Well wall stability model construction prediction method based on easy-to-collapse layer instability mechanism, terminal equipment and storage medium
By using a wellbore stability model based on the instability mechanism of easily collapsible layers, combined with SVR and neural networks, wellbore stability can be monitored in real time, solving the problems of insufficient accuracy and poor adaptability in existing technologies, and improving the safety and efficiency of drilling operations.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2026-03-31
Smart Images
Figure CN121765495A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of drilling engineering technology, and is a method, terminal equipment and storage medium for constructing and predicting wellbore stability models based on the instability mechanism of easily collapsing layers. Background Technology
[0002] Wellbore prediction primarily involves the prediction and analysis of wellbore stability, a crucial aspect of drilling engineering. Wellbore instability is a common problem during drilling, potentially leading to drilling operations interruption or even safety accidents. Therefore, accurate wellbore prediction is essential for ensuring the safety and efficiency of drilling operations.
[0003] Wellbore stability prediction technology involves multiple aspects, including the measurement of in-situ stress, the mechanical properties of rock materials, the stress state of the surrounding rock, wellbore collapse and fracturing, plastic shrinkage, and viscoelastic deformation. These technologies aim to predict wellbore stability by analyzing factors such as the mechanical properties of the rock, the distribution of in-situ stress, and the influence of drilling fluid on wellbore stability.
[0004] Existing technologies rely heavily on empirical formulas, statistical analysis, or simple physical models, which have limited accuracy when dealing with complex geological conditions and nonlinear relationships. Furthermore, they are often based on fixed parameters and assumptions, resulting in poor adaptability to different geological strata. They only consider a few directly influencing factors, ignoring the combined effects of numerous potential variables, making real-time monitoring and dynamic adjustment of prediction models difficult, leading to poor adaptability and crude risk assessment. In addition, traditional methods are ineffective at handling complex nonlinear problems such as wellbore deformation. Summary of the Invention
[0005] This invention provides a wellbore stability model construction and prediction method, terminal equipment and storage medium based on the instability mechanism of easily collapsible layers, which overcomes the shortcomings of the above-mentioned prior art and can effectively solve the problems of insufficient prediction accuracy, poor adaptability and rough risk assessment of traditional methods.
[0006] One of the technical solutions of this invention is achieved through the following measures: a wellbore stability model construction and prediction method based on the instability mechanism of easily collapsing layers, comprising the following steps:
[0007] S1. Equipment arrangement: Based on the downhole conditions, select a fiber optic endoscope with high temperature and high pressure resistance and corrosion resistance. Based on the drilling depth and the well wall area to be observed, determine the length, diameter and lens type of the fiber optic endoscope. Integrate the fiber optic endoscope with the drill string to ensure that it can be used along the drill pipe or the measurement while drilling system during the drilling process or specific operation stages.
[0008] S2. Data Collection and Preprocessing: Collect drilling data, geological data, and wellbore image data acquired by fiber optic endoscope, and preprocess the collected data to remove outliers and standardize the numerical range; the drilling data includes drilling pressure, rotation speed, and drilling fluid properties, the geological data includes rock strata characteristics and geostress distribution, and the wellbore image data includes fracture width and spacing.
[0009] S3. Structural Features: Extract key features from the preprocessed image data in step S2, and normalize all features based on drilling parameters and geological information; divide the data into training and testing sets.
[0010] S4. Model Construction: Support Vector Regression (SVR) is used to predict the degree of wellbore deformation, and a neural network is used to evaluate the stability level. The training set data is used to fit the model parameters, and cross-validation is used to optimize the model performance.
[0011] S5. Model Validation: Test the model on the validation set and evaluate the prediction accuracy;
[0012] S6. Real-time monitoring and prediction: Input the real-time acquired downhole images and drilling parameters into the trained model to predict the stability of the wellbore in real time. Combine the prediction results with preset thresholds to activate the early warning system and adjust the drilling strategy in a timely manner.
[0013] S7. Feedback and Iteration: New data collected during actual drilling is continuously fed back into the model, continuously updating and optimizing the prediction model.
[0014] The following are further optimizations and / or improvements to one of the above-mentioned technical solutions:
[0015] In step S2 above, based on drilling data and wellbore image data, features highly correlated with wellbore stability can be selected to perform data cleaning.
[0016] In step S3 above, when normalizing all features, Min-Max scaling can be used to map the values of crack width and spacing features to the range [0,1]. Through linear transformation of the data, the minimum value of the data is 0 and the maximum value is 1; the specific formula is as follows:
[0017]
[0018] In equation (2), X represents the original data. min and X max These are the minimum and maximum values of the data, respectively.
[0019] In step S4 above, the RBF kernel function is selected in the support vector regression (SVR) prediction of wellbore deformation. Its mathematical expression can be:
[0020] K(x,y)=exp(-γ||xy|| 2 (3)
[0021] In equation (3), x and y are the input data points, and γ is the width parameter of the kernel function;
[0022] The RBF kernel function maps the original feature space to a higher-dimensional space, maximizing the distance from the hyperplane to each class's data points. The hyperplane is defined by the normal vector ω and the bias b, i.e.:
[0023] ω·x+b=0 (4)
[0024] The optimal hyperplane is obtained by determining the optimal support vector through a classification decision function. Specifically:
[0025]
[0026] In equations (4) and (5), x is the feature vector of the data point, y is the category label of the data point, and α i It is the optimal Lagrange multiplier;
[0027] The optimal hyperplane is obtained through calculation, and all data points satisfying (ω·x+b)≥1 are used to predict the degree of deformation of the wellbore.
[0028] The deformation degree predicted by SVR can be used as an input feature of the neural network. The original crack width and spacing features are also added to the input of the neural network. After normalizing the input features, the neural network model is constructed. Its specific expression is as follows:
[0029]
[0030] In equation (6), a represents the output of the neuron, g(·) represents the activation function, and w i Let x represent the weight of the i-th input signal. i Let x represent the i-th input signal and b represent the bias; the neuron receives multiple input signals x. i Each input signal is multiplied by its corresponding weight w. i They are weighted and a bias b is added, and their weighted sum is fed into the activation function g(·) to produce the neuron's output a;
[0031] The ReLU activation function is used to change all input values less than 0 to 0, while keeping values greater than 0 unchanged. The expression for the ReLU function is:
[0032] g(z) = max(0, z) (7)
[0033] Map the input of neuron z to a specific output value g(z), and calculate the output of neuron a = g(z).
[0034] When using cross-validation to evaluate model performance, the specific steps include:
[0035] We choose k-fold cross-validation, which divides the dataset into k parts. Each time, we use k-1 parts as the training set and the remaining part as the validation set. We repeat this process k times to obtain k validation results.
[0036] Calculate the average of the model performance metrics in k validations, and use it as the evaluation metric for the neural network model;
[0037] Compare the average performance metrics of all hyperparameter configurations, select the configuration with the highest average performance as the optimal model parameter setting, retrain the model using this parameter setting, and evaluate its performance on the test set.
[0038] In step S5 above, model validation may specifically include the following steps:
[0039] Assuming we are currently considering the j-th neuron in the l-th layer, calculate the error of the output layer:
[0040]
[0041] In equation (8), J is the loss function. It is the input of the j-th neuron in the l-th layer;
[0042] Recursively calculate the error of the previous layer:
[0043]
[0044] In equation (9), n l It is the number of neurons in the i-th layer. is the weight connecting the j-th neuron in layer (l-1) and the k-th neuron in layer l, and g′(·) is the derivative of the activation function;
[0045] Calculate the gradients of the weights and biases:
[0046]
[0047] In equation (10), It is the output of the i-th neuron in the (l-1)-th layer;
[0048] Update weights and biases:
[0049]
[0050] In equations (12) and (13), α is the learning rate. The formulas are used repeatedly until the convergence condition is met.
[0051] The loss function J mentioned above can be expressed as mean squared error (MSE), where the expression for mean squared error (MSE) is:
[0052]
[0053] In equation (14), n is the sample size, h i For the true value, These are predicted values.
[0054] The second technical solution of the present invention is achieved through the following measures: a terminal device, including a memory and a processor, wherein the memory stores a program that can run on the processor, and the processor executes the program to implement the above-mentioned wellbore stability model construction and prediction method based on the unstable mechanism of easily collapsing layers.
[0055] The third technical solution of the present invention is achieved through the following measures: a storage medium storing one or more programs, which can be executed by one or more processors to realize the above-mentioned wellbore stability model construction and prediction method based on the unstable mechanism of easily collapsing layers.
[0056] This invention combines the powerful predictive capabilities of SVR (Self-Range Dynamics) with the pattern recognition capabilities of neural networks, forming an integrated "prediction + evaluation" method that improves the overall system's predictive performance and the reliability of stability assessment. SVR effectively maps to a high-dimensional space using kernel tricks, capturing the nonlinear relationship between input features and wellbore deformation, thus improving the accuracy and adaptability of the prediction model. SVR effectively handles high-dimensional feature spaces and does not fail due to the curse of dimensionality, allowing for a more comprehensive consideration of all relevant factors. The neural network, acting as a classifier, can subdivide wellbore stability into multiple levels based on the deformation degree predicted by SVR and other features, providing a more detailed risk assessment and facilitating early preventative measures. This invention allows the model to continuously learn and optimize with the arrival of new data, improving the model's timeliness and accuracy. The combination of SVR and neural networks significantly improves the safety and efficiency of drilling operations, solving problems such as insufficient prediction accuracy, poor adaptability, and coarse risk assessment in traditional methods. This invention not only increases the measurement frequency, enabling continuous testing of wellbore fluid levels, but also provides feasibility for real-time monitoring of fluid level changes. Furthermore, it allows for parallel operation with drilling, greatly improving wellbore fluid level safety and effectively enhancing work efficiency. Attached Figure Description
[0057] Appendix Figure 1 This is a flowchart illustrating the wellbore stability model construction and prediction method based on the instability mechanism of easily collapsing layers in an embodiment of the present invention. Detailed Implementation
[0058] The present invention is not limited to the following embodiments, and the specific implementation can be determined according to the technical solution of the present invention and the actual situation.
[0059] The present invention will be further described below with reference to embodiments:
[0060] Example 1: As shown in the attached document Figure 1 As shown, the wellbore stability model construction and prediction method based on the instability mechanism of easily collapsing layers includes the following steps:
[0061] S1. Equipment Setup: Based on downhole conditions, select a fiber optic endoscope with high temperature and pressure resistance and corrosion resistance. Determine the length, diameter, and lens type of the fiber optic endoscope according to the drilling depth and the expected wellbore area to be observed. Integrate the fiber optic endoscope with the drill string to ensure that it can move along the drill pipe or the measurement-while-drilling system during drilling or specific operation stages. In this embodiment, precise control is required during deployment to avoid damaging the fiber optic cable.
[0062] S2. Data Collection and Preprocessing: Collect drilling data, geological data, and wellbore image data acquired by fiber optic endoscope, and preprocess the collected data to remove outliers and standardize the numerical range; the drilling data includes drilling pressure, rotation speed, and drilling fluid properties, the geological data includes rock strata characteristics and geostress distribution, and the wellbore image data includes fracture width and spacing.
[0063] S3. Structural Features: Key features are extracted from the preprocessed image data in step S2. Based on drilling parameters and geological information, structural features are constructed, and all features are normalized. The dataset is divided into training and testing sets. In this embodiment, the dataset can be divided into a training set and a testing set, with feature X representing fracture width and spacing and target variable Y representing wellbore stability index.
[0064] S4. Model Construction: Support Vector Regression (SVR) is used to predict the degree of wellbore deformation, and a neural network is used to evaluate the stability level. The training set data is used to fit the model parameters, and cross-validation is used to optimize the model performance.
[0065] S5. Model Validation: Test the model on the validation set and evaluate the prediction accuracy;
[0066] S6. Real-time monitoring and prediction: Input the real-time acquired downhole images and drilling parameters into the trained model to predict the stability of the wellbore in real time. Combine the prediction results with preset thresholds to activate the early warning system and adjust the drilling strategy in a timely manner, such as increasing the drilling fluid density and reducing the drilling speed to prevent wellbore instability.
[0067] S7. Feedback and Iteration: New data collected during actual drilling is continuously fed back into the model to continuously update and optimize the prediction model, thereby ensuring its accuracy and practicality.
[0068] In step S2, based on drilling data and wellbore image data, features highly correlated with wellbore stability are selected to perform data cleaning. In this embodiment, specifically, the isnull() or notnull() functions can be used to mark the location of missing values, and the missing values can be filled by linear interpolation.
[0069] The specific steps of isnull() are as follows: If (C ij If the expression is NaN, the result is (T)(True); otherwise, it is (F)(False).
[0070] The specific steps of notnull() are as follows:
[0071] If (C) ij If the expression is not NaN, the result is (T)(True); otherwise, it is (F)(False).
[0072] The calculation formula for linear interpolation is as follows:
[0073]
[0074] In equation (1), (x1, y1) and (x2, y2) are known data points, x is the x-coordinate of the point to be estimated, and x is the estimated y-coordinate.
[0075] In step S3, when normalizing all features, Min-Max scaling is used to map the values of crack width and spacing features to the range [0,1]. Through linear transformation of the data, the minimum value of the data is 0 and the maximum value is 1; the specific formula is:
[0076]
[0077] In equation (2), X represents the original data. min and X max These are the minimum and maximum values of the data, respectively.
[0078] In step S4, the RBF kernel function is selected in the support vector regression (SVR) prediction of wellbore deformation. Its mathematical expression is:
[0079] K(x,y)=exp(-γ||xy|| 2 (3)
[0080] In equation (3), x and y are the input data points, and γ is the width parameter of the kernel function;
[0081] The RBF kernel function maps the original feature space to a higher-dimensional space, maximizing the distance from the hyperplane to each class of data points. In this process, the hyperplane is defined by the normal vector ω and the bias b, i.e.:
[0082] ω·x+b=0 (4)
[0083] The optimal hyperplane is obtained by determining the optimal support vector through a classification decision function. Specifically:
[0084]
[0085] In equations (4) and (5), x is the feature vector of the data point, y is the category label of the data point, and α i It is the optimal Lagrange multiplier;
[0086] The optimal hyperplane is obtained through calculation, and all data points satisfying (ω·x+b)≥1 are used to predict the degree of deformation of the wellbore.
[0087] The deformation degree predicted by SVR is used as an input feature of the neural network. The original crack width and spacing features are also added to the input of the neural network. After normalizing the input features, the neural network model is constructed. Its specific expression is as follows:
[0088]
[0089] In equation (6), a represents the output of the neuron, g(·) represents the activation function, and w i Let x represent the weight of the i-th input signal. i Let x represent the i-th input signal and b represent the bias; the neuron receives multiple input signals x. i Each input signal is multiplied by its corresponding weight w. i They are weighted and a bias b is added, and their weighted sum is fed into the activation function g(·) to produce the neuron's output a;
[0090] The ReLU activation function is used to change all input values less than 0 to 0, while keeping values greater than 0 unchanged. The expression for the ReLU function is:
[0091] g(z) = max(0, z) (7)
[0092] Map the input of neuron z to a specific output value g(z), and calculate the output of neuron a = g(z).
[0093] When using cross-validation to evaluate model performance, the specific steps include:
[0094] We choose k-fold cross-validation, which divides the dataset into k parts. Each time, we use k-1 parts as the training set and the remaining part as the validation set. We repeat this process k times to obtain k validation results.
[0095] Calculate the average of the model performance metrics in k validations, and use it as the evaluation metric for the neural network model;
[0096] Compare the average performance metrics of all hyperparameter configurations, select the configuration with the highest average performance as the optimal model parameter setting, retrain the model using this parameter setting, and evaluate its performance on the test set.
[0097] In step S5, model validation specifically includes the following steps:
[0098] Assuming we are currently considering the j-th neuron in the l-th layer, calculate the error of the output layer:
[0099]
[0100] In equation (8), J is the loss function. It is the input of the j-th neuron in the l-th layer;
[0101] Recursively calculate the error of the previous layer:
[0102]
[0103] In equation (9), n l It is the number of neurons in the i-th layer. is the weight connecting the j-th neuron in layer (l-1) and the k-th neuron in layer l, and g′(·) is the derivative of the activation function;
[0104] Calculate the gradients of the weights and biases:
[0105]
[0106] In equation (10), It is the output of the i-th neuron in the (l-1)-th layer;
[0107] Update weights and biases:
[0108]
[0109] In equations (12) and (13), α is the learning rate. The formulas are used repeatedly until the convergence condition is met.
[0110] The loss function J uses the mean squared error (MSE), where the expression for the mean squared error (MSE) is:
[0111]
[0112] In equation (14), n is the sample size, h i For the true value, These are predicted values.
[0113] This wellbore stability model prediction method based on the instability mechanism of easily collapsible layers utilizes an online wellbore level gauge to measure the annular fluid level in real time. Drilling fluid parameters, such as density, are collected using on-site monitoring equipment such as integrated logging tools or drilling instruments. By initializing the wellbore structure, the bottom hole pressure parameters are quickly calculated using the measured data and associated parameters and transmitted in real time to the integrated logging tool. The integrated logging tool, according to pre-set system functions, performs real-time monitoring, remote transmission, and rapid alarm functions based on changes in bottom hole pressure. This invention combines the powerful predictive capabilities of SVR (Self-Range Dynamics) with the pattern recognition capabilities of neural networks, forming an integrated "prediction + evaluation" method. This synergy improves the overall system's predictive efficiency and the reliability of stability assessment. SVR effectively maps to a high-dimensional space through kernel tricks, capturing the nonlinear relationship between input features and wellbore deformation, thus improving the accuracy and adaptability of the prediction model. SVR can effectively handle high-dimensional feature spaces and will not fail due to the curse of dimensionality, thereby more comprehensively considering all relevant factors. Neural networks, acting as classifiers, can subdivide wellbore stability into multiple levels, such as stable, slightly deformed, moderately deformed, and severely deformed, based on the degree of deformation predicted by SVR and other features. This provides a more detailed risk assessment and facilitates the early implementation of preventative measures.
[0114] This wellbore stability model construction and prediction method based on the instability mechanism of easily collapsible layers allows the model to continuously learn and optimize with the arrival of new data, improving the model's timeliness and accuracy. The method combining SVR with neural networks significantly improves the safety and efficiency of drilling operations, solving problems such as insufficient prediction accuracy, poor adaptability, and coarse risk assessment in traditional methods. It not only increases the measurement frequency, enabling continuous testing of wellbore fluid levels, but also provides feasibility for real-time monitoring of fluid level changes. Furthermore, it allows for parallel operation with drilling, greatly improving wellbore fluid level safety and effectively enhancing work efficiency.
[0115] Example 2: This example provides a terminal device, which includes a memory, a processor, a communication interface, and a communication bus. The memory stores a program that can run on the processor. When the processor executes the program, it implements the wellbore stability model construction and prediction method based on the unstable mechanism of easily collapsing layers in the above example.
[0116] The processor can be a central processing unit, or it can be other general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.
[0117] The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and units, such as the program units corresponding to those in the above-described method embodiments of the present invention. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory, thereby realizing the wellbore stability model construction and prediction method based on the unstable mechanism of easily collapsing layers as described in the above embodiments.
[0118] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. The memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. One or more programs are stored in the memory and, when executed by the processor, perform the wellbore stability model construction and prediction method based on the unstable mechanism of easily collapsible layers described in the above embodiments.
[0119] Example 3: This example provides a storage medium for computer-readable storage. The storage medium stores one or more programs, which can be executed by one or more processors to implement the wellbore stability model construction and prediction method based on the unstable mechanism of easily collapsing layers as described in the above examples.
[0120] The storage medium can be an internal storage unit of the terminal device, such as the hard drive or memory of the terminal device. Alternatively, the storage medium can be an external storage device of the terminal device, such as a plug-in hard drive, smart memory card, secure digital card, or flash memory card installed on the terminal device.
[0121] The above technical features constitute the embodiments of the present invention, which have strong adaptability and implementation effect. Unnecessary technical features can be added or removed according to actual needs to meet the needs of different situations.
Claims
1. A method for constructing and predicting wellbore stability models based on the instability mechanism of easily collapsing layers, characterized in that... Includes the following steps: S1. Equipment arrangement: Based on the downhole conditions, select a fiber optic endoscope with high temperature and high pressure resistance and corrosion resistance. Based on the drilling depth and the well wall area to be observed, determine the length, diameter and lens type of the fiber optic endoscope. Integrate the fiber optic endoscope with the drill string to ensure that it can be used along the drill pipe or the measurement while drilling system during the drilling process or specific operation stages. S2. Data Collection and Preprocessing: Collect drilling data, geological data, and wellbore image data acquired by fiber optic endoscope, and preprocess the collected data to remove outliers and standardize the numerical range; the drilling data includes drilling pressure, rotation speed, and drilling fluid properties, the geological data includes rock strata characteristics and geostress distribution, and the wellbore image data includes fracture width and spacing. S3. Structural features: Extract key features from the preprocessed image data in step S2, and normalize all features based on drilling parameters and geological information. Divided into training set and test set; S4. Model Construction: Support Vector Regression (SVR) is used to predict the degree of wellbore deformation, and a neural network is used to evaluate the stability level. The training set data is used to fit the model parameters, and cross-validation is used to optimize the model performance. S5. Model Validation: Test the model on the validation set and evaluate the prediction accuracy; S6. Real-time monitoring and prediction: Input the real-time acquired downhole images and drilling parameters into the trained model to predict the stability of the wellbore in real time. Combine the prediction results with preset thresholds to activate the early warning system and adjust the drilling strategy in a timely manner. S7. Feedback and Iteration: New data collected during actual drilling is continuously fed back into the model, continuously updating and optimizing the prediction model.
2. The wellbore stability model construction and prediction method based on the unstable mechanism of easily collapsing layers as described in claim 1, characterized in that... In step S2, based on drilling data and wellbore image data, features highly correlated with wellbore stability are selected to perform data cleaning.
3. The wellbore stability model construction and prediction method based on the unstable mechanism of easily collapsing layers according to claim 1 or 2, characterized in that... In step S3, when normalizing all features, Min-Max scaling is used to map the values of crack width and spacing features to the range [0,1]. Through linear transformation of the data, the minimum value of the data is 0 and the maximum value is 1; the specific formula is: In equation (2), X represents the original data. min and X max These are the minimum and maximum values of the data, respectively.
4. The wellbore stability model construction and prediction method based on the unstable mechanism of easily collapsing layers according to claim 1 or 2, characterized in that... In step S4, the RBF kernel function is selected in the support vector regression (SVR) prediction of wellbore deformation. Its mathematical expression is: K(x,y)=exp(-γ||x-y|| 2 ) (3) In equation (3), x and y are the input data points, and γ is the width parameter of the kernel function; The RBF kernel function maps the original feature space to a higher-dimensional space, maximizing the distance from the hyperplane to each class's data points. The hyperplane is defined by the normal vector ω and the bias b, i.e.: ω·x+b=0 (4) The optimal hyperplane is obtained by determining the optimal support vector through a classification decision function. Specifically: In equations (4) and (5), x is the feature vector of the data point, y is the category label of the data point, and α i It is the optimal Lagrange multiplier; The optimal hyperplane is obtained through calculation, and all data points satisfying (ω·x+b)≥1 are used to predict the degree of deformation of the wellbore.
5. The wellbore stability model construction and prediction method based on the instability mechanism of easily collapsing layers as described in claim 4, characterized in that... The deformation degree predicted by SVR is used as an input feature of the neural network. The original crack width and spacing features are also added to the input of the neural network. After normalizing the input features, the neural network model is constructed. Its specific expression is as follows: In equation (6), a represents the output of the neuron, g(·) represents the activation function, and w i Let x represent the weight of the i-th input signal. i Let x represent the i-th input signal and b represent the bias; the neuron receives multiple input signals x. i Each input signal is multiplied by its corresponding weight w. i They are weighted and a bias b is added, and their weighted sum is fed into the activation function g(·) to produce the neuron's output a; The ReLU activation function is used to change all input values less than 0 to 0, while keeping values greater than 0 unchanged. The expression for the ReLU function is: g(z) = max(0, z) (7) Map the input of neuron z to a specific output value g(z), and calculate the output of neuron a = g(z).
6. The wellbore stability model construction and prediction method based on the instability mechanism of easily collapsing layers according to claim 1, 2, or 5, characterized in that... When using cross-validation to evaluate model performance, the specific steps include: We choose k-fold cross-validation, which divides the dataset into k parts. Each time, we use k-1 parts as the training set and the remaining part as the validation set. We repeat this process k times to obtain k validation results. Calculate the average of the model performance metrics in k validations, and use it as the evaluation metric for the neural network model; Compare the average performance metrics of all hyperparameter configurations, select the configuration with the highest average performance as the optimal model parameter setting, retrain the model using this parameter setting, and evaluate its performance on the test set.
7. The wellbore stability model construction and prediction method based on the instability mechanism of easily collapsing layers according to claim 1, 2, or 5, characterized in that... In step S5, model validation specifically includes the following steps: Assuming we are currently considering the j-th neuron in the l-th layer, calculate the error of the output layer: In equation (8), J is the loss function. It is the input of the j-th neuron in the l-th layer; Recursively calculate the error of the previous layer: In equation (9), n l It is the number of neurons in the i-th layer. is the weight connecting the j-th neuron in layer (l-1) and the k-th neuron in layer l, and g′(·) is the derivative of the activation function; Calculate the gradients of the weights and biases: In equation (10), It is the output of the i-th neuron in the (l-1)-th layer; Update weights and biases: In equations (12) and (13), α is the learning rate. The formulas are used repeatedly until the convergence condition is met.
8. The wellbore stability model construction and prediction method based on the instability mechanism of easily collapsing layers according to claim 7, characterized in that... The loss function J uses the mean squared error (MSE), where the expression for the mean squared error (MSE) is: In equation (14), n is the sample size, h i For the true value, These are predicted values.
9. A terminal device, comprising a memory and a processor, wherein the memory stores a program executable on the processor, characterized in that, When the processor executes the program, it implements the wellbore stability model construction and prediction method based on the instability mechanism of easily collapsible layers as described in any one of claims 1 to 8.
10. A storage medium, characterized in that, The storage medium stores one or more programs, which can be executed by one or more processors to implement the wellbore stability model construction and prediction method based on the instability mechanism of easily collapsible layers as described in any one of claims 1 to 8.