, the thermal balance equation of the casing inner wall unit, and the formation temperature change equation.

[0011] Furthermore, the coupled alternating parameters include axial load, axial stress amplitude, axial stress alternating frequency, holding time, temperature amplitude, temperature alternating frequency, axial displacement / strain, number of cycles, and current cycle state.

[0012] Furthermore, using the coupled alternating parameters of the casing service temperature and equivalent axial load as load conditions, a thermo-mechanical coupled alternating test is conducted to obtain the casing yield strength test data, including the following steps: The coupled alternating parameters of the casing service temperature and equivalent axial load are used as load conditions, including: selecting the maximum axial load fluctuation range and fluctuation frequency, and the maximum temperature fluctuation range and fluctuation frequency among the coupled alternating parameters as load conditions. A thermo-mechanical coupling alternating test was conducted on the casing. After experiencing different thermo-mechanical coupling alternating cycles, a tensile test was immediately performed to obtain the casing yield strength test data.

[0013] Furthermore, the experimental data on microstructure characteristics include grain size, the proportion of high-angle grain boundaries, and dislocation density.

[0014] Furthermore, the casing material parameters, coupling alternating parameters, casing yield strength test data, and microstructure characteristic test data are preprocessed to obtain training samples, including the following steps: Collect timestamps, coupled alternating parameters, casing yield strength test data, and microstructure characteristic test data to obtain a dataset; Data preprocessing and feature extraction were performed on the dataset. The yield strength in the casing yield strength test data was correlated with the actual working condition parameters of the casing at the corresponding time, the microstructure characteristics data, and the microstructure characteristics data of the casing before the thermo-coupling alternating test to obtain training samples.

[0015] Furthermore, a casing yield strength prediction model integrating physical constraints and data-driven approaches is constructed, including the following steps: A dynamic prediction model framework is constructed, which includes an LSTM layer, a Dense layer, an Attention layer, a BatchNormalization layer, and a NonNeg layer. Embed material constitutive parameters in the casing yield strength prediction model; A multi-branch feature extraction architecture is constructed in the casing yield strength prediction model. The multi-branch feature extraction architecture includes mechanical load branch, thermal load branch and microstructure and strength branch. Define the physical constraint models for the NonNeg layer, including the Hall-Petch grain refinement strengthening model, the Taylor dislocation strengthening model, and the cyclic softening model; Define an Attention layer to fuse material mechanical and thermal characteristics; Define the physical baseline term and data-driven correction term for the casing yield strength prediction model; Define the input layer, output layer, and loss function of the casing yield strength prediction model to obtain a casing yield strength prediction model that integrates physical constraints and data-driven approaches.

[0016] Furthermore, the mechanical load branch is used to extract the pressure from the actual working parameters of the casing, as well as the axial stress amplitude, alternating frequency, and holding time from the coupled alternating parameters; The thermal load branch is used to extract the fracturing temperature of the actual casing operating parameters, as well as the temperature amplitude and temperature alternation frequency in the coupled alternating parameters; The microstructure and strength branches are used to extract experimental data on microstructure properties.

[0017] Furthermore, the casing yield strength prediction model is trained using training samples, including the following steps: The input and output features in the training samples are normalized, and the normalized training samples are divided into training set, validation set and test set according to the proportion. The casing yield strength prediction model is trained in stages using the training set, including: the first stage freezes the data-driven layer for physical term pre-training; the second stage, during joint fine-tuning, adopts a dynamic weight adjustment strategy to adjust the weights of the physical constraint terms and the data fitting terms. The hyperparameters of the casing yield strength prediction model were adjusted using the validation set; The performance of the casing yield strength prediction model is evaluated using a test set. If the performance indicators of the casing yield strength prediction model meet the requirements, the training is complete.

[0018] Further, data preprocessing and feature extraction are performed on the dataset, including the following steps: Remove outliers, noise, and faulty data from the dataset; The test data of fracturing temperature, pressure, axial load, axial displacement / strain, yield strength and microstructure characteristics are synchronized according to the timestamp. The feature data extracted from the dataset includes input features, cumulative damage features, and material features; The extracted feature data is preprocessed to obtain key feature combinations, including thermo-coupling factors, microstructure characteristics test data interaction terms, and asymmetric loading coefficients.

[0019] This invention also provides a system for predicting the strength of bushings under alternating temperature and pressure conditions, comprising: The calculation module is used to establish a calculation model of the casing's physical parameters based on the actual working parameters of the casing, and to calculate the coupled alternating parameters of the casing's service temperature and equivalent axial load based on the calculation model of the casing's physical parameters. The yield test apparatus is used to conduct thermo-mechanical coupled alternating tests using the coupled alternating parameters of the casing service temperature and equivalent axial load as load conditions to obtain test data on the casing yield strength. The first data processing module is used to characterize the microstructure of the original casing and the casing after the thermo-coupling alternating test, and to obtain experimental data on the microstructure characteristics that affect the casing strength. The second data processing module is used to preprocess the casing material parameters, coupling alternating parameters, casing yield strength test data and microstructure characteristic test data to obtain training samples. The model training module is used to build a casing yield strength prediction model that integrates physical constraints and data-driven approaches, and to train the casing yield strength prediction model using training samples. The prediction module is used to input the real-time acquired actual working condition parameters of the casing and the experimental data of microstructure characteristics related to the casing strength into the trained casing yield strength prediction model, and output the yield strength of the casing.

[0020] Furthermore, the yield test apparatus is specifically used for: The coupled alternating parameters of the casing service temperature and equivalent axial load are used as load conditions, including: selecting the maximum axial load fluctuation range and fluctuation frequency, and the maximum temperature fluctuation range and fluctuation frequency among the coupled alternating parameters as load conditions. A thermo-mechanical coupling alternating test was conducted on the casing. After experiencing different thermo-mechanical coupling alternating cycles, a tensile test was immediately performed to obtain the casing yield strength test data.

[0021] Furthermore, the model training module is specifically used for: A dynamic prediction model framework is constructed, which includes an LSTM layer, a Dense layer, an Attention layer, a BatchNormalization layer, and a NonNeg layer. Embed material constitutive parameters in the casing yield strength prediction model; A multi-branch feature extraction architecture is constructed in the casing yield strength prediction model. The multi-branch feature extraction architecture includes mechanical load branch, thermal load branch and microstructure branch. Define the physical constraint models for the NonNeg layer, including the Hall-Petch grain refinement strengthening model, the Taylor dislocation strengthening model, and the cyclic softening model; Define an Attention layer to fuse material mechanical and thermal characteristics; Define the physical baseline term and data-driven correction term for the casing yield strength prediction model; Define the input layer, output layer, and loss function of the casing yield strength prediction model to obtain a casing yield strength prediction model that integrates physical constraints and data-driven approaches.

[0022] The beneficial effects of this invention are: 1. This invention uses coupled alternating parameter load as the load condition for thermo-mechanical coupled alternating test to obtain experimental data on microstructure characteristics, and uses the casing material parameters, coupled alternating parameters, casing yield strength test data and microstructure characteristic test data as sample data. The sample data has load, temperature and microstructure evolution characteristics, so that the casing yield strength prediction model trained by the sample data can accurately predict the yield strength of the casing.

[0023] 2. This invention is based on a multi-scale physical constraint-data-driven fusion model to obtain a casing yield strength prediction model. It adopts a multi-branch feature extraction architecture to extract load, temperature and microstructure evolution features respectively, further improving the model prediction accuracy and providing technical support for the design and safety evaluation of unconventional well casing materials.

[0024] 3. This invention is based on a simulated thermal coupling alternating strength evaluation test of casing material. It obtains parameters of temperature and pressure alternating conditions, stress and temperature alternating amplitude and frequency of different thermal coupling alternating tests, original microstructure characteristics of casing material, and data such as the strength and microstructure evolution characteristics of casing material after different thermal coupling alternating cycles. A multi-scale dataset is established, and deep learning algorithms such as Long Short-Term Memory Recurrent Neural Network (LSTM) + Dense Fully Connected Layer + Attention Mechanism are used to construct a dynamic prediction model of the yield strength of casing material under temperature and pressure alternating conditions that integrates physical constraints and data-driven approaches. This model can significantly improve the prediction accuracy and efficiency of casing strength under temperature and pressure alternating conditions.

[0025] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description and the drawings. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 A flowchart illustrating a method for predicting the strength of a bushing under alternating temperature and pressure conditions according to an embodiment of the present invention is shown. Figure 2 A schematic diagram of the casing yield strength prediction model according to an embodiment of the present invention is shown; Figure 3 A schematic diagram of the loading path of the load / temperature over time in a thermo-coupling alternating test tension-tension / temperature alternating test according to an embodiment of the present invention is shown. Figure 4 A schematic diagram of the loading path of the load / temperature over time in a thermo-coupled alternating test tension-compression / temperature coupled alternating test according to an embodiment of the present invention is shown. Figure 5 A schematic diagram of strain-time curves (plastic accumulation) in a thermo-coupling alternating test according to an embodiment of the present invention is shown. Figure 6 The following is an illustration of the tension-tension / 25℃~100℃ coupled alternating yield strength variation curve according to an embodiment of the present invention; Figure 7 The tensile-compression / 25℃~100℃ coupled alternating yield strength variation curve according to an embodiment of the present invention is shown; Figure 8 An IPF diagram of the microstructure of the sleeve in its initial state according to an embodiment of the present invention is shown; Figure 9 The IPF diagram of the microstructure of the sleeve after 20 cycles of tension-tension / coupling at 25°C to 100°C according to an embodiment of the present invention is shown. Figure 10 An IPF diagram of the microstructure of the sleeve after 40 cycles of tension-tension / 25°C~100°C coupling alternation according to an embodiment of the present invention is shown. Figure 11 A schematic diagram of the initial state statistics of the sleeve grain size according to an embodiment of the present invention is shown; Figure 12 A statistical schematic diagram is shown below after 20 cycles of tension-tension / 25℃~100℃ coupling alternation of the sleeve grain size according to an embodiment of the present invention. Figure 13 A statistical schematic diagram is shown below after 40 cycles of tension-tension / 25℃~100℃ coupling alternation of the sleeve grain size according to an embodiment of the present invention. Figure 14 A schematic diagram showing the initial state calculation results of the sleeve geometry dislocation density according to an embodiment of the present invention is shown; Figure 15 A schematic diagram of the calculation results after 20 cycles of tension-tension / 25℃~100℃ coupling alternation of the sleeve geometry dislocation density according to an embodiment of the present invention is shown; Figure 16 A schematic diagram of the calculation results after 40 cycles of tension-tension / 25℃~100℃ coupling alternation of the sleeve geometry dislocation density according to an embodiment of the present invention is shown. Figure 17 A comparison chart of the yield strength decay curves of Q125 bushing under different thermo-coupling alternating cycles obtained by the prediction method according to an embodiment of the present invention and experimental data is shown. Figure 18 A schematic diagram of a casing strength prediction system under alternating temperature and pressure conditions according to an embodiment of the present invention is shown. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] It should be noted that the terms "first," "second," etc., used in this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein.

[0030] This invention provides a method and system for predicting the strength of casing under alternating temperature and pressure conditions. It integrates a hybrid model of physical microstructure evolution and deep learning to achieve dynamic yield strength prediction of casing materials.

[0031] like Figure 1 As shown, a method for predicting the strength of a bushing under alternating temperature and pressure conditions includes the following steps: S1. Based on the actual working parameters of the casing, establish a calculation model for the physical parameters of the casing, and calculate the coupled alternating parameters of the casing service temperature and equivalent axial load based on the calculation model.

[0032] The actual operating parameters of the casing include wellbore parameters, fracturing process parameters, and casing structure parameters. Wellbore parameters include wellbore trajectory, wellbore structure, wellhead pressure, wellhead temperature, and formation temperature gradient. Fracturing process parameters include fracturing fluid performance parameters, fracturing-plug-fracturing operation time, and displacement of multi-stage volumetric fracturing process, as well as fracturing temperature T and pressure P. Casing structure parameters include casing size and casing material parameters.

[0033] Among them, the coupled alternating parameters of the casing service temperature and the equivalent axial load include axial load F, axial stress amplitude, axial stress alternating frequency, holding time, temperature amplitude, temperature alternating frequency, axial displacement / strain ε, number of cycles N, and current cycle status (such as peak value, valley value).

[0034] The casing physical parameter calculation model includes the casing axial load calculation model and the wellbore temperature field calculation model during hydraulic fracturing.

[0035] For example, based on the actual operating parameters of the casing, a calculation model for the axial load of the casing during hydraulic fracturing is established, including the following steps: Based on the experimental simulation results, the axial stress of the casing during hydraulic fracturing is determined to be the sum of the compressive stress generated by the buoyancy of the casing in the cementing, the tensile stress caused by the casing's own weight, the axial stress caused by the casing bending, and the additional axial stress generated by the casing during the hydraulic fracturing process, as detailed below: (1) In the formula, σ oe The axial stress is measured in MPa. σ c The compressive stress on the casing caused by the buoyancy of the casing in the cementing cement, in MPa; σ M The tensile stress caused by the casing's own weight, in MPa; σ z The axial stress caused by bending is expressed in MPa. σ H This represents the axial stress caused by hydraulic fracturing, in MPa.

[0036] The compressive stress on the casing caused by the buoyancy of the casing string in the cementing cement is determined based on the specific gravity of the drilling mud, the casing length, the specific gravity of the cementing mud, the cement return depth, and the cross-sectional area of ​​the casing, as detailed below: (2) In the formula, γ 1 represents the specific gravity of drilling mud, in kg / m³. γ 2 represents the specific gravity of the cementing mud, in kg / m³ 3 ; L The length of the sleeve is in meters (m). l For cement back depth, m; s The cross-sectional area of ​​the casing is m. 2 .

[0037] The tensile stress caused by the self-weight of the casing is determined based on the weight per unit length of casing, casing length, depth of the calculation point, and cross-sectional area of ​​the casing, as detailed below: (3) In the formula, q The weight of the sleeve per unit length is in kg; h The depth of the point is calculated in meters (m).

[0038] The axial force caused by casing bending is determined based on the elastic modulus of the casing, the bending curvature of the casing, and the radial coordinate, as follows: (4) In the formula, E Let be the elastic modulus of the casing, in GPa; cLet m be the curvature of the casing. -1 ; r b For radial coordinates, the range of values ​​is... , r 2, r 3 represents the inner and outer diameters of the casing, respectively, in meters (m).

[0039] The axial stress of the casing caused by hydraulic fracturing is determined by calculation using Lamé's formula in elasticity mechanics, including the following steps: Assuming the casing-cement sheath-formation composite model is an axisymmetric plane strain model, the stress field of each layer includes radial stress and circumferential stress, as detailed below: (5) In the formula, σ r , σ φ These are radial stress and circumferential stress, respectively, in MPa; q a The pressure inside the casing is in MPa. q b External pressure of the casing, MPa; r 2 is the inner radius of the casing, in meters; r 3 represents the outer radius of the casing, in meters (m). r Let be the radial coordinate of the casing, in meters (m).

[0040] The expression for radial displacement is obtained from the physical equations under plane strain conditions: (6) In the formula, For radial displacement, μ The Poisson's ratio of the casing. This represents the radial strain of the sleeve.

[0041] Since the radial displacements of the formation, cement sheath, and casing at the first and second cemented surfaces are the same, the radial pressure of the outer layer of the casing can be calculated, and then the radial stress and circumferential stress of the casing can be obtained through formula (5).

[0042] Based on the generalized Hooke's law and the assumptions of the plane strain model, the additional axial stress generated in the casing during hydraulic fracturing can be calculated as follows: (7) During fracturing, temperature changes follow the law of conservation of energy. Taking any unit cell in the study, the energy conservation relationship at a certain moment is: total heat flowing into the unit cell - total heat flowing out of the unit cell + instantaneous power of the heat source inside the unit cell = increase in heat stored inside the unit cell.

[0043] For example, based on the actual operating parameters of the casing, a wellbore temperature field calculation model is established during hydraulic fracturing. The wellbore temperature field calculation model during hydraulic fracturing includes the fluid temperature change equation inside the casing, the heat balance equation of the casing inner wall unit, and the formation temperature change equation.

[0044] Taking fracturing fluid injection into the casing as the research object, assuming the volumetric flow rate is... Q The fracturing fluid flows from top to bottom inside the casing string, at a distance from the wellhead. h Based on the energy conservation equation, an equation for the temperature change of the fluid inside the casing is established, as follows: (8) In the formula: Q The volumetric flow rate of the fluid is m. 3 / h; ρ 0 represents the density of the fracturing fluid, in kg / m³. 3 ; C 0 represents the specific heat capacity of the fracturing fluid, in kJ / (kg·℃). T 0, T 1 represents the temperature inside the casing and the tubing wall, in °C; Let... R 0 =(r 0 -r 1 ) / 2λ 1, λ 1 represents the fluid heat transfer coefficient, with units of kJ / (m·min·℃). f Friction coefficient, dimensionless; v Fluid velocity, m / s; r 0, r 1 represents the inner and outer diameters of the tubing, respectively, and m and τ represent the frictional force on the casing wall caused by viscosity.

[0045] The heat balance equation for the inner wall element of the casing is established as follows: (9) In the formula: ρ 1 represents the density of the fracturing fluid, in kg / m³ 3 ; C 1 represents the specific heat capacity of the fracturing fluid, kJ / (kg·℃); T 2, T 1 represents the temperature of the outer wall and the inside of the casing, in °C. R 1 =(r 2 -r 1 ) / 2λ 2, λ2 is the fluid heat transfer coefficient, in kJ / (m·min·℃); v Fluid velocity, m / s; r 2 is the inner diameter of the casing, in meters (m).

[0046] Assuming only radial heat transfer occurs and no circumferential heat transfer, the formation temperature variation equation is as follows: (10) In the formula: α e =λ / (ρ e C） , =lnr , ρ e The density of the formation is kg / m³. 3 ; C Specific heat of the formation, kJ / (kg·℃); λ is the thermal conductivity coefficient of the formation, kJ / (m·℃); T e Formation temperature, °C; r' Let be the radius (m) of any point on the wellbore from the wellbore axis.

[0047] S2. Using the coupled alternating parameters of the casing service temperature and equivalent axial load as load conditions, conduct a thermo-mechanical coupled alternating test to obtain the casing yield strength test data, including the following steps: Using a high-temperature fatigue testing machine that can simultaneously control the axial load loading program and the temperature loading program, the coupled alternating parameters of the casing service temperature and the equivalent axial load are used as load conditions to carry out the thermo-mechanical coupled alternating test of the casing. After experiencing different thermo-mechanical coupled alternating cycles, a tensile test is immediately carried out to obtain the casing yield strength test data.

[0048] The coupled alternating parameters of the casing service temperature and the equivalent axial load are used as load conditions. Specifically, the maximum axial load fluctuation range and fluctuation frequency, the maximum temperature fluctuation range and fluctuation frequency are selected from the coupled alternating parameters as load conditions.

[0049] S3. Characterize the microstructure of the original casing and the casing after the thermo-coupling alternating test, and obtain experimental data on the microstructure characteristics that affect the casing strength, including the following steps: Microstructure characterization techniques such as SEM-EBSD and TEM were used to characterize the microstructure characteristics of the original casing and the casing after different thermo-mechanical coupling alternations, and experimental data on the microstructure characteristics that affect the strength of the pipe were collected.

[0050] For example, the typical microstructure of commonly used casing materials is tempered sorbite (S... 回 The test data collected included microstructure characteristics that affect the strength of the pipe, such as grain size, high-angle grain boundary ratio, and dislocation density.

[0051] S4. Preprocess the casing material parameters, coupling alternating parameters, casing yield strength test data, and microstructure characteristic test data to obtain training samples, including the following steps: S41. Collect timestamps, coupled alternating parameters, casing yield strength test data, and microstructure characteristic test data to obtain a dataset.

[0052] S42. Perform data preprocessing and feature extraction on the dataset, extracting the yield strength from the casing yield strength test data. σ y The training samples were obtained by correlating the actual operating parameters of the casing at the corresponding time, the microstructure characteristics data, and the casing microstructure characteristics data before the thermo-coupling alternating test. Features, σ y The details are as follows: 1) Perform data cleaning on the dataset: remove outliers, noise, and faulty data from the dataset.

[0053] 2) Data Alignment: Align fracturing temperature T, pressure P, axial load F, and axial displacement / strain according to timestamps. ε Yield strength σ y Synchronized with microstructure characteristics test data.

[0054] 3) Feature extraction The feature data extracted from the dataset includes input features, cumulative damage features, and material features. The input features include axial load alternation amplitude and frequency, axial average stress, the ratio of maximum axial stress to minimum axial stress, and temperature alternation amplitude and frequency.

[0055] Cumulative damage characteristics include plastic strain accumulation, duration of maximum / minimum temperature / axial load, equivalent cycle number (N) of a specific temperature-load coupled alternation, and physics-based damage parameters (such as low-cycle fatigue Coffin-Manson parameter corrections and creep damage parameters).

[0056] Material characteristics include experimental data on casing material parameters and microstructure properties. Among them, casing material parameters include the initial yield strength and chemical composition of the casing.

[0057] 4) Preprocess the extracted feature data to obtain key feature combinations, including thermo-coupling factors, microstructure characteristics test data interaction terms, and asymmetric load coefficients.

[0058] The thermodynamic coupling factor is as follows: Thermodynamic coupling factor = load amp ×temp amp / (frequency+ε) In the formula, load amp Indicates the magnitude of the load. tempamp Indicates the amplitude of temperature. frequency Indicates the coupling alternating frequency. ε =10 -5 This is a small constant used to avoid division by zero.

[0059] The data interaction item for microstructure characteristics test is as follows:

[0060] In the formula, △σ This represents the change in the yield strength of the casing after different cycles of thermo-coupling alternation. ρ For dislocation density, f HAGB For HAGB scores.

[0061] The asymmetric load factor is as follows: Asymmetric load factor = (σ) max -σ min ) / (σ max +ε) In the formula, σ max For the maximum axial stress, σ min For minimum axial stress, ε =10 -5 This is a small constant used to avoid division by zero.

[0062] Temporal feature enhancement is performed on key feature combinations, as detailed below: Calculate the cumulative damage for each cycle: Calculate the stress difference between adjacent time steps: Δ σ = Current axial stress - Previous axial stress

[0063] Calculate the cumulative sum of axial stress changes: calculate the total axial stress change from the beginning to the current moment along the time axis.

[0064] Feature splicing: The microstructure characteristics data of the casing before the thermo-coupling alternating test, the instantaneous axial stress change, and the cumulative axial stress change are spliced ​​along the feature dimension.

[0065] 5) Construct output features.

[0066] The measured yield strength σ in each of the key feature combinations y The training samples were obtained by correlating the actual operating parameters (T, P, N, etc.) of the casing at the corresponding time, experimental data on microstructure characteristics, and microstructure characteristics of the casing before the thermo-coupling alternating test. Features, σy ).

[0067] S5. Construct a casing yield strength prediction model that integrates physical constraints and data-driven approaches, and train the casing yield strength prediction model using training samples.

[0068] S501, a framework for building dynamic prediction models based on TensorFlow / Keras, such as Figure 2 As shown, the prediction model framework includes an LSTM layer, a Dense layer, an Attention layer, a BatchNormalization layer, and a NonNeg layer.

[0069] The LSTM layer is a Long Short-Term Memory network used to process time-series data (such as cyclic load history). The Dense layer is a fully connected layer used for non-linear feature combination. The Attention layer is an attention mechanism that helps the model focus on key time steps (such as high-stress cycles).

[0070] The BatchNormalization layer is a batch normalization layer that accelerates training and improves stability.

[0071] The constraint weights in the NonNeg layer are non-negative to ensure that the outputs (such as lifetime and strength) are physically reasonable.

[0072] S5011. Define a prediction model class, specifically as follows: inherit from tf.Keras.Model, and define a custom Keras model cyclicstrengthpredictor.

[0073] S5012. Initialize the prediction model parameters, as follows: lstm_units: The number of neurons in the LSTM layer (default 64), which controls the ability to extract temporal features.

[0074] dense_units: The number of neurons in the fully connected layer (default 32), used to learn non-linear relationships.

[0075] attention_heads: The number of attention heads (default 4), which enhances the model's ability to identify key loops.

[0076] dropout_rate: Dropout rate (default 0.2) to prevent overfitting.

[0077] S5013. Define the network layers of the prediction model framework, as follows: For example, the LSTM layer uses a dual LSTM structure, where the first LSTM layer is used to extract primary temporal features, and the second LSTM layer is used to deepen the feature representation.

[0078] `return_sequences=True` retains the output of all time steps for use by subsequent Attention layers.

[0079] For example, the BatchNormalization layer accelerates convergence and reduces sensitivity to initialization.

[0080] For example, the Attention layer automatically learns which time steps (such as high-stress cycles) contribute more to fatigue damage.

[0081] use_scale=True: Allows scaling of attention scores, enhancing flexibility.

[0082] For example, Dense layers are used for nonlinear transformations, learning complex relationships through the ReLU activation function.

[0083] For example, Dropout randomly blocks some neurons to improve generalization.

[0084] For example, the output layer can be defined with linear activation: directly predicting the intensity value.

[0085] For example, defining a NonNeg layer ensures that the output value is non-negative (which is consistent with physical meaning).

[0086] For example, define forward propagation, which is the process of computing and passing data layer by layer from the input layer to the output layer.

[0087] For example, the input layer consists of time-series data (such as stress-time curves) with a shape of (batch_size, timesteps, features).

[0088] The key steps of the network layer in the prediction model framework include: LSTM layer processing time series → BatchNormalization layer.

[0089] The Attention layer calculates the weights for each time step and then calculates a weighted average to obtain the global features.

[0090] Dense layers progressively reduce dimensionality, then output predicted values.

[0091] training parameters: control the behavioral differences between Dropout and Batch Normalization during training / inference.

[0092] S502. Embed material constitutive parameters into the casing yield strength prediction model, including: embedding the basic physical constants of the material (e.g., shear modulus G, Burroughs vector b, Taylor factor M, etc.) as non-trainable parameters into the casing yield strength prediction model to ensure that the prediction always conforms to physical laws and avoid non-physical interpretations that may occur with pure data-driven approaches.

[0093] S503. Construct a multi-branch feature extraction architecture in the casing yield strength prediction model. The multi-branch feature extraction architecture includes mechanical load branch, thermal load branch and microstructure and strength branch.

[0094] Among them, the mechanical load branch uses a dual LSTM model with LSTM layers to capture cyclic hardening / softening features, which is used to extract the pressure in the actual working parameters of the casing, as well as the axial stress amplitude, axial stress alternation frequency and holding time in the coupled alternating parameters.

[0095] The thermal load branch (temperature amplitude / frequency) uses an independent LSTM model with the same structure in the LSTM layer to extract the fracturing temperature of the actual casing operating parameters, as well as the temperature amplitude and alternating frequency in the coupled alternating parameters.

[0096] The microstructure and strength branches employ a Dense layer network to process non-temporal parameter microstructures, which is used to extract experimental data on microstructure characteristics (grain size, high-angle grain boundary ratio, and dislocation density, etc.).

[0097] The sigmoid activation function is used to improve gradient flow, and the BatchNormalization layer is used to enhance training stability.

[0098] S504. Define the physical constraint model for the NonNeg layer, including the Hall-Petch grain refinement strengthening model, the Taylor dislocation strengthening model, and the cyclic softening model.

[0099] The Hall-Petch grain refinement and strengthening model is detailed below: σ hp =k / d 1 / 2 In the formula, σ hp This indicates the contribution of grain refinement to yield strength, which is related to grain size. d It is inversely proportional to the square root, k=0.5.

[0100] Among them, Taylor's dislocation reinforcement model: σ taylor =αMGbρ 1 / 2 In the formula, σ taylor The value represents the contribution of dislocation density to the yield strength of the casing. The material constant is α = 0.3, M is the Taylor factor (3.06 for face-centered cubic metal), G is the shear modulus, b is the Burgers vector, and ρ is the dislocation density.

[0101] Among them, the cyclic softening model (load history effect) is used to predict the strength decay of materials under cyclic loading + holding time conditions.

[0102] For example, defining a NonNeg layer divides the input features into three physical fields, including load features, temperature features, and holding time.

[0103] For example, load characteristics: 16 dimensions (stress amplitude, average stress, plastic strain amplitude, number of cycles, etc.); Temperature characteristics: 16 dimensions (current temperature, temperature amplitude, frequency, heat flux density, etc.); Hold-load time: 1 dimension (the time to maintain maximum / minimum load, which is related to creep).

[0104] For example, the physical constraint model for the NonNeg layer is modeled using a softening mechanism, as follows: For the temperature effect, the aging process is simulated, and the Arrhenius-type exponential decay is used to simulate the softening of strength caused by the increase in temperature. For mechanical load effects, the dislocation reorganization caused by the simulated load amplitude is represented by the nonlinear damage accumulation using the square relationship with the plastic strain; For the load holding time effect, time is linearly correlated, simulating diffusion creep.

[0105] For example, the output features of the NonNeg layer are defined as follows: Strength attenuation coefficient = 0.1 × temperature effect × mechanical effect × creep effect.

[0106] S505. Define an Attention layer to fuse material mechanical and thermal features. The input of the Attention layer includes mechanical and thermal features as sequences of two modes, and the output of the Attention layer is the fused features, which contain cross-modal interaction information.

[0107] The initialization of the Attention layer includes: multi-head attention is enabled by default, which can capture multiple interaction modes between different physical fields in parallel, and a scaling factor of the science system is used to adjust the numerical stability of dot product attention.

[0108] S506. Define the physical baseline term and data-driven correction term for the casing yield strength prediction model.

[0109] For example, the physical benchmark term is as follows: σ=σ hp +σ taylor -σ soft In the formula, σIndicates the yield strength of the casing; Hall-Petch term σ hp Indicates grain size strengthening effect; Taylor strengthening term σ taylor Indicates dislocation reinforcement effect; cyclic softening Item σ soft This indicates the decrease in strength caused by fatigue / creep.

[0110] Among them, the data-driven correction term is to use neural networks to build a data-driven correction model, which is used to learn the residual error of the physical model and improve the overall prediction accuracy.

[0111] The input features of the data-driven correction term include: the main features after cross-modal attention fusion (taking the 0th time step), and the material microstructure features (such as grain size, phase composition, etc.).

[0112] The specific process of learning the residual error of the physical model through data-driven correction terms includes: splicing two types of features along the last axis (axis=-1); and predicting the correction amount through the neural network strength_predictor.

[0113] The prediction targets of data-driven correction terms include: residual errors that the physical model failed to capture (such as those caused by complex effects like grain boundary slip and local deformation).

[0114] S507. Define the input layer, output layer, and loss function of the casing yield strength prediction model to obtain a casing yield strength prediction model that integrates physical constraints and data-driven approaches.

[0115] The input layer data of the casing yield strength prediction model includes actual casing operating parameters, coupling alternating parameters, experimental data of microstructure characteristics, and microstructure characteristics data of the casing before the thermo-coupling alternating test.

[0116] The output layer of the casing yield strength prediction model includes the yield strength σ. y The constraint of the output layer is: yield strength σ y It is non-negative, and the correction term can be positive or negative (allowing to compensate for deficiencies in the physical model).

[0117] For example, the loss function adopts a multi-objective loss function, and on the basis of the traditional mean square error (MSE), two types of physical constraints are introduced as regularization terms: the predicted sleeve yield strength is forced to be non-negative, and the model is guided to learn the known cyclic softening model.

[0118] For example, the mean square error of data fitting:

[0119] In the formula, LMSE Mean square error, σ true To determine the yield strength of the bushing after different alternation cycles in a thermo-coupling alternating test, σ pred denoted as the predicted yield strength of the casing by the model, and n is the total number of data points.

[0120] For example, physical constraints stipulate that the yield strength is non-negative and monotonically decreasing:

[0121] In the formula, L non-neg This is a non-negative constraint on yield strength.

[0122]

[0123] In the formula, L mono The increasing cycle number increases the monotonically decreasing constraint of the casing yield strength. E Indicates the expected value. Δ This represents the change in yield strength.

[0124] For example, the multi-objective loss function is as follows:

[0125] In the formula, L total Total constraint, α=0.7.

[0126] S508, the casing yield strength prediction model is used for online learning.

[0127] S5081. Establish a memory buffer (memory_buffer) and train the casing yield strength prediction model through the memory buffer.

[0128] For example, a memory buffer uses a deque to store the latest 1000 sets of observations. The memory buffer is used to prevent catastrophic forgetting and to preserve the distribution characteristics of historical data. The memory buffer prioritizes the replay of key samples.

[0129] S5082. When the number of samples in the memory buffer exceeds the set value, the samples in the memory buffer are updated.

[0130] For example, when the memory buffer accumulates ≥100 samples, 32 groups are randomly sampled for training. During random sampling, key samples are limited to prevent important samples from being lost during simple random sampling.

[0131] S509. After completing the online learning, perform uncertainty quantification on the casing yield strength prediction model.

[0132] S5091. The casing yield strength prediction model is modified to be probabilistic, so that the prediction model outputs both the mean and variance, and the casing yield strength output by the prediction model follows a normal distribution N(μ, σ). 2 ).

[0133] S5092. Define the variance of the output value of the prediction model to be greater than 0, as follows: σ 2 =exp(f θ (x)）＞0 In the formula, f θ (x) represents the output value of the casing yield strength prediction model, and exp represents the exponential function, which has better numerical stability in backpropagation.

[0134] S510. After completing the uncertainty quantification, the casing yield strength prediction model is trained using training samples.

[0135] S5101. Normalize the input and output features in the training samples, and divide the normalized training samples into training set, validation set and test set according to the proportion.

[0136] Normalizing input and output features accelerates model training and improves performance.

[0137] For example, the training samples can be randomly divided into training, validation, and test sets according to a ratio (e.g., 70:15:15) to ensure that different working conditions are represented in the dataset.

[0138] S5102. The casing yield strength prediction model is trained in stages using the training set, including: in the first stage, the data-driven layer is frozen for physical term pre-training; in the second stage, during joint fine-tuning, a dynamic weight adjustment strategy is used to adjust the weights of the physical constraint term and the data fitting term.

[0139] Initial stage: Focus on physical constraints (alpha=0.5) to ensure the model converges quickly to a physically reasonable solution space; Later stage: Focus on data fitting (alpha→0.9) to optimize prediction accuracy. The principle of parameter selection for joint fine-tuning is shown in Table 1.

[0140] Table 1

[0141] S5102. Adjust the hyperparameters of the casing yield strength prediction model using the validation set.

[0142] The hyperparameters tuned include the number of neurons in the LSTM layer (range 32-64), microstructure-sensitive batch processing (range 16-64), temperature-dependent learning rate (range 0.5-2.0), initial weights of the physics term (range 0.1-0.9), and the temperature reference term loss_T_ref in the loss function (range 293~473). The optimal combination is selected by evaluating the performance of the validation set through random search combined with 5-fold cross-validation.

[0143] S5103. Evaluate the performance of the casing yield strength prediction model using the test set. If the performance indicators of the casing yield strength prediction model meet the requirements, the training is complete.

[0144] The performance indicators include relative error of yield strength, trend consistency index, extreme condition prediction deviation and physical constraint violation rate. The calculation formulas and thresholds for the performance indicators are shown in Table 2.

[0145] Table 2

[0146] The casing yield strength prediction model of the present invention has the following advantages: 1. Multiscale coupling: cross-scale prediction from dislocation motion (μm scale) to macroscopic yield strength (mm scale).

[0147] 2. Load history awareness: LSTM layers can capture the cumulative damage effect of cyclic loads.

[0148] 3. Physical interpretability: Each physical module outputs intermediate variables for engineers to analyze.

[0149] 4. Online adaptability: It can integrate new experimental data in real time to update model parameters.

[0150] S6. Input the real-time acquired actual working condition parameters of the casing and the experimental data of microstructure characteristics related to the casing strength into the trained casing yield strength prediction model, and output the yield strength of the casing.

[0151] This invention utilizes a physical model to calculate the temperature field and axial equivalent stress distribution of the casing string under multi-stage volumetric fracturing. The maximum axial load fluctuation range and frequency, as well as the maximum temperature fluctuation range and frequency, are selected as the load conditions for the indoor thermo-mechanical coupling alternating casing strength evaluation test. The yield strength value of the casing after different thermo-mechanical coupling alternating cycles is obtained. At the same time, EBSD is used to collect the microstructural evolution characteristics of the casing material after different thermo-mechanical coupling alternating cycles. A multi-scale physical-data coupling model is established using a long short-term memory recurrent neural network (LSTM) + fully connected layer (Dense) + attention mechanism for dynamic prediction of the yield strength of the casing material under multi-stage volumetric fracturing conditions.

[0152] For example, the prediction method of this invention is used to predict the yield strength of Q125 casing material under multi-stage volumetric fracturing conditions in shale gas development of an oilfield. The specific process is as follows: Step 1: Based on the casing physical parameter calculation model, calculate the coupled alternating parameters of casing service temperature and equivalent axial load. This yields the tensile load fluctuation range of the casing during single-stage fracturing in a multi-stage volumetric fracturing process: 85%~75% of yield strength, frequency 1Hz; coupled temperature load fluctuation range: 25℃~100℃, frequency once every 30 minutes; single-stage fracturing load holding time: 4 hours; unloading to 0MPa. The number of fracturing stages is 40, which means 40 large-scale thermo-mechanical coupled alternating cycles.

[0153] Step 2: Conduct an indoor thermo-mechanical coupling alternating test based on the above test conditions. The test loading path is as follows: Figure 3 and Figure 4 As shown, Figure 3 The load / temperature loading path over time in a tension-tension / temperature alternating test is shown. Figure 4 The load / temperature loading path over time in the tension-compression / temperature coupled alternating test is shown, and the strain-time curves during the test are as follows: Figure 5 As shown, the strain fluctuation pattern is consistent with the temperature fluctuation frequency.

[0154] Tensile tests and microstructure characterization tests were conducted at the initial stage, after 20 cycles, and after 40 cycles, respectively, to obtain the corresponding yield strength and microstructure characteristics test data. Figure 6 and Figure 7 The strength variation curves of the casing material after different cycles of thermo-mechanical coupling are shown. Figure 6 The yield strength variation curves of three groups of samples under tension-tension / 25℃~100℃ coupled alternating stress are shown. Figure 7 The yield strength variation curves of three groups of samples under tension-compression / coupling alternating temperature from 25℃ to 100℃ are shown.

[0155] Figures 8-10 The image shows the Inverted Pole Figure (IPF) plots illustrating the microstructure evolution of the sleeve after different thermo-coupling alternating cycles. IPF plots are graphical representations used in materials science to characterize crystal orientation distribution, primarily for displaying grain orientation differences and texture analysis. Figure 8 An IPF diagram of the microstructure of the sleeve in its initial state according to an embodiment of the present invention is shown; Figure 9 The IPF diagram of the microstructure of the sleeve after 20 cycles of tension-tension / coupling at 25°C to 100°C according to an embodiment of the present invention is shown. Figure 10 The IPF diagram of the microstructure of the sleeve after 40 cycles of tension-tension / coupling at 25°C to 100°C according to an embodiment of the present invention is shown.

[0156] Figures 11-13 The following is a statistical chart of the grain size of the sleeve material after different cycles of thermo-mechanical coupling. Figure 11 A schematic diagram of the initial state statistics of the sleeve grain size according to an embodiment of the present invention is shown; Figure 12 A statistical schematic diagram is shown below after 20 cycles of tension-tension / 25℃~100℃ coupling alternation of the sleeve grain size according to an embodiment of the present invention. Figure 13 A statistical schematic diagram is shown below after 40 cycles of tension-tension / 25°C to 100°C coupling alternation of the sleeve grain size according to an embodiment of the present invention.

[0157] Figures 14-16 The diagram shows the calculation of the geometric dislocation density of the casing after different thermo-coupling alternation cycles. Figure 14 A schematic diagram showing the initial state calculation results of the sleeve geometry dislocation density according to an embodiment of the present invention is shown; Figure 15 A schematic diagram of the calculation results after 20 cycles of tension-tension / 25℃~100℃ coupling alternation of the sleeve geometry dislocation density according to an embodiment of the present invention is shown; Figure 16 A schematic diagram of the calculation results after 40 cycles of tension-tension / 25℃~100℃ coupling alternation of the sleeve geometry dislocation density according to an embodiment of the present invention is shown.

[0158] Step 3: According to Figure 1 Construct a casing strength prediction model that integrates multi-scale physical constraints and data-driven thermo-mechanical coupling cycles.

[0159] Step 4: Implementation of the physical constraint layer, integrating the classic strength model Hall-patch relationship, Taylor reinforcement model and cyclic softening model.

[0160] Step 5: Implement the recurrent softening model. A three-layer fully connected neural network is defined. The first hidden layer performs primary feature abstraction, inputting a 32-dimensional space and activating it with the sigmoid function. The second hidden layer performs advanced feature combination, compressing it to a 16-dimensional feature space while maintaining the sigmoid function for activation. The output layer performs scalar prediction output, which is finally mapped to a 1-dimensional output. ReLU activation ensures non-negativity. The physical constraint is that the material softening amount is always ≥0, which conforms to the second law of thermodynamics.

[0161] Step 6: Model Training. Data acquisition: Conduct 50 sets of experiments on the same grade of Q125 steel as described above, and supplement with 150 sets of finite element simulations, for a total of 200 sets as the dataset. The dataset is divided into training set, validation set, and test set according to a ratio of 60%, 20%, and 20%, respectively.

[0162] Step 7: Data preprocessing, data cleaning to remove outlier data points, data alignment, and filling in missing items using interpolation.

[0163] Step 8: Dataset pipeline construction. Establish a multi-branch dataset pipeline and classify features into four categories: mechanical load, thermal load, microstructure, and thermo-coupling factor.

[0164] Step 9: Compile the hybrid loss function.

[0165] Step 10: Optimizer configuration. Create the AdamW optimizer, set the learning rate to 0.0003, and the weight decay factor to 1×10⁻⁶. -5 Perform L2 regularization.

[0166] Step 11: Evaluation metrics, which include mean absolute error (MAE), mean absolute percentage error (MAPE), and a custom physical consistency metric.

[0167] Step 10: The model training adopts a phased training strategy. Phase 1 freezes the physical layer and trains only the data-driven part. Phase 2 performs joint fine-tuning.

[0168] Step 11: Prediction model validation, including yield strength relative error validation and extreme condition validation (maximum amplitude of axial load / temperature load across the entire well section, 40 cycles). The calculation formulas and thresholds for the performance indicators used in the model are shown in Table 3.

[0169] Table 3

[0170] Step 12: Online learning updates.

[0171] Step 13: Input the actual working parameters of the casing and the experimental data of the microstructure characteristics related to the casing strength into the trained casing yield strength prediction model, and output the yield strength of Q125 casing.

[0172] Figure 17 The diagram shows a comparison between the yield strength decay curves of Q125 tubing under different thermo-coupling alternating cycles obtained using the prediction method of this invention and the experimental data. It can be seen that the predicted values ​​are quite close to the experimental data, which shows that the prediction method of this application has a high accuracy.

[0173] Based on the above-mentioned methods for improving bushing strength under alternating temperature and pressure conditions, such as Figure 18 As shown, this embodiment of the invention also provides a prediction system for the strength of the casing under alternating temperature and pressure conditions, including a calculation module, a yield test device, a first data processing module, a second data processing module, a model training module, and a prediction module.

[0174] The calculation module is used to establish a calculation model of the casing's physical parameters based on the actual operating parameters of the casing, and to calculate the coupled alternating parameters of the casing's service temperature and equivalent axial load based on the calculation model of the casing's physical parameters.

[0175] The yield test apparatus is used to conduct thermo-mechanical coupled alternating tests using the coupled alternating parameters of the casing service temperature and equivalent axial load as load conditions to obtain test data on the casing yield strength.

[0176] The first data processing module is used to characterize the microstructure of the original casing and the casing after the thermo-coupling alternating test, and to obtain experimental data on the microstructure characteristics that affect the casing strength.

[0177] The second data processing module is used to preprocess the casing material parameters, coupling alternating parameters, casing yield strength test data, and microstructure characteristic test data to obtain training samples.

[0178] The model training module is used to build a casing yield strength prediction model that integrates physical constraints and data-driven approaches, and to train the casing yield strength prediction model using training samples.

[0179] The prediction module is used to input the real-time acquired actual working condition parameters of the casing and the experimental data of microstructure characteristics related to the casing strength into the trained casing yield strength prediction model, and output the yield strength of the casing.

[0180] This invention uses coupled alternating parameter load as the load condition for thermo-mechanical coupled alternating test to obtain experimental data on microstructure characteristics. The casing material parameters, coupled alternating parameters, casing yield strength test data, and microstructure characteristic test data are used as sample data. The sample data has load, temperature, and microstructure evolution characteristics, thereby enabling the casing yield strength prediction model trained from the sample data to accurately predict the yield strength of the casing.

[0181] This invention is based on a multi-scale physical constraint-data-driven fusion model to obtain a casing yield strength prediction model. It adopts a multi-branch feature extraction architecture to extract load, temperature and microstructure evolution features respectively, thereby improving the model prediction accuracy and providing technical support for the design and safety evaluation of unconventional well casing materials.

[0182] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the strength of a bushing under alternating temperature and pressure conditions, characterized in that, Includes the following steps: Based on the actual working parameters of the casing, a calculation model for the physical parameters of the casing is established, and based on the calculation model for the physical parameters of the casing, the coupled alternating parameters of the casing service temperature and the equivalent axial load are calculated. The coupled alternating parameters of the casing service temperature and the equivalent axial load are used as load conditions to conduct a thermo-mechanical coupled alternating test to obtain the casing yield strength test data. The microstructure characteristics of the original casing and the casing after thermo-coupling alternating test were characterized to obtain experimental data on the microstructure properties that affect the casing strength. Preprocessing of casing material parameters, coupling alternating parameters, casing yield strength test data, and microstructure experimental data is performed to obtain training samples. A casing yield strength prediction model integrating physical constraints and data-driven approaches was constructed, and the model was trained using training samples. The actual working parameters of the casing, which are acquired in real time, and the experimental data of the microstructure characteristics related to the casing strength are input into the trained casing yield strength prediction model, and the yield strength of the casing is output.

2. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to claim 1, characterized in that, The casing physical parameter calculation model includes the casing axial load calculation model and the wellbore temperature field calculation model during hydraulic fracturing.

3. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to claim 2, characterized in that, Based on the actual operating parameters of the casing, a calculation model for the axial load of the casing during hydraulic fracturing is established, including the following steps: Based on the experimental simulation results, the axial stress of the casing during hydraulic fracturing is determined to be the sum of the compressive stress generated by the buoyancy of the casing in the cementing cement, the tensile stress caused by the casing's own weight, the axial stress caused by the casing bending, and the additional axial stress generated by the casing during the hydraulic fracturing process.

4. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to claim 2, characterized in that, The calculation model for the wellbore temperature field during hydraulic fracturing includes equations for the temperature change of the fluid inside the casing, thermal balance equations for the inner wall units of the casing, and equations for the temperature change of the formation.

5. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to claim 1, characterized in that, The coupled alternating parameters include axial load, axial stress amplitude, axial stress alternating frequency, holding time, temperature amplitude, temperature alternating frequency, axial displacement / strain, number of cycles, and current cycle status.

6. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to claim 5, characterized in that, Using the coupled alternating parameters of the casing service temperature and equivalent axial load as load conditions, a thermo-mechanical coupled alternating test is conducted to obtain the casing yield strength test data, including the following steps: The coupled alternating parameters of the casing service temperature and equivalent axial load are used as load conditions, including: selecting the maximum axial load fluctuation range and fluctuation frequency, and the maximum temperature fluctuation range and fluctuation frequency among the coupled alternating parameters as load conditions. A thermo-mechanical coupling alternating test was conducted on the casing. After experiencing different thermo-mechanical coupling alternating cycles, a tensile test was immediately performed to obtain the casing yield strength test data.

7. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to claim 1, characterized in that, Microstructure characteristics experimental data include grain size, high-angle grain boundary ratio, and dislocation density.

8. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to claim 1, characterized in that, Preprocessing of casing material parameters, coupling alternating parameters, casing yield strength test data, and microstructure characteristic test data to obtain training samples includes the following steps: Collect timestamps, coupled alternating parameters, casing yield strength test data, and microstructure characteristic test data to obtain a dataset; Data preprocessing and feature extraction were performed on the dataset. The yield strength in the casing yield strength test data was correlated with the actual working condition parameters of the casing at the corresponding time, the microstructure characteristics data, and the microstructure characteristics data of the casing before the thermo-coupling alternating test to obtain training samples.

9. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to any one of claims 1-8, characterized in that, Constructing a casing yield strength prediction model that integrates physical constraints and data-driven approaches includes the following steps: A dynamic prediction model framework is constructed, which includes an LSTM layer, a Dense layer, an Attention layer, a BatchNormalization layer, and a NonNeg layer. Embed material constitutive parameters in the casing yield strength prediction model; A multi-branch feature extraction architecture is constructed in the casing yield strength prediction model. The multi-branch feature extraction architecture includes mechanical load branch, thermal load branch and microstructure and strength branch. Define the physical constraint models for the NonNeg layer, including the Hall-Petch grain refinement strengthening model, the Taylor dislocation strengthening model, and the cyclic softening model; Define an Attention layer to fuse material mechanical and thermal characteristics; Define the physical baseline term and data-driven correction term for the casing yield strength prediction model; Define the input layer, output layer, and loss function of the casing yield strength prediction model to obtain a casing yield strength prediction model that integrates physical constraints and data-driven approaches.

10. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to claim 9, characterized in that, The mechanical load branch is used to extract the pressure from the actual working parameters of the casing, as well as the axial stress amplitude, alternation frequency, and holding time from the coupled alternating parameters; The thermal load branch is used to extract the fracturing temperature of the actual casing operating parameters, as well as the temperature amplitude and temperature alternation frequency in the coupled alternating parameters; Microstructure and strength branches are used to extract experimental data on microstructure properties.

11. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to claim 9, characterized in that, The casing yield strength prediction model is trained using training samples, including the following steps: The input and output features in the training samples are normalized, and the normalized training samples are divided into training set, validation set and test set according to the proportion. The casing yield strength prediction model is trained in stages using the training set, including: the first stage freezes the data-driven layer for physical term pre-training; the second stage, during joint fine-tuning, adopts a dynamic weight adjustment strategy to adjust the weights of the physical constraint terms and the data fitting terms. The hyperparameters of the casing yield strength prediction model were adjusted using the validation set; The performance of the casing yield strength prediction model is evaluated using a test set. If the performance indicators of the casing yield strength prediction model meet the requirements, the training is complete.

12. The method for predicting the strength of a bushing under alternating temperature and pressure conditions according to claim 8, characterized in that, Data preprocessing and feature extraction of the dataset include the following steps: Remove outliers, noise, and faulty data from the dataset; The test data of fracturing temperature, pressure, axial load, axial displacement / strain, yield strength and microstructure characteristics are synchronized according to the timestamp; The feature data extracted from the dataset includes input features, cumulative damage features, and material features; The extracted feature data is preprocessed to obtain key feature combinations, including thermo-coupling factors, microstructure characteristics test data interaction terms, and asymmetric loading coefficients.

13. A system for predicting the strength of a bushing under alternating temperature and pressure conditions, characterized in that, include: The calculation module is used to establish a calculation model of the casing's physical parameters based on the actual working parameters of the casing, and to calculate the coupled alternating parameters of the casing's service temperature and equivalent axial load based on the calculation model of the casing's physical parameters. The yield test apparatus is used to conduct thermo-mechanical coupled alternating tests using the coupled alternating parameters of the casing service temperature and equivalent axial load as load conditions to obtain test data of the casing yield strength. The first data processing module is used to characterize the microstructure of the original casing and the casing after the thermo-coupling alternating test, and to obtain experimental data on the microstructure characteristics that affect the casing strength. The second data processing module is used to preprocess the casing material parameters, coupling alternating parameters, casing yield strength test data and microstructure characteristic test data to obtain training samples. The model training module is used to build a casing yield strength prediction model that integrates physical constraints and data-driven approaches, and to train the casing yield strength prediction model using training samples. The prediction module is used to input the real-time acquired actual working condition parameters of the casing and the experimental data of microstructure characteristics related to the casing strength into the trained casing yield strength prediction model, and output the yield strength of the casing.

14. The bushing strength prediction system under alternating temperature and pressure conditions according to claim 13, characterized in that, The yield test apparatus is specifically used for: The coupled alternating parameters of the casing service temperature and equivalent axial load are used as load conditions, including: selecting the maximum axial load fluctuation range and fluctuation frequency, and the maximum temperature fluctuation range and fluctuation frequency among the coupled alternating parameters as load conditions. A thermo-mechanical coupling alternating test was conducted on the casing. After experiencing different thermo-mechanical coupling alternating cycles, a tensile test was immediately performed to obtain the casing yield strength test data.

15. The bushing strength prediction system under alternating temperature and pressure conditions according to claim 13, characterized in that, The model training module is specifically used for: A dynamic prediction model framework is constructed, which includes an LSTM layer, a Dense layer, an Attention layer, a BatchNormalization layer, and a NonNeg layer. Embed material constitutive parameters in the casing yield strength prediction model; A multi-branch feature extraction architecture is constructed in the casing yield strength prediction model. The multi-branch feature extraction architecture includes mechanical load branch, thermal load branch and microstructure branch. Define the physical constraint models for the NonNeg layer, including the Hall-Petch grain refinement strengthening model, the Taylor dislocation strengthening model, and the cyclic softening model; Define an Attention layer to fuse material mechanical and thermal characteristics; Define the physical baseline term and data-driven correction term for the casing yield strength prediction model; Define the input layer, output layer, and loss function of the casing yield strength prediction model to obtain a casing yield strength prediction model that integrates physical constraints and data-driven approaches.

Images