Fault area underground engineering surrounding rock stress and deformation prediction method
Through a combination of multi-task learning algorithm and numerical calculation, a stress deformation prediction model was constructed, which solved the problem of low prediction efficiency and accuracy in the study of the stress and deformation of faults on the surrounding rocks of the chamber, and achieved more efficient underground engineering surrounding rocks. Stress and deformation prediction.
Patent Information
- Application Number
- CN202411528673.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-30
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing research methods on the impact of faults on the stress and deformation of the surrounding rock of the cave chamber have poor prediction efficiency and low accuracy.
Using multi-task learning algorithm combined with numerical calculations, a stress deformation prediction model is constructed. Through the training data set, a number of preset working conditions parameters, stress data and deformation data are included, and the adaptive task weight adjustment and Huber loss function are used to generate a prediction model to predict. Stress and deformation after excavation of underground engineering.
It improves the accuracy and efficiency of surrounding rock stress and deformation prediction, reduces the calculation amount of finite element analysis, and can quickly and accurately predict the stress and deformation of underground engineering surrounding rocks.
Smart Images

Figure CN119047061A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a method for predicting surrounding rock stress and deformation of underground engineering in a fault area, belonging to the technical field of underground rock engineering. Background Art
[0002] Underground projects face a variety of geological and engineering risks during construction and operation, among which the stability of the surrounding rock has always been a key factor affecting the safety and economy of the project. The existence of faults has a significant impact on the stress and deformation of the surrounding rock of underground projects. In order to study the influence of faults on the stress and deformation of the surrounding rock of caverns, many scholars have conducted a large number of studies using numerical simulation and model experiments, but these studies usually only consider a single or two influencing factors, so their prediction efficiency and accuracy of stress and deformation are poor. Summary of the invention
[0003] The present invention provides a method for predicting stress and deformation of surrounding rock of underground engineering in fault area, which can solve the problems of poor prediction efficiency and low accuracy of existing research methods on the influence of fault on stress and deformation of surrounding rock of cavern.
[0004] The present invention provides a method for predicting stress and deformation of surrounding rock in underground engineering in a fault area, the method comprising: S1. Construct a training data set based on multiple sets of preset working condition parameters of the surrounding rock of the underground engineering in the fault area and the stress data and deformation data corresponding to each set of preset working condition parameters; wherein each set of preset working condition parameters includes initial ground stress conditions and fault geometry parameters; S2. Using a multi-task learning algorithm to train the training data set to obtain a stress-deformation prediction model; S3. Inputting the measured working condition parameters before the underground engineering excavation into the stress-deformation prediction model to obtain the predicted stress value and predicted deformation value of the surrounding rock after the underground engineering excavation.
[0005] Optionally, the S1 specifically includes: Conduct model simulation on underground engineering in the fault area to obtain a simulation model; A plurality of sets of preset working condition parameters are respectively input into the simulation model to obtain stress data and deformation data corresponding to each set of preset working condition parameters, and the plurality of sets of preset working condition parameters and their corresponding stress data and deformation data are combined into a training data set.
[0006] Optionally, the fault geometric parameters include fault dip, fault width and minimum distance between the fault and the underground engineering structure.
[0007] Optionally, the stress-deformation prediction model uses an adaptive task weight adjustment method to adjust the weight coefficient of the training loss value corresponding to each task.
[0008] Optionally, the weight coefficient of the training loss value of each task at each training is determined according to the rate of change of the training loss of each task.
[0009] Optionally, the weight coefficient of the training loss value of each task during each training is the ratio of the natural exponential function of the training loss change rate of the task to the maximum value of the natural exponential function of the training loss change rate of all tasks.
[0010] Optionally, the training loss change rate of each task during each training is a ratio of a training loss value of the task in the current training to a training loss value of the task in the previous training.
[0011] Optionally, the stress-deformation prediction model is trained using a Huber loss function.
[0012] Optionally, the multi-task learning algorithm is trained in a hard parameter sharing manner.
[0013] The beneficial effects that the present invention can produce include: The method for predicting stress and deformation of surrounding rock of underground engineering in fault area provided by the present invention predicts stress and deformation of surrounding rock of underground engineering in fault area by combining numerical calculation with multi-task learning. The trained stress-deformation prediction model can effectively capture the influence of fault geometric parameters on surrounding rock stress and deformation, and improve prediction efficiency and accuracy. Compared with traditional methods, the present invention uses a multi-task learning algorithm to effectively determine the characteristic relationship between various factors of surrounding rock and stress and deformation after excavation, which reduces the calculation amount of finite element analysis, greatly improves calculation efficiency, and can quickly and accurately predict the stress and deformation of surrounding rock of underground engineering after excavation. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 A flow chart of a method for predicting stress and deformation of surrounding rock in underground engineering in a fault region provided by an embodiment of the present invention; Figure 2 A numerical calculation model diagram provided for an embodiment of the present invention; Figure 3 The embodiment of the present invention provides Schematic diagram of the effect of the relative distance between the fault and the cave on the displacement of the surrounding rock; Figure 4 The embodiment of the present invention provides Schematic diagram of the effect of the relative distance between the fault and the cave on the displacement of the surrounding rock; Figure 5 The embodiment of the present invention provides Schematic diagram of the effect of the relative distance between the fault and the cave on the displacement of the surrounding rock; Figure 6 The embodiment of the present invention provides Schematic diagram of the effect of fault width on surrounding rock stress; Figure 7 The embodiment of the present invention provides Schematic diagram of the effect of fault width on surrounding rock stress; Figure 8 The embodiment of the present invention provides Schematic diagram of the effect of fault width on surrounding rock stress. DETAILED DESCRIPTION
[0015] The present invention is described in detail below in conjunction with embodiments, but the present invention is not limited to these embodiments.
[0016] The embodiment of the present invention provides a method for predicting stress and deformation of surrounding rock in underground engineering in a fault area. Figure 1 As shown, the method includes: S1. Construct a training data set based on multiple sets of preset working condition parameters of the surrounding rock of the underground engineering in the fault area and the stress data and deformation data corresponding to each set of preset working condition parameters; wherein each set of preset working condition parameters includes initial ground stress conditions and fault geometry parameters.
[0017] In an embodiment of the present invention, the fault geometric parameters may include the fault dip angle, the fault width and the minimum distance between the fault and the underground engineering structure.
[0018] S1 specifically includes: firstly, model simulation is performed on the underground engineering in the fault area to obtain a simulation model; then, multiple sets of preset working condition parameters are respectively input into the simulation model to obtain stress data and deformation data corresponding to each set of preset working condition parameters, and the multiple sets of preset working condition parameters and their corresponding stress data and deformation data are combined into a training data set.
[0019] The above-mentioned underground projects may be engineering structures such as underground caverns.
[0020] First, the underground caverns in the fault area are modeled by numerical calculation methods such as finite element analysis. The coupling of multiple factors needs to be considered in the simulation, including fault dip, fault width, minimum distance between the fault and the cavern, and initial geostress conditions. By setting these parameters in the simulation model and performing numerical simulation, the stress and deformation values of the surrounding rock after excavation at various locations can be obtained. The numerical simulation results will be used to form a training data set, which contains multidimensional data such as stress values, deformation values, initial geostress conditions, and fault geometric parameters. Finally, the training data set will be stored in a structured form for subsequent multi-task learning model training.
[0021] S2. Use a multi-task learning algorithm to train the training data set to obtain a stress-deformation prediction model.
[0022] Specifically, each set of preset operating parameters of the surrounding rock of the underground cavern in the training data set can be used as input, and the simulated stress and deformation values of the surrounding rock after excavation at various locations can be used as output to perform multi-task learning training to obtain a stress-deformation prediction model, which can reflect the characteristic relationship between different factors of the surrounding rock of the underground cavern and the stress and deformation values of the surrounding rock after excavation at various locations.
[0023] In the embodiment of the present invention, the multi-task learning algorithm is trained by sharing hard parameters.
[0024] For the prediction task of surrounding rock stress and deformation in the fault area, the two tasks are closely related, and the data input during the training process is the same. Therefore, the training is performed in a hard parameter sharing manner, which can obtain a stress-deformation prediction model with better prediction effect.
[0025] The basic network structure of the hidden layer of multi-task learning is a deep neural network structure. A basic deep neural network consists of three parts: input layer, hidden layer and output layer. The hidden layer is responsible for the training of the deep neural network and consists of multiple layers of fully connected neurons. Layer neurons, and the input data is Dimension, Layer The output of a neuron for: (1) in, and Respectively The weight coefficients and bias coefficients of the layers, is the activation function. Represents the set of weight matrices and bias vectors, denoted as .when When The input layer is The training of the neural network is done by continuously updating and Until the output condition is met. Since deep neural networks have a large number of parameters, equations with a gradient of 0 are usually difficult to solve directly, but the gradient can be directly calculated to update the network parameters. The generalized neural network parameter gradient equation is expanded according to the first-order Taylor formula as follows: (2) in is a small vector, and we can use it as a unit vector Indicates that the parameter gradient equation is: (3) in is a scalar representing the update step size. The purpose of local gradient descent is to update Will reduce neurons ,Right now .because is a scalar, then the inner product of the remaining two vectors needs to be less than 0, satisfying:
[0026] in is the angle between two vectors, when and When the direction is opposite, It is the negative direction of the current gradient direction that ensures The direction of is the direction where the local gradient descent is fastest. Therefore, the deep neural network parameter update rule is: (5) merge To step size factor The neural network learning rate is obtained , learning rate It controls the parameter update step size and model learning speed, and its relationship is expressed as: (6) in Indicates the gradient value of the parameter to be updated. Before training, the initial value of the parameter should be set. Different initialization methods may have different training effects.
[0027] In an embodiment of the present invention, the stress-deformation prediction model is trained using the Huber loss function, which can enhance the robustness of the model to outlier data by combining the advantages of mean square error (MSE) and mean absolute error (MAE).
[0028] Specifically, after the training data is calculated by the model, a set of predicted values will be obtained. The difference between the predicted value and the true value is called loss. The training process of deep neural networks is to achieve the minimum loss while regularizing the parameters. This process requires the introduction of a loss function. In the learning process of the model, the feature data predicts the output value through the model. This process is called forward propagation, and the loss between the output value and the true value needs to be continuously reduced by updating the parameters of each neuron in the model. This process is called backward propagation. The training process continues when the loss function starts the backward propagation by receiving the result of forward propagation. In this process, the choice of loss function is very critical. The commonly used loss functions in data regression prediction are: 1) Mean Absolute Loss (MAE): It measures the average magnitude of the model prediction error by calculating the absolute value of the difference between the output value and the true value. Its mathematical definition is as follows: (7) in is the true value, is the output value of the model. The unique characteristics of this loss function include robustness to outliers. Because it is based on absolute error, the impact of outliers increases linearly over the entire error range. This makes the MAE loss function more robust in the presence of outliers. The MAE loss function is widely used in sparse modeling. This loss function causes some coefficients to tend to zero, thereby achieving feature selection. This is particularly useful in high-dimensional data sets and helps improve the generalization ability of the model.
[0029] 2) Squared Error Loss (MSE): Calculate the sum of the squares of the difference between the output value and the true value to measure the magnitude of the model prediction error. Its mathematical definition is as follows: (8) The primary feature of MSE is that it is the square form of the MAE loss function, so the penalty of MSE on the error increases quadratically as the error increases. This makes the MSE loss function smoother in the optimization problem, which helps to use gradient-based optimization algorithms such as gradient descent. Therefore, considering the advantages of the above two loss functions, the switching node is selected by the data distribution characteristics during the training process to give full play to the advantages of the two loss functions in the training process. The present invention adopts the Huber loss function as the loss function for model training. The mathematical definition of the Huber loss function is as follows: (9) In the formula is the actual value, is the predicted value, which is a threshold used to determine whether to use square loss or absolute loss in training. When the difference is greater than When the model is trained, the absolute loss is used. This makes it robust to outliers, because for large residuals, its loss grows slower than the squared loss function. The Huber loss function balances the sensitivity to outliers and the efficiency of squared errors. This feature ensures that outliers have a more significant impact on the model. The degree of outliers in the data during training and the importance of outliers affect the update of neural network parameters. The differences in the variance, median, upper quartile, and lower quartile in the training data and the variance, median, upper quartile, and lower quartile in the data set are used to update the neural network parameters. Make adjustments: (10) In the formula are the standardized variance, first, second, and third quartiles of each task in the training set, respectively. To predict the variance, first, second, and third quartiles of the data, weights of 1, 0.1, 0.05, and 0.05 are given respectively according to the degree of influence of the variance, second quartile, first and third quartiles on the data distribution.
[0030] The Huber loss function is determined by calculating the variance, median, upper quartile and lower quartile of the predicted data for each task. This can monitor the data distribution during the training process, induce the predicted data to present the distribution characteristics of the training data, fully consider the outlier data in the stress and deformation data, and improve the accuracy of the model prediction.
[0031] In an embodiment of the present invention, preferably, the stress-deformation prediction model adopts an adaptive task weight adjustment method to adjust the weight coefficient of the training loss value corresponding to each task.
[0032] Furthermore, the weight coefficient of the training loss value of each task at each training is determined according to the training loss change rate of each task.
[0033] Specifically, the weight coefficient of the training loss value of each task at each training is the ratio of the natural exponential function of the task's training loss change rate to the maximum value of the natural exponential function of the training loss change rate of all tasks. Among them, the training loss change rate of each task at each training is the ratio of the task's training loss value in this training to the training loss value of the previous training.
[0034] The adaptive task weight adjustment method dynamically adjusts the weight of each task based on the change rate of the training loss of each task to balance the training progress of the multi-task learning model.
[0035] In practical applications, the adaptive task weight adjustment method uses an optimization algorithm based on gradient descent to continuously update task weights so that the loss function value of each task converges faster, thereby improving the overall training efficiency of the model.
[0036] In multi-task learning, the training of the model is usually obtained by weighting multiple loss functions. The dimensions of the loss functions used by different tasks and the importance of the task require additional setting of new parameters, and the parameter adjustment process usually consumes a lot of time. Therefore, the present invention considers the method of automatically adjusting the weights of each loss function to liberate training efficiency and optimize model performance. The overall loss value for the surrounding rock stress and deformation prediction task is defined as follows: (11) In the formula , Corresponding to The weight coefficient of the adaptive training loss value of each task and the training loss value. Define the training loss change rate of each task for: (12) In the formula Indicates Task No. The training loss value of the training, Indicates Task No. The training loss value of the training, It means the Mission No. The training loss change rate of the training, that is, the ratio of the current training loss value to the previous training loss value. The larger this ratio is, the slower the decrease speed of the training loss value of the task in the current training, and the more difficult the task is to train. Therefore, in order to balance the training speed of each task, a larger loss weight should be given to the task with a slow decrease in training loss value. By introducing an adaptive task weight coefficient, the difference in loss value changes between tasks is automatically balanced. The weight coefficient of the adaptive training loss value is: (13) In the formula It is a natural constant that amplifies the loss change rate and improves the difference in the loss change rate between tasks. The stress of characteristic points on the left, middle and right sides of the surrounding rock of the cavern; It is the deformation of characteristic points on the left, middle and right sides of the cave surrounding rock.
[0037] According to the above rules, each task after each training is re-assigned a new weight coefficient to balance the impact of the difference in the loss function value of different tasks on the training effect between multiple tasks. In addition, in order to reduce the uncertainty caused by stochastic gradient descent and random training data selection during training, the process of updating the weight coefficient is completed at the epoch level, where and The loss value is the mean of all batch Loss loss values.
[0038] In practical applications, the open source machine learning platform TensorFlow2.4 can be used to build and train the model. The training code can be written in Python, and the NVIDIA RTX3080 device can be used for training. The training process lasts for 5000 epochs to ensure the adequacy and stability of the model under different training parameters.
[0039] S3. Input the measured working condition parameters before the underground engineering excavation into the stress-deformation prediction model to obtain the predicted stress value and predicted deformation value of the surrounding rock after the underground engineering excavation.
[0040] After the stress-deformation prediction model is trained and optimized, the measured working condition parameters before the underground cavern excavation are used as input for prediction. The input data include the measured fault geometry parameters and initial geostress conditions. Based on these inputs, the stress-deformation prediction model will predict the stress and deformation of the surrounding rock after the underground cavern excavation. In order to improve the accuracy of the prediction results, the prediction results can be compared and corrected with the limited measured data to ensure that the stress and deformation values output by the stress-deformation prediction model can accurately reflect the actual situation of the underground project.
[0041] The following will take the surrounding rock of underground engineering in a typical two-dimensional fault area as the engineering background to analyze and summarize the stress and deformation of the surrounding rock. During the excavation of the cavern, the originally stable geological body is unloaded and deformed, causing the stress state of the surrounding rock to change significantly. This stress redistribution process becomes extremely complicated when encountering geological discontinuities such as faults. Due to the inherent differences in physical and mechanical properties, the transmission and distribution patterns of stress in faults are fundamentally different from those of the surrounding continuous rock mass. Near the fault, the stress concentration phenomenon is particularly obvious, which can easily lead to deformation or even destruction of the rock, thereby affecting the stability and safety of the cavern. More importantly, the diversity of fault geometric parameters makes its influence on the stress state of the surrounding rock highly uncertain. Therefore, it is necessary to explore the influence of various factors of the fault on the stress and deformation of the cavern surrounding rock to ensure the stability and safety of the underground structure.
[0042] In order to determine the influence of faults on the stress and deformation of the surrounding rock in the range of the cavern roof arch, the model established this time takes a span of 25m and sets the cavern height to 1.5 times the span, that is, 37.5m. Figure 2 (a) and (b) show the schematic diagram of the simulation model parameter setting, taking the horizontal direction as Coordinate axis, vertical direction is Two faults are set in the simulation model. The fault on the left side of the plant is The fault on the right side of the plant is The definitions of the parameters in the simulation model are shown in Table 1. Considering the calculation efficiency and accuracy, the modeling area is three times the span of the cavern, and the model size is ,See Figure 2 (a) shown.
[0043] Table 1. Definition of simulation model parameter symbols
[0044] The vault and arch ends of underground caverns are not only key structures for bearing and dispersing ground pressure, but also the cornerstone for maintaining the stability of the entire underground space. During the design and construction process, the structural deformation and stress characteristics of the vault and arch ends and their interactions should be fully considered to ensure the safety and stability of underground projects. Figure 2 The stress and deformation definition symbols of the cavern vault and the arch ends on both sides shown in (b) are shown in Table 2.
[0045] Table 2 Symbolic definitions of simulation model parameters
[0046] right Figure 2 The simulation model given can obtain a variety of combined working conditions by considering the change of its value range. The fault dip angles on both sides are 30°, 45°, and 60°; the distance between the left fault and the tangent line of the cavern vault is The value range is 0m, 1m, 3m, 5m; the distance between the right fault and the tangent line of the cavern vault The value range is 0m, ±1m, ±3m, ±5m (when the value is negative, it indicates the distance between the fault and the cavern); the left and right fault widths are 0.5m, 1m, 1.5m; the initial geostress condition is selected ; There are 5 groups in total. Table 3 shows the orthogonal combinations of all influencing factors, and the total number of calculation conditions is 11,340.
[0047] Table 3 Working condition combination table
[0048] The parameter values are shown in Table 4 The surrounding rock and fault parameters of the underground powerhouse of Yebatan Hydropower Station. The numerical calculation was performed using Flac2D9.0, and the null model was selected to simulate the excavation of the underground powerhouse cavern, and the Morh-Coulomb model was selected for the constitutive relationship of the surrounding rock.
[0049] Table 4 Regional parameters of simulation model
[0050] Excavation calculation and analysis were carried out on the 11340 fault and surrounding rock stress combination conditions obtained in Table 3, and the stress and deformation (i.e. displacement) data of the characteristic points of each condition were recorded as samples for deep learning, forming a training data set of surrounding rock stress and deformation values. Each condition in this training data set contains 14 data, including: , fault geometry parameters And the corresponding characteristic point stress and displacement data .
[0051] The combined working condition data of 11,340 faults and surrounding rocks, as well as the corresponding stress data and deformation data, are used to form a training data set, and then the training data set is trained using a multi-task learning algorithm to obtain a stress-deformation prediction model. The input of the stress-deformation prediction model includes fault geometry parameters and initial geostress conditions, and the output is the stress and deformation values of the surrounding rock. During the training process, the data set is divided into a training set and a validation set, with the initial ratio set to 70% for training and 30% for validation. By gradually increasing the proportion of the training set, the model gradually establishes the characteristic relationship between various fault factors and the stress and deformation values of the surrounding rock.
[0052] In order to further improve the prediction accuracy, the Huber loss function is introduced in the training process. This loss function can effectively deal with outliers that may exist in the data set and reduce the interference of these outliers on the model prediction results. δ By making dynamic adjustments, the model is optimized and has higher robustness when facing complex stress distribution situations.
[0053] So far, the multi-task learning of surrounding rock stress and deformation of underground engineering in two-dimensional fault area has been completed. In order to predict the influence of the distance between fault and cave on surrounding rock deformation and stress, it can be fixed The fault position makes its influence on the displacement of each feature point minimal. The fault geometry parameters are , , and then change Distance from the cave, research The influence of the relative distance between the fault and the cave on the deformation and stress of the characteristic points of the surrounding rock. The stress-deformation prediction model is used to predict the influence of the relative distance between the fault and the cave on the surrounding rock stress and the influence of the relative distance between the fault and the cave on the surrounding rock displacement. Figure 3 for The influence of the relative distance between the fault and the cave on the displacement of the surrounding rock, where (a) represents the displacement of the left arch end under different ground stress conditions , (b) shows the influence of different ground stress conditions on the displacement of the arch crown , (c) shows the displacement of the right arch end under different ground stress conditions The influence of, (d) is the legend and schematic diagram of parameter changes; Figure 4 for The influence of the relative distance between the fault and the cave on the displacement of the surrounding rock, where (a) represents the displacement of the left arch end under different ground stress conditions , (b) shows the influence of different ground stress conditions on the displacement of the arch crown , (c) shows the displacement of the right arch end under different ground stress conditions The influence of, (d) is the legend and schematic diagram of parameter changes; Figure 5 for The influence of the relative distance between the fault and the cave on the displacement of the surrounding rock, where (a) represents the displacement of the left arch end under different ground stress conditions , (b) shows the influence of different ground stress conditions on the displacement of the arch crown , (c) shows the displacement of the right arch end under different ground stress conditions The influence of, (d) is the legend and schematic diagram of parameter changes; Figure 6 for The influence of fault width on surrounding rock stress, where (a) represents the stress on the left arch end under different ground stress conditions. , (b) shows the effect of different ground stress conditions on the arch stress , (c) shows the effect of stress on the right arch end under different ground stress conditions The influence of, (d) is the legend and schematic diagram of parameter changes; Figure 7 for The influence of fault width on surrounding rock stress, where (a) represents the stress on the left arch end under different ground stress conditions. , (b) shows the effect of different ground stress conditions on the arch stress , (c) shows the effect of stress on the right arch end under different ground stress conditions The influence of, (d) is the legend and schematic diagram of parameter changes; Figure 8 for The influence of fault width on surrounding rock stress, where (a) represents the stress on the left arch end under different ground stress conditions. , (b) shows the effect of different ground stress conditions on the arch stress , (c) shows the effect of stress on the right arch end under different ground stress conditions , (d) is the legend and schematic diagram of parameter changes.
[0054] The embodiment of the present invention successfully predicted the stress and deformation of the surrounding rock of underground engineering in the fault area by combining numerical calculation with multi-task learning. The results show that the trained stress-deformation prediction model can effectively capture the influence of fault geometric parameters on the stress and deformation of the surrounding rock, and improve the accuracy and robustness of the prediction through dynamic optimization. This method not only reduces the computational complexity in traditional finite element analysis, but also improves the prediction efficiency, providing reliable technical support for practical engineering applications.
[0055] The above are only a few embodiments of the present application and do not constitute any form of limitation to the present application. Although the present application is disclosed as above with preferred embodiments, it is not intended to limit the present application. Any technician familiar with the profession, without departing from the scope of the technical solution of the present application, using the technical content disclosed above to make slight changes or modifications are equivalent to equivalent implementation cases and fall within the scope of the technical solution.
Claims
1. A method for predicting surrounding rock stress and deformation of underground engineering in fault areas, characterized in that: The method comprises: S1. Construct a training data set based on multiple sets of preset working condition parameters of the surrounding rock of the underground engineering in the fault area and the stress data and deformation data corresponding to each set of preset working condition parameters; wherein each set of preset working condition parameters includes initial ground stress conditions and fault geometry parameters; S2. Using a multi-task learning algorithm to train the training data set to obtain a stress-deformation prediction model; S3. Inputting the measured working condition parameters before the underground engineering excavation into the stress-deformation prediction model to obtain the predicted stress value and predicted deformation value of the surrounding rock after the underground engineering excavation.
2. The method according to claim 1, characterized in that: The S1 specifically includes: Conduct model simulation on underground engineering in the fault area to obtain a simulation model; A plurality of sets of preset working condition parameters are respectively input into the simulation model to obtain stress data and deformation data corresponding to each set of preset working condition parameters, and the plurality of sets of preset working condition parameters and their corresponding stress data and deformation data are combined into a training data set.
3. The method according to claim 1, characterized in that: The fault geometric parameters include fault dip, fault width and the minimum distance between the fault and the underground engineering structure.
4. The method according to claim 1, characterized in that: The stress-deformation prediction model adopts an adaptive task weight adjustment method to adjust the weight coefficient of the training loss value corresponding to each task.
5. The method according to claim 4, characterized in that The weight coefficient of the training loss value of each task at each training is determined according to the rate of change of the training loss of each task.
6. The method according to claim 5, characterized in that The weight coefficient of the training loss value of each task at each training is the ratio of the natural exponential function of the training loss change rate of the task to the maximum value of the natural exponential function of the training loss change rate of all tasks.
7. The method according to claim 5, characterized in that The training loss change rate of each task at each training is the ratio of the training loss value of the task in this training to the training loss value of the previous training.
8. The method according to claim 1, characterized in that The stress-strain prediction model is trained using the Huber loss function.
9. The method according to claim 1, characterized in that: The multi-task learning algorithm is trained in a hard parameter sharing manner.
Citation Information
Patent Citations
Remote sensing image cultivated land parcel segmentation method based on multi-task learning
CN116188993A
Ground pressure disaster prediction method and device based on multi-task learning
CN117094235A
Model and method for judging ground stress state of tunnel crossing fault
CN118070388A
Truss stress prediction and weight lightening method based on transfer learning fusion model
WO2023115596A1
Cited By
Surrounding rock deformation prediction method for tunnel construction in soft rock area
CN119475122A
A method for predicting the deformation of surrounding rock in tunnel construction in soft rock areas
CN119475122B
Fault geology cavern surrounding rock deformation prediction method based on excavation sequence optimization and NPDOA-LSTM model
CN120805663A
A fault geologic cavern surrounding rock deformation prediction method based on excavation sequence optimization and NPDOA-LSTM model
CN120805663B
