Hydropower engineering underground cavern group crustal stress field prediction method

Abstract

The invention relates to the technical field of crustal stress field prediction, in particular to a hydropower engineering underground cavern group crustal stress field prediction method. According to the technical scheme, the method comprises the following steps of collecting geological data of an underground cavern group area of the water-power engineering, wherein the geological data comprises landform data, stratum lithology data and geological structure data; and constructing a crustal stress field prediction model by using an artificial intelligence algorithm, the artificial intelligence algorithm being a deep neural network algorithm, and the number of neurons in an input layer being determined according to the geological data feature dimensions. According to the method, the crustal stress field prediction model is constructed and trained by using the deep neural network algorithm in combination with various geological data, so that the crustal stress field is accurately predicted, the design construction scheme of the underground cavern group can be further optimized, the decision is aided by visual display, the model performance is guaranteed through various technical means, and the construction efficiency is improved. And the stability, safety, economical efficiency and construction efficiency of underground cavern group construction of the water-power engineering are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of geostress field prediction, and in particular to a method for predicting geostress fields of underground cavern groups in hydropower projects. Background Art

[0002] With the rapid development of underground engineering construction, the number and scale of underground caverns such as underground powerhouses, traffic tunnels, and mine tunnels in large hydropower projects are increasing. In the design and construction process of these underground caverns, accurate prediction of the geostress field is crucial, which is directly related to the stability analysis, support design, construction safety and other issues of the caverns.

[0003] However, the existing method for predicting the geostress field of underground caverns has an obvious flaw. When faced with complex geological conditions, traditional prediction methods seem to be powerless. The geological environment in which the underground caverns are located is a complex system, covering a variety of topography and landforms. For example, different terrain features such as high mountains and canyons, hilly plains, etc. will have different effects on underground stress; the lithology of the formations is even more diverse, from hard granite to soft shale, and their mechanical properties vary greatly, and the transmission and distribution of stress have completely different performances; coupled with the intricate geological structures, such as folds, faults, joints and fissures, the existence of these structures destroys the integrity of the rock mass, making the distribution law of geostress extremely complicated.

[0004] Traditional methods for predicting geostress fields mostly rely on simplified theoretical models and empirical formulas. When considering the influence of topography and geomorphology, simple terrain correction coefficients are usually used for approximate processing, which cannot accurately reflect the complex effects of stress concentration and unloading caused by the actual terrain fluctuations. For the differences in stratum lithology, calculations can often only be made based on a few typical lithology parameters, and it is difficult to fully cover the comprehensive effects of various lithologies and their combinations on geostress. In terms of dealing with geological structures, due to the irregularity and uncertainty of the structures, it is difficult for traditional methods to accurately simulate their interference and changes in the stress field. Generally, only the influence of large structures such as faults is roughly considered, and it is difficult to effectively include many small-scale structures and their interactions in the analysis.

[0005] For example, in the construction of underground caverns for hydropower projects in mountainous canyon areas, the deep valley terrain causes strong stress redistribution, and the stress state on both sides of the valley is very different from that at the bottom of the valley. At the same time, the multiple faults and dense joints developed in the strata further disrupt the normal transmission path of the ground stress. However, due to the above defects, the traditional prediction method cannot accurately determine the magnitude and direction of ground stress at different locations, which brings many risks to the design and construction of the cavern group, such as unreasonable support structure design, excessive deformation of surrounding rock, collapse and other safety accidents during construction, affecting the progress of the project and economic benefits.

[0006] In summary, the present application proposes a method for predicting the geostress field of an underground cavern group in a hydropower project. Summary of the invention

[0007] The purpose of the present invention is to propose a method for predicting the geostress field of an underground cavern group in a hydropower project in view of the problem that the existing method for predicting the geostress field of an underground cavern group in the background technology is insufficient in dealing with complex geological conditions.

[0008] The technical solution of the present invention is a method for predicting the geostress field of underground caverns in a hydropower project, comprising the following steps:

[0009] Step 1: Collect geological data of the underground cavern group area of ​​the hydropower project, wherein the geological data includes topographic data, stratum lithology data and geological structure data;

[0010] Step 2: Use an artificial intelligence algorithm to construct a geostress field prediction model. The artificial intelligence algorithm is a deep neural network algorithm. The number of neurons in the input layer is determined according to the characteristic dimension of the geological data. When the characteristic dimension of the geological data is [20,50], the number of neurons in the input layer is set to the corresponding characteristic dimension value; the hidden layer adopts a 3-5 layer structure, in which the number of neurons in the first hidden layer is 2-3 times the number of neurons in the input layer, and the number of neurons in subsequent hidden layers decreases successively, with a decreasing ratio of 0.6-0.8; the number of neurons in the output layer is 6, corresponding to the three principal stress components of the geostress field and their direction cosines, a total of 6 parameters;

[0011] Step 3: inputting the geological data as input data into the deep neural network algorithm for training to obtain a trained geostress field prediction model;

[0012] Step 4: Use the trained geostress field prediction model to predict the geostress field of the underground cavern group of the hydropower project and output the prediction results.

[0013] Optionally, the topographic data include regional elevation data, which ranges from [-2000m, 3000m], slope data ranging from [0°, 90°] and terrain undulation characteristic data; the stratigraphic lithology data include elastic modulus of different rock layers ranging from [1GPa, 100GPa], Poisson's ratio ranging from [0.1, 0.45], thickness data and rock layer distribution data; the geological structure data include the location coordinates of the fault, the strike range from [0°, 360°], the dip data range from [0°, 90°] and the distribution density of joints and fissures ranging from [0.1 / m, 10 / m], the length range from [0.1m, 10m] and the opening data range from [0.01mm, 100mm].

[0014] Optionally, when constructing a deep neural network algorithm, the number of input layer neurons is set to correspond to the characteristic dimension of the geological data, the hidden layer adopts a multi-layer structure, and the number of hidden layer neurons is adjusted according to the complexity of the geological data and the prediction accuracy requirements, and the number of output layer neurons corresponds to the number of stress components of the geostress field. The output calculation formula of the jth neuron in the lth layer in the deep neural network algorithm is:

[0015]

[0016] in is the output of the jth neuron in the lth layer, is the connection weight from the i-th neuron in the l-1th layer to the j-th neuron in the lth layer, is the output of the i-th neuron in the l-1th layer, is the bias of the jth neuron in the lth layer, is the number of neurons in the l-1th layer, f(x) is the activation function, and the ReLU function f(x)=max(0,x) is selected.

[0017] Optionally, the training process of the deep neural network algorithm adopts a back propagation algorithm, and the sum of squares of the errors between the predicted geostress value and the measured geostress value is used as the loss function. The loss function formula is:

[0018]

[0019] Among them, m is the number of samples, is the i-th true geostress field parameter value of the k-th sample, is the i-th predicted geostress field parameter value of the k-th sample, and θ is all the parameters of the neural network (including weights and biases). The loss function is minimized by adjusting the weights and thresholds of the neural network. The stochastic gradient descent method is used to optimize the update of weights and thresholds during the training process. The weight update formula is:

[0020]

[0021] Where α is the learning rate, and its value range is [0.001, 0.1].

[0022] Optionally, before using the trained geostress field prediction model for prediction, a step of validating the model is performed. A 5-10-fold cross-validation method is used to divide the collected geological data into a training set, a validation set, and a test set. The model is trained using the training set, the model hyperparameters are adjusted using the validation set, and the prediction performance of the model is evaluated using the test set. When the prediction root mean square error (RMSE) of the model is less than 1 MPa, a formal geostress field prediction is performed.

[0023] Optionally, after collecting geological data, the step of preprocessing the geological data is also included, wherein the preprocessing includes data cleaning, removing outliers and missing values ​​that exceed the reasonable range of the data by 3 times the standard deviation, and filling the missing values ​​with the mean or median of the feature column; data standardization, using the Z-score standardization method to unify data of different dimensions into the same numerical range, the formula is:

[0024]

[0025] Among them, x std is the standardized value, x is the original value, μ is the mean of the feature column, and σ is the standard deviation of the feature column; data enhancement, by performing random rotation, translation, scaling and other transformation operations on the existing geological data, the transformation parameter range is rotation angle [-10°, 10°], translation distance [0, 100m], scaling ratio [0.9, 1.1], each transformation operation generates 2-5 times the amount of original data to expand the data volume and improve the generalization ability of the model.

[0026] Optionally, the deep neural network algorithm introduces regularization terms during the training process, and the regularization terms include L1 regularization and L2 regularization, which are used to prevent the model from overfitting and improve the stability and accuracy of the model under complex geological conditions. The loss function formula after regularization is:

[0027]

[0028] Among them, λ is the regularization coefficient, and its value range is [0.001, 0.1].

[0029] Optionally, after obtaining the prediction results of the geostress field, the prediction results are visualized, including:

[0030] The topography, geological structure and geostress distribution of the underground cavern group area are presented in the form of three-dimensional images through three-dimensional modeling technology, which can intuitively show the changing laws of geostress fields under different geological conditions.

[0031] The OpenGL 3D graphics library is used for 3D modeling. The visualization of the geostress field uses color mapping or arrows to represent the stress magnitude and direction. The color mapping can use a gradient from blue to red to represent the stress from small to large. The arrow length is proportional to the stress magnitude, and the arrow direction represents the stress direction.

[0032] Optionally, in step 4, the design and construction of the underground cavern group is optimized according to the prediction results, specifically including:

[0033] Adjustment of cave shape: The radius of the circular cave is selected according to the predicted maximum principal stress in [5m, 50m] according to different stress value ranges; the length-to-width ratio of the rectangular cave is selected according to the angle between the predicted maximum principal stress direction and the long axis of the rectangle in [1:1, 1:5];

[0034] Cavern size and layout: The cavern size should be determined to meet engineering requirements and the distribution of the ground stress field, and the layout should minimize the stress interference between adjacent caverns;

[0035] Selection of support structure: Select different types of support structures such as anchor support, cable support or concrete lining according to the predicted maximum principal stress and ground stress distribution.

[0036] Determination of anchor rod parameters: The tensile strength of the anchor rod is 1.2-1.5 times the predicted maximum principal stress. Its installation position is determined based on the stress concentration area. The stress concentration area is determined by the stress gradient, and then the installation range is determined based on the stress concentration coefficient.

[0037] Optionally, when constructing the geostress field prediction model, physical constraints are integrated, and the physical constraints include the equilibrium equation of the rock mass:

[0038]

[0039]

[0040] Constitutive relation: {σ}=[D]{ε} and boundary conditions, where [D] is the elastic matrix, {σ} is the stress vector, and {ε} is the strain vector. The residual sum of squares of the equilibrium equation is added to the loss function as part of the regularization term.

[0041] Compared with the prior art, the present invention has at least one of the following beneficial technical effects:

[0042] By comprehensively collecting a variety of geological data and building a model using a deep neural network algorithm, the complex geological conditions can be fully considered. It can accurately capture the complex relationship between geological factors and the geostress field. Compared with traditional methods, it greatly improves the accuracy of geostress field prediction and reduces engineering risks caused by prediction deviations.

[0043] Based on the accurate prediction results, the shape, size, and layout of the cavern can be adjusted in a targeted manner, and the type of support structure and its parameters can be reasonably selected. For example, the parameters of circular and rectangular caverns can be optimized according to the stress conditions, and appropriate support can be used in appropriate areas, which can not only enhance the stability of the cavern, but also avoid resource waste, reduce project costs, and improve construction efficiency.

[0044] With the help of 3D modeling technology, the prediction results of the geostress field can be visualized and its relationship with geological elements can be intuitively presented, which is convenient for engineering personnel to understand and provide a reliable decision-making basis for design and construction. It is helpful to improve the coordination and work efficiency of all stages of the project, ensure the smooth progress and long-term stable operation of the project, and improve the overall benefits of the project.

[0045] Data preprocessing improves data quality and model generalization ability, cross-validation ensures model reliability, regularization terms prevent overfitting, and integration of physical constraints makes the model conform to the principles of mechanics. These measures jointly ensure that the model runs stably and accurately under complex geological conditions and enhance the credibility and practicality of the prediction results.

[0046] The present invention uses a deep neural network algorithm in combination with a variety of geological data to construct and train a geostress field prediction model, thereby achieving accurate prediction of the geostress field, and further optimizing the design and construction plans of underground cavern groups. At the same time, it uses visualization to assist decision-making and uses a variety of technical means to ensure model performance, effectively improving the stability, safety, economy and construction efficiency of the construction of underground cavern groups in hydropower projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The figure is a flow chart of a method for predicting the geostress field of underground caverns in hydropower projects. DETAILED DESCRIPTION

[0048] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0049] Example 1

[0050] like Figure 1 This embodiment provides a method for predicting the geostress field of an underground cavern group of a hydropower project, including data collection, data preprocessing, building a deep neural network model, model training, model verification, geostress field prediction and result visualization. The following is a detailed description of each process.

[0051] 1. Data Collection

[0052] First, the geological data of the underground cavern group area of ​​the hydropower project is collected comprehensively. Using geological exploration technology, drilling is performed to obtain stratigraphic lithology samples at different depths, and then the stratigraphic lithology data, including the elastic modulus of different rock layers (in the range of 1GPa to 100GPa), Poisson's ratio (0.1 to 0.45), thickness data and rock layer distribution data, are determined. Through high-precision measuring instruments, total stations and GPS, topographic and geomorphological data are obtained, including regional elevation data (-2000m to 3000m), slope data (0° to 90°) and terrain undulation feature data. For geological structural data, the location coordinates, strike (0° to 360°), inclination (0° to 90°) of the fault, and the distribution density (0.1 / m to 10 / m), length (0.1m to 10m), and opening data (0.01mm to 100mm) of the joints and fissures are determined by means of geological mapping and geophysical exploration. These data are organized into a structured data set to prepare for subsequent model construction and training.

[0053] 2. Data Preprocessing

[0054] Data cleaning: perform outlier detection on the collected data, calculate the mean and standard deviation of each feature column, and remove outliers that exceed the range of 3 times the standard deviation of the mean. For missing values, if they are numerical features, use the mean or median of the feature column to fill in. If there are missing values ​​in the elastic modulus of a certain formation lithology data, fill in the missing values ​​based on the mean or median of the elastic modulus of other measured data in the area where the formation is located.

[0055] Data standardization, using the Z-score standardization method to process the data. For each eigenvalue x, according to the formula Calculate, where μ is the mean of the feature column and σ is the standard deviation of the feature column. This can unify data of different dimensions into a similar numerical range, which is convenient for training the neural network model. Both the elevation data and the elastic modulus data are standardized to make them comparable in value.

[0056] Data enhancement, perform data enhancement operations on the preprocessed data. Expand the data volume through random rotation, translation, scaling and other transformations. The rotation angle is randomly selected in the range of [-10°, 10°], the translation distance is determined in the range of [0, 100m], and the scaling ratio is selected between [0.9, 1.1]. Each transformation operation generates 2-5 times the amount of new data of the original data. For example, the topographic data of a certain area is randomly rotated and translated to generate multiple sets of similar but slightly different data to improve the generalization ability of the model so that it can better adapt to the prediction of geostress fields under different geological conditions.

[0057] 3. Building a Deep Neural Network Model

[0058] The network structure is determined, and the number of input layer neurons is determined according to the characteristic dimension of the geological data. When the characteristic dimension is in [20,50], the number of input layer neurons corresponds to it. After the geological data is sorted, there are 30 features, so the number of input layer neurons is set to 30.

[0059] The hidden layer adopts a 3-5 layer structure. The number of neurons in the first hidden layer is set to 2-3 times the number of neurons in the input layer. Assuming that the number of neurons in the input layer is 30, the number of neurons in the first hidden layer can be set to 60-90. The number of neurons in subsequent hidden layers decreases successively, with a decreasing ratio of 0.6-0.8. For example, the number of neurons in the second hidden layer can be the number of neurons in the first hidden layer multiplied by a coefficient of 0.6-0.8.

[0060] The number of neurons in the output layer is 6, corresponding to the three principal stress components and their direction cosines of the geostress field.

[0061] Neuron calculation and activation function selection, set the number of input layer neurons to correspond to the characteristic dimension of the geological data, the hidden layer adopts a multi-layer structure, and the number of hidden layer neurons is adjusted according to the complexity of the geological data and the prediction accuracy requirements, the number of output layer neurons corresponds to the number of stress components of the geostress field, the output calculation formula of the jth neuron in the lth layer in the deep neural network algorithm is:

[0062]

[0063] in is the output of the jth neuron in the lth layer, is the connection weight from the i-th neuron in the l-1th layer to the j-th neuron in the lth layer, is the output of the i-th neuron in the l-1th layer, is the bias of the jth neuron in the lth layer, is the number of neurons in the l-1th layer, f(x) is the activation function, and the ReLU function f(x)=max(0,x) is selected. The activation function uses the ReLU function, which can speed up the training of the neural network and performs well in processing complex nonlinear relationships.

[0064] 4. Model Training

[0065] The loss function and regularization term are set, and the sum of square errors between the predicted ground stress value and the measured ground stress value is used as the loss function. The formula is:

[0066]

[0067] m is the number of samples, is the i-th true geostress field parameter value of the k-th sample, is the i-th predicted geostress field parameter value of the k-th sample, and θ is all the parameters of the neural network (including weights and biases).

[0068] Regularization terms are introduced, including L1 regularization and L2 regularization. The loss function formula after regularization is:

[0069]

[0070] Among them, λ is the regularization coefficient, and its value range is [0.001, 0.1]. The role of the regularization term is to prevent the model from overfitting and improve the stability and accuracy of the model under complex geological conditions.

[0071] Training algorithm selection and parameter setting

[0072] The back propagation algorithm is used, and the square sum of the errors between the predicted ground stress value and the measured ground stress value is used as the loss function. The loss function formula is:

[0073]

[0074] Among them, m is the number of samples, is the i-th true geostress field parameter value of the k-th sample, is the i-th predicted geostress field parameter value of the k-th sample, θ is all the parameters of the neural network, and the loss function is minimized by adjusting the weights and thresholds of the neural network. The stochastic gradient descent method is used to optimize the update of weights and thresholds during the training process. The weight update formula is:

[0075]

[0076] Where α is the learning rate, and its value range is [0.001, 0.1]. During the training process, set the number of training iterations, for example, 1000-5000 times, and determine whether to stop training early based on the performance of the model on the validation set.

[0077] 5. Model Validation

[0078] The 5-10 fold cross validation method was used to divide the collected geological data into training set, validation set and test set. The model was trained using the training set. During the training process, the validation set was used to evaluate the performance of the model after each training iteration. The root mean square error (RMSE) of the model on the validation set was calculated. If the RMSE is less than 1MPa, it is considered that the prediction performance of the model meets the preset accuracy requirements. Through cross validation, the generalization ability and stability of the model can be more comprehensively evaluated to avoid overfitting. At the same time, the hyperparameters of the model, such as the number of hidden layer neurons, learning rate, regularization coefficient, etc., can also be adjusted to obtain the best model performance.

[0079] VI. Prediction of Geostress Field

[0080] The geostress field of the underground caverns of hydropower projects is predicted using a trained and verified geostress field prediction model. The geological data of the area to be predicted is input into the model, and the model outputs the prediction results after calculation, namely the three principal stress components of the geostress field and their direction cosines.

[0081] 7. Results Visualization

[0082] Through 3D modeling technology, such as using 3D graphics libraries such as OpenGL or VTK, the topography, geological structure and geostress field distribution of the underground cavern group area are presented in the form of 3D images. For the visualization of the geostress field, color mapping or arrows are used to indicate the magnitude and direction of stress. The color mapping can be from blue to red to indicate the gradual change of stress from small to large, the arrow length is proportional to the stress magnitude, and the arrow direction indicates the stress direction. This can intuitively show the changing law of the geostress field under different geological conditions, helping engineers to better understand the distribution of the geostress field in the underground cavern group area.

[0083] Example 2

[0084] Based on Example 1, this example optimizes the design and construction of the underground cavern group according to the prediction results, specifically:

[0085] According to the size, direction and distribution characteristics of the geostress field, the shape of the cavern is adjusted. When a circular cavern is used, its radius is adjusted within the range of [5m, 50m] according to the predicted geostress value. The adjustment rule is: if the predicted maximum principal stress is less than or equal to 5MPa, the radius is selected within [5m, 20m]; if the predicted maximum principal stress is greater than 5MPa and less than or equal to 10MPa, the radius is selected within [10m, 30m]; if the predicted maximum principal stress is greater than 10MPa, the radius is selected within [20m, 50m]; when a rectangular cavern is used, its length-to-width ratio is adjusted within [1 :1,1:5], if the angle between the predicted maximum principal stress direction and the major axis of the rectangle is less than or equal to 30°, the aspect ratio is selected in [1:1,1:3]; if the angle between the predicted maximum principal stress direction and the major axis of the rectangle is greater than 30° and less than or equal to 60°, the aspect ratio is selected in [1:2,1:4]; if the angle between the predicted maximum principal stress direction and the major axis of the rectangle is greater than 60°, the aspect ratio is selected in [1:3,1:5]; determine the size of the cavern to meet the needs of hydropower engineering equipment installation and operation and adapt to the distribution of the ground stress field. When laying out the caverns, the stress interference between adjacent caverns should be minimized;

[0086] The type of support structure is selected according to the characteristics of the geostress field. When the predicted maximum principal stress is less than or equal to 8MPa and the geostress distribution is relatively uniform, anchor support is preferred; when the predicted maximum principal stress is greater than 8MPa and less than or equal to 15MPa, anchor cable support is preferred; when the predicted geostress field changes are complex and there are large stress concentration areas, concrete lining or combined support of anchor rods and anchor cables is used; the tensile strength of the anchor rod is 1.2-1.5 times the predicted maximum principal stress, and its installation position is determined according to the stress concentration area. The stress concentration area is determined by analyzing the predicted results of the geostress field. Specifically, if the stress gradient in a certain area is greater than the average of the surrounding areas, If the stress concentration coefficient is 1.5 times of the stress gradient, it is determined as a stress concentration area. The anchor rods should be installed in this area and a certain range around it along the direction of the principal stress. The range is determined according to the degree of stress concentration. If the stress concentration coefficient (the ratio of the maximum principal stress in the stress concentration area to the average maximum principal stress in the surrounding area) is less than or equal to 2, the installation range is 2-3m extending outward from the stress concentration area; if the stress concentration coefficient is greater than 2 and less than or equal to 3, the installation range is 3-5m extending outward from the stress concentration area; if the stress concentration coefficient is greater than 3, the installation range is 5-8m extending outward from the stress concentration area, so as to ensure the stability and safety of the underground cavern group during construction and operation.

[0087] In this embodiment, targeted cavern shape design: adjusting the cavern shape according to the accurate prediction results of the geostress field, such as the reasonable determination of the circular cavern radius and the aspect ratio of the rectangular cavern, can make the cavern structure better adapt to the geostress distribution. Properly increasing the circular cavern radius or adjusting the aspect ratio of the rectangular cavern in the high stress area can effectively reduce the stress concentration of the cave wall, reduce the risk of surrounding rock damage, improve the stability of the cavern, and also help optimize the use of engineering materials and reduce construction costs.

[0088] Furthermore, reasonable support structure selection and layout: Based on the predicted ground stress size and distribution characteristics, the appropriate support structure type (anchor support, cable support or concrete lining) is selected, and its parameters (such as anchor tensile strength and installation position) are determined to achieve accurate configuration of the support structure. Relatively simple anchor support is used in areas with small and uniform stress, and more powerful cable support or concrete lining is used in areas with high stress or complex stress. This can not only ensure the safety and stability of the cavern, but also avoid the waste of resources caused by excessive support, improve construction efficiency and reduce project costs.

[0089] The present invention optimizes the size, layout and support structure of the caverns and fully considers the influence of the ground stress field, which can effectively reduce the stress interference between adjacent caverns and enhance the overall stability of the entire underground cavern group. This helps to avoid chain reactions caused by local cavern instability, ensure the long-term safe and reliable operation of hydropower projects during construction and operation, reduce the cost of later maintenance and repair, and extend the service life of the project.

[0090] It is worth mentioning that the prediction results are displayed in three-dimensional visualization, showing the distribution of the geostress field in the underground cavern group area and its relationship with the topography and geological structure in an intuitive image form. This enables engineering designers, construction personnel and management personnel to understand the complex characteristics of the geostress field more clearly, facilitates communication and decision-making during the design and construction process, and improves work efficiency and coordination.

[0091] In addition, accurate prediction results and visual displays provide a scientific and reliable basis for decision-making in the design and construction of hydropower projects. In the project planning stage, the overall layout and structural form of the cavern group can be reasonably determined based on the geostress field information; during the construction process, the construction process and support measures can be adjusted in time to deal with possible geostress problems, thereby ensuring the smooth progress of the project, reducing project risks, and improving the economic and social benefits of the project.

[0092] The above specific embodiments are only several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A method for predicting the geostress field of underground caverns in hydropower projects, characterized in that: The following steps are involved: Step 1: Collect geological data of the underground cavern group area of ​​the hydropower project, wherein the geological data includes topographic data, stratum lithology data and geological structure data; Step 2: Use an artificial intelligence algorithm to construct a prediction model for the geostress field. The artificial intelligence algorithm is a deep neural network algorithm. The number of neurons in the input layer is determined according to the characteristic dimension of the geological data. When the characteristic dimension of the geological data is [20,50], the number of neurons in the input layer is set to the corresponding characteristic dimension value; the hidden layer adopts a 3-5 layer structure, in which the number of neurons in the first hidden layer is 2-3 times the number of neurons in the input layer, and the number of neurons in subsequent hidden layers decreases successively, with a decreasing ratio of 0.6-0.8; the number of neurons in the output layer is 6; Step 3: inputting the geological data as input data into the deep neural network algorithm for training to obtain a trained geostress field prediction model; Step 4: Use the trained geostress field prediction model to predict the geostress field of the underground cavern group of the hydropower project and output the prediction results.

2. A method for predicting geostress field of underground caverns in hydropower projects according to claim 1, characterized in that: The topographic data include elevation data of the region, ranging from -2000m to 3000m, slope data ranging from 0° to 90°, and topographic undulation characteristic data; the stratigraphic lithology data include elastic modulus of different rock layers ranging from 1GPa to 100GPa, Poisson's ratio ranging from 0.1 to 0.45, thickness data, and rock layer distribution data; the geological structure data include the location coordinates of the fault, with a strike range of 0° to 360°, an inclination data range of 0° to 90°, and the distribution density of joints and fissures ranging from 0.1 to 10, a length range of 0.1m to 10m, and an opening data range of 0.01mm to 100mm.

3. A method for predicting geostress field of underground caverns in hydropower projects according to claim 1, characterized in that: When constructing a deep neural network algorithm, the number of input layer neurons is set to correspond to the characteristic dimension of the geological data. The hidden layer adopts a multi-layer structure, and the number of hidden layer neurons is adjusted according to the complexity of the geological data and the prediction accuracy requirements. The number of output layer neurons corresponds to the number of stress components of the geostress field. The output calculation formula of the jth neuron in the lth layer of the deep neural network algorithm is: in is the output of the jth neuron in the lth layer, is the connection weight from the i-th neuron in the l-1th layer to the j-th neuron in the lth layer, is the output of the i-th neuron in the l-1th layer, is the bias of the jth neuron in the lth layer, is the number of neurons in the l-1th layer, f(x) is the activation function, and the ReLU function f(x)=max(0,x) is selected.

4. A method for predicting geostress field of underground caverns in hydropower projects according to claim 1, characterized in that: The training process of the deep neural network algorithm adopts the back propagation algorithm, and the square sum of the errors between the predicted ground stress value and the measured ground stress value is used as the loss function. The loss function formula is: Among them, m is the number of samples, is the i-th true geostress field parameter value of the k-th sample, is the i-th predicted geostress field parameter value of the k-th sample, θ is all the parameters of the neural network, and the loss function is minimized by adjusting the weights and thresholds of the neural network. The stochastic gradient descent method is used to optimize the update of weights and thresholds during the training process. The weight update formula is: Among them, α is the learning rate, and its value range is [0.001, 0.1].

5. A method for predicting geostress field of underground caverns in hydropower projects according to claim 1, characterized in that: Before using the trained geostress field prediction model for prediction, the model is verified by using a 5-10-fold cross-validation method. The collected geological data are divided into a training set, a validation set, and a test set. The model is trained using the training set, the model hyperparameters are adjusted using the validation set, and the prediction performance of the model is evaluated using the test set. When the model's prediction root mean square error (RMSE) is less than 1 MPa, formal geostress field prediction is performed.

6. A method for predicting geostress field of underground caverns in hydropower projects according to claim 1, characterized in that: After collecting geological data, the geological data is preprocessed. The preprocessing includes data cleaning, removing outliers and missing values ​​that exceed the reasonable range of data by 3 times the standard deviation, and filling the missing values ​​with the mean or median of the feature column; data standardization, using the Z-score standardization method to unify data of different dimensions into the same numerical range, the formula is: Among them, x std is the standardized value, x is the original value, μ is the mean of the feature column, and σ is the standard deviation of the feature column; data enhancement, by performing random rotation, translation, scaling and other transformation operations on the existing geological data, the transformation parameter range is rotation angle [-10°, 10°], translation distance [0, 100m], scaling ratio [0.9, 1.1], each transformation operation generates 2-5 times the amount of original data to expand the data volume.

7. A method for predicting geostress field of underground caverns in hydropower projects according to claim 1, characterized in that: The deep neural network algorithm introduces regularization terms during the training process. The regularization terms include L1 regularization and L2 regularization. The loss function formula after regularization is: Among them, λ is the regularization coefficient, and its value range is [0.001, 0.1].

8. A method for predicting geostress field of underground caverns in hydropower projects according to claim 1, characterized in that: After obtaining the prediction results of the geostress field, the prediction results are visualized, including: The topography, geological structure and geostress distribution of the underground cavern group area are presented in the form of three-dimensional images through three-dimensional modeling technology, which can intuitively show the changing laws of geostress fields under different geological conditions. The OpenGL 3D graphics library is used for 3D modeling. The visualization of the geostress field uses color mapping or arrows to represent the stress magnitude and direction. The color mapping can use a gradient from blue to red to represent the stress from small to large. The arrow length is proportional to the stress magnitude, and the arrow direction represents the stress direction.

9. A method for predicting geostress field of underground caverns in hydropower projects according to claim 1, characterized in that: In step 4, the design and construction of the underground cavern group are optimized according to the prediction results, including: Adjustment of cave shape: The radius of the circular cave is selected according to the predicted maximum principal stress in [5m, 50m] according to different stress value ranges; the length-to-width ratio of the rectangular cave is selected according to the angle between the predicted maximum principal stress direction and the long axis of the rectangle in [1:1, 1:5]; Cavern size and layout: The cavern size should be determined to meet engineering requirements and the distribution of the ground stress field, and the layout should minimize the stress interference between adjacent caverns; Support structure selection: Select different types of support structures such as anchor support, cable support or concrete lining according to the predicted maximum principal stress and ground stress distribution; Determination of anchor rod parameters: The tensile strength of the anchor rod is 1.2-1.5 times the predicted maximum principal stress. Its installation position is determined based on the stress concentration area. The stress concentration area is determined by the stress gradient, and then the installation range is determined based on the stress concentration coefficient.

10. A method for predicting geostress field of underground caverns in hydropower projects according to claim 1, characterized in that: When constructing the geostress field prediction model, physical constraints are integrated, and the physical constraints include the equilibrium equation of the rock mass: Constitutive relation: {σ}=[D]{ε} and boundary conditions, where [D] is the elastic matrix, {σ} is the stress vector, and {ε} is the strain vector. The residual sum of squares of the equilibrium equation is added to the loss function as part of the regularization term.

Images