A method and system for evaluating grouting effect of fissured surrounding rock of underground cavern

Abstract

The application discloses a kind of underground cavern fissured surrounding rock grouting effect evaluation method and system, method includes: obtaining the measurement-while-drilling parameter and surrounding rock image data of underground cavern construction reconnaissance stage;The measurement-while-drilling parameter is input to random forest model, and the uniaxial compressive strength of rock mass is obtained;The surrounding rock image data is input to improved Unet model, and the fissure quantification parameter is obtained;The improved Unet model integrates CBAM attention mechanism and KAN module;Grouting construction data is obtained, the uniaxial compressive strength of rock mass, the fissure quantification parameter and the grouting construction data are input to gradient boosting tree model, and unit consumption of ash amount is obtained, and the grouting effect of fissured surrounding rock is evaluated by the unit consumption of ash amount.The application can effectively improve the accuracy of unit consumption of ash amount and uniaxial compressive strength prediction by integrating learning model, the nonlinear relationship processing capacity of random forest and gradient boosting tree meta-learner.

Description

Technical Field

[0001] This invention belongs to the field of rock mass grouting construction technology in underground water-sealed cavern engineering, and particularly relates to a method for evaluating the grouting effect of fractured surrounding rock in underground caverns. Background Technology

[0002] Water-sealed oil cavern technology is a method of excavating underground spaces in rock masses below the groundwater level to store crude oil. This technology achieves a water seal by maintaining a water level higher than the petroleum products within the cavern, preventing oil leaks. It is widely popular due to its high safety, environmental friendliness, economy, and concealment. In underground engineering design and construction, drill-and-blast (DUB) and tunnel boring machines (TBMs) are two major methods for excavating underground rock masses. DUB is known for its flexibility and adaptability, suitable for underground caverns of various shapes and sizes. Before construction, key geological information is obtained through both contact acquisition (such as measurement-while-drilling parameters: drill pressure, torque, rotational speed, and drilling rate) and non-contact acquisition (such as using digital images to obtain information on the surrounding rock and face fractures of the excavated cavern). Furthermore, with the development of artificial intelligence technology, machine learning and deep learning methods have been used to predict rock mass mechanical parameters and grouting effects, as well as to identify and quantify rock mass fractures.

[0003] Despite the progress made in the design and construction of underground water-sealed caverns, existing technologies still have limitations. First, traditional drilling data processing methods rely on experience and qualitative judgment, lacking quantitative analysis tools, resulting in limited data processing capabilities and difficulty in achieving real-time updates and feedback of drilling data, thus limiting the accuracy and reliability of prediction results. Second, in non-contact acquisition methods, manually identifying and marking rock fracture information is inefficient, inaccurate, and labor-intensive. Furthermore, the grouting process is concealed, making it difficult to directly observe grout distribution, complicating the evaluation of construction quality and effectiveness. While numerical simulation methods can evaluate grouting effects, the modeling process is complex and time-consuming, unsuitable for the rapid demands of on-site construction. In summary, existing technologies in the construction exploration stage suffer from problems such as unclear rock transparency information, difficulty in evaluating rock grouting effects, limitations of numerical simulation methods, high complexity of prediction models, and a lack of real-time prediction and decision support. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a method for evaluating the grouting effect of fractured surrounding rock in underground caverns, thereby resolving the issues present in the existing technologies.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for evaluating the grouting effect of fractured surrounding rock in underground caverns, comprising:

[0006] Obtain drilling measurement parameters and surrounding rock image data during the underground cavern construction and exploration phase;

[0007] The drilling measurement parameters are input into the random forest model to obtain the uniaxial compressive strength of the rock mass;

[0008] The surrounding rock image data is input into the improved Unet model to obtain a binary image of the fracture; the improved Unet model integrates the CBAM attention mechanism and the KAN module; and the fracture quantization parameters are then obtained through the fracture geometric feature quantization ensemble algorithm.

[0009] Obtain grouting construction data, input the uniaxial compressive strength of the rock mass, the fissure quantification parameters, and the grouting construction data into the gradient boosting tree model to obtain the unit ash consumption, and evaluate the grouting effect of the fissure surrounding rock through the unit ash consumption.

[0010] Preferably, the measurement-while-drilling parameters include drilling speed, rotational speed, and rotational torque, and the surrounding rock image data includes the original surrounding rock image and its corresponding binarized fracture image.

[0011] Preferably, the drilling measurement parameters and the uniaxial compressive strength of the rock mass are integrated into a first dataset;

[0012] The surrounding rock images and their corresponding binary fracture images are integrated into a second dataset;

[0013] The uniaxial compressive strength of the rock mass, the quantitative parameters of fractures, and the grouting construction data are integrated into a third dataset.

[0014] Preferably, a random forest model is trained using the first dataset, wherein the random forest model is based on decision trees as base learners and integrates multiple decision trees through the Bagging method;

[0015] The gradient boosting tree model is trained using the third dataset. The gradient boosting tree model is based on decision trees as base learners and integrates multiple decision trees through the boosting method.

[0016] Preferably, an improved Unet model is trained using the second dataset, wherein the improved Unet model incorporates a CBAM module after each convolutional block and integrates a KAN module to identify crack features;

[0017] The improved Unet model uses a loss function that combines binary cross-entropy loss and Dice loss, and is trained using an optimizer.

[0018] Preferably, the process of inputting the surrounding rock image data into the improved Unet model includes:

[0019] The fracture features of the surrounding rock image data are extracted using an improved Unet model to obtain fracture quantification parameters, which include fracture length, width, area, and main direction.

[0020] The crack features were processed and analyzed using morphological closing operations, contour extraction, and skeleton extraction algorithms.

[0021] Preferably, the grouting construction data includes pre-grouting permeability, grouting hole sequence, grouting pressure, and elevation parameters.

[0022] Secondly, the present invention provides a system for evaluating the grouting effect of fractured surrounding rock in underground caverns, comprising:

[0023] The data acquisition module is used to acquire drilling measurement parameters and surrounding rock image data during the underground cavern construction and exploration phase.

[0024] The first prediction module is used to input the drilling measurement parameters into the random forest model to obtain the uniaxial compressive strength of the rock mass;

[0025] The second prediction module is used to input the surrounding rock image data into the improved Unet model to obtain a binary image of the fracture; the improved Unet model integrates the CBAM attention mechanism and the KAN module; and then obtains the fracture quantization parameters through the fracture geometric feature quantization ensemble algorithm.

[0026] The third prediction module is used to acquire grouting construction data. It inputs the uniaxial compressive strength of the rock mass, the fissure quantification parameters, and the grouting construction data into the gradient boosting tree model to obtain the unit ash consumption. The grouting effect of the fissure surrounding rock is evaluated by the unit ash consumption.

[0027] Thirdly, the present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0028] Fourthly, the present invention also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0029] Compared with the prior art, the present invention has the following advantages and technical effects:

[0030] This invention provides a method for evaluating the grouting effect of fractured surrounding rock in underground caverns. First, drilling measurement parameters and surrounding rock image data are obtained during the underground cavern construction and exploration phase. Second, the drilling measurement parameters are input into a random forest model to obtain the uniaxial compressive strength of the rock mass. Next, the surrounding rock image data is input into an improved Unet model to obtain a binary image of the fractures. The improved Unet model integrates the CBAM attention mechanism and the KAN module. Then, a fracture quantification parameter is obtained through a fracture geometric feature quantification integration algorithm. Finally, grouting construction data is obtained, and the uniaxial compressive strength of the rock mass, the fracture quantification parameter, and the grouting construction data are input into a gradient boosting tree model to obtain the unit ash consumption. The grouting effect of the fractured surrounding rock is evaluated based on the unit ash consumption.

[0031] This invention employs an ensemble learning-based prediction model. By combining the nonlinear relationship processing capabilities of random forests and gradient boosting tree learners, it effectively improves the accuracy of predicting unit ash consumption and uniaxial compressive strength. This ensemble method fully utilizes the advantages of different learners. Random forests, due to the randomness of their construction process, possess excellent robustness and are less prone to overfitting; gradient boosting trees, through iterative optimization, can adapt to various loss functions and achieve high accuracy.

[0032] This invention provides an efficient and accurate solution for evaluating rock mass information transparency and grouting effect in the field of underground engineering. It significantly enhances the accuracy and real-time performance of predictions, providing strong scientific support for engineering decision-making and construction safety. This invention significantly improves the performance of evaluating grouting effects, enabling rapid and accurate guidance for on-site construction decisions, ensuring the economic benefits and project stability of grouting. In actual grouting projects, this invention provides reliable methodological support for estimating the grouting volume in the area to be grouted, possessing significant engineering application value. Furthermore, due to the high requirements for controlling rock mass seepage in underground water-sealed cavern projects, the model is not only applicable to specific projects but also has the potential to be extended to other underground grouting projects. Attached Figure Description

[0033] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0034] Figure 1 This is a flowchart of a method according to an embodiment of the present invention;

[0035] Figure 2 This is a technical roadmap of an embodiment of the present invention;

[0036] Figure 3 This is a diagram of the crack identification model architecture according to an embodiment of the present invention. Detailed Implementation

[0037] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0038] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0039] Example 1

[0040] like Figure 1-2 As shown, this embodiment provides a method for evaluating the grouting effect of fractured surrounding rock in underground caverns, including:

[0041] S1. Obtain drilling measurement parameters and surrounding rock image data during the underground cavern construction and exploration phase;

[0042] Furthermore, the measurement-while-drilling parameters include drilling speed, rotational speed, and rotational torque, and the surrounding rock image data includes the original surrounding rock image and its corresponding binarized fracture image.

[0043] Specifically, parameters such as drilling speed, rotational speed, and rotational torque are obtained through measurement-while-drilling (MWD) technology.

[0044] S2. Input the drilling measurement parameters into the random forest model to obtain the uniaxial compressive strength of the rock mass;

[0045] As an innovative implementation method, the drilling measurement parameters and the uniaxial compressive strength of the rock mass are integrated into a first dataset; a random forest model is trained using the first dataset, wherein the random forest model is based on decision trees as base learners and integrates multiple decision trees through the Bagging method;

[0046] Specifically, the random forest model first constructs multiple decision trees, each trained based on different random samples and feature subsets to reduce the correlation between trees. For a single decision tree, its output is the average value of that node, assigned to the input data x.

[0047]

[0048] Among them, w i R is the weight of the leaf node. i The region is the leaf node, I(x∈R) i ) is an indicator function.

[0049] The overall prediction result of a random forest is the average of the predictions from all decision trees, i.e.

[0050]

[0051] Where M is the number of decision trees.

[0052] S3. Input the surrounding rock image data into the improved Unet model to obtain a binary image of the fracture; the improved Unet model integrates the CBAM attention mechanism and the KAN module; then, the fracture quantization parameters are obtained through the fracture geometric feature quantization ensemble algorithm.

[0053] As an innovative implementation method, the surrounding rock image and its corresponding binarized fracture image are integrated into a second dataset; an improved Unet model is trained using the second dataset, wherein the improved Unet model adds a CBAM module after each convolutional block and integrates a KAN module to identify fracture features;

[0054] Specifically, the improved Unet model integrates the CBAM and KAN modules to enhance its ability to identify rock mass fracture features. Specifically, the CBAM module is added after each convolutional block of the Unet model, enhancing the model's recognition ability through channel focusing and spatial attention mechanisms; the KAN module is further integrated based on the CBAM module, improving the model's accuracy in identifying fracture features in complex backgrounds through kernel attention mechanisms.

[0055] The parameter configuration step includes defining key parameters for model training to ensure the stability and tunability of the model. Specific parameters include the number of training epochs, batch size, optimizer type, learning rate (Lr), and weight decay. To ensure the reproducibility of the experiment, a random seed is set to ensure consistency in data loading and model initialization. This step involves specifying a fixed random seed value to initialize the random number generator.

[0056] Furthermore, the loss function of the improved Unet model is a combination of binary cross-entropy loss and Dice loss, and the model is trained using an optimizer.

[0057] Specifically, based on the parameter configuration, a loss function is defined to quantify the difference between the model's predictions and the actual results. This includes selecting the Binary Cross-Entropy Loss (BCEWithLogitsLoss), a loss function that combines the Binary Cross-Entropy Loss (BCE Loss) and the Dice Loss, commonly used in image segmentation tasks. The purpose of this loss function is to combine the advantages of both loss functions to improve model performance. BCE Loss is a loss function used for binary classification problems, and its mathematical formula is:

[0058]

[0059] In the formula, N is the number of samples, y i It is the actual label of the i-th sample, with a value of 0 or 1. This is the predicted value of the i-th sample, ranging from (0,1). This loss function measures the difference between the probability distribution predicted by the model and the true label.

[0060] Dice Loss transforms the Dice coefficients into a loss form:

[0061]

[0062] In the formula, X and Y represent the predicted result and the ground truth label, respectively; |X∩Y| is the intersection area of ​​the predicted bounding box X and the ground truth bounding box Y; |X| is the area of ​​the predicted bounding box X; and |Y| is the area of ​​the ground truth bounding box Y. This loss function focuses on the overlapping area between the predicted region and the ground truth region.

[0063] BCEDiceLoss combines the two loss functions mentioned above, and its calculation method is usually a weighted sum of the two losses:

[0064] BCEDiceLoss=α×BCE+β×DiceLoss

[0065] In the formula, α and β are weighting parameters used to balance the influence of the two loss functions. In practical applications, these weights can be adjusted according to the specific needs of the task.

[0066] Based on the parameter configuration, an image segmentation model is created. The model structure includes an encoder, a CBAM attention mechanism, a KAN module, and a decoder. Specifically, it includes convolutional layers for extracting image features, a self-attention mechanism module for further extracting and fusing features, and a decoder for mapping the features extracted by the encoder and KAN module back to the image space for pixel-level prediction.

[0067] Furthermore, based on the parameter configuration, optimizers and learning rate schedulers are set to dynamically adjust the learning rate and optimize the model training process. This specifically includes selecting optimizers such as Adam or SGD, and setting learning rate schedulers such as CosineAnnealingLR or ReduceLROnPlateau.

[0068] Furthermore, the Albumentations library is used to preprocess the data, including random rotation, flipping, and normalization. The dataset is divided into training and validation sets. Specifically, this includes image enhancement and normalization to improve the model's generalization ability, and dividing the dataset into training and validation sets for model training and performance evaluation.

[0069] Furthermore, the loss is calculated using BCEDiceLoss, and the model parameters are updated.

[0070] Furthermore, for each training epoch, the following steps are performed: train the model using the training data, calculate the loss, and update the model parameters; evaluate the model performance on the validation set, and record the loss and IoU metric; adjust the learning rate according to the learning rate scheduler's strategy; if the IoU metric on the validation set improves, save the current best model. The formula for calculating IoU is as follows:

[0071]

[0072] In the formula, |X∩Y| represents the intersection area of ​​the predicted bounding box X and the ground truth bounding box Y; |X∪Y| represents the union area of ​​the two bounding boxes.

[0073] Furthermore, if the IoU metric on the validation set does not improve over multiple consecutive epochs, an early stopping mechanism is triggered, ending the training process. This involves setting a threshold to determine when to trigger the early stopping mechanism.

[0074] Furthermore, the loss, IoU metric, and performance on the validation set during training are recorded in a log file and visualized using TensorboardX. This includes recording key metrics during training to facilitate subsequent analysis and model selection, and using tools like TensorboardX to visualize the training process and intuitively demonstrate changes in model performance.

[0075] Furthermore, the process of inputting the surrounding rock image data into the improved Unet model includes:

[0076] The fracture features of the surrounding rock image data are extracted using an improved Unet model to obtain fracture quantification parameters, which include fracture length, width, area, and main direction.

[0077] The crack features were processed and analyzed using morphological closing operations, contour extraction, and skeleton extraction algorithms.

[0078] As an innovative implementation method, this embodiment implements the skeleton extraction step through the following algorithm: morphological processing, including erosion and dilation operations, to process the image segmentation results to remove noise points and highlight the continuity of the cracks, followed by applying skeletonization algorithms such as iterative erosion to extract the center line of the cracks, and using a crack contour extraction algorithm to extract the contour.

[0079] As an innovative implementation method, the geometric feature quantization steps include: length measurement, which calculates the total length of the crack by summing the distances between points on the crack skeleton; tilt angle calculation, which calculates the tilt angle of the crack relative to the horizontal plane using the endpoint coordinates of the crack skeleton; and area and width statistics, which obtain the area by counting the number of pixels in the crack region and calculate the average width by measuring the width variation of the skeleton.

[0080] This embodiment innovatively combines the skeleton extraction step, the geometric feature quantization step, and the CBAM-UKAN model to form a technical solution for automated extraction and quantification of cracks.

[0081] This embodiment utilizes advanced deep learning technology to automate the extraction of crack features and optimizes the model training process, ensuring the reproducibility of experimental results. Through detailed performance evaluation and intuitive visualization tools, this embodiment not only improves the efficiency of model training but also effectively avoids overfitting.

[0082] In this embodiment, rock fractures are a key structural feature, manifested as measurable line segments between discontinuous surfaces on the rock mass. Accurately grasping the quantitative characteristics of rock joints and fractures is crucial for determining the surrounding rock grade and setting construction parameters. These characteristics include trace length, dip angle, width, and area. Traditionally, drawing fracture trace diagrams relies on engineers' on-site visual observation and contact tools, which is not only inefficient but also poses safety risks and is easily influenced by personal experience and perspective, leading to judgment biases.

[0083] With advancements in computer vision and deep learning technologies, data sample collection and information analysis methods have been significantly improved. Deep learning networks can achieve end-to-end feature extraction, eliminating the tedious step of manually adjusting pixel thresholds compared to traditional image processing methods (such as Otsu's algorithm, Canny's algorithm, and Laplacian algorithm), resulting in simpler operation and higher accuracy.

[0084] Please see Figure 3 This embodiment provides a fracture identification model architecture diagram to improve the accuracy and efficiency of fracture detection in diverse rock masses and complex background environments. The specific implementation steps are as follows:

[0085] 1. Model Initialization

[0086] Create an instance of the CBAM-UKAN class and set the parameters as follows:

[0087] Input_channels: 3, because the input is an RGB image.

[0088] Img_size: 512, consistent with the input image size.

[0089] Patch_size: 16, determined according to the design of Patch Embedding.

[0090] Embed_dims: [256,320,512], defines the embedding dimensions at different stages.

[0091] Drop_rate: 0.1, sets the dropout rate to prevent overfitting.

[0092] Drop_path_rate: 0.1, sets the drop path rate for regularization.

[0093] Norm_layer: Uses nn.LayerNorm as the normalization layer.

[0094] 2. Data Preparation

[0095] Collect image data of the rock tunnel working face, ensuring that each image is 512×512 pixels in size. Preprocess the images, including: normalization: scaling pixel values ​​to the range [0,1]. Data augmentation: including random rotation, flipping, etc., to increase the model's generalization ability.

[0096] 3. Data Loading

[0097] Use DataLoader to load the preprocessed image data in batches, setting an appropriate batch size of Batchsize=32.

[0098] 4. Model Training

[0099] Define the loss function as cross-entropy loss, select Adam as the optimizer, and set the learning rate to 0.001.

[0100] The model is trained by iterating through 500 specified epochs. In each epoch: image data is input into the UKAN model. The output of the loss function is calculated and backpropagation is performed. The model weights are updated. The model performance is evaluated on the validation set, and the weights of the best-performing model are saved.

[0101] 5. Feature Extraction and Attention Applications

[0102] In the encoder stage, features are extracted and their representations are enhanced using the ConvLayer and CBAM modules. In the KAN module, features are further extracted, with the KANLayer utilizing the characteristics of Kolmogorov-Arnold Networks for nonlinear transformation.

[0103] 6. Decoding and Classification

[0104] In the decoder stage, the spatial resolution of the image is gradually restored and the feature representation is enhanced through the D_ConvLayer and CBAM modules. Finally, classification is performed through a final layer, using nn.Conv2d for 1x1 convolution to convert the feature map into a class probability map.

[0105] 7. Model Evaluation

[0106] The model performance was evaluated on an independent test set using metrics such as accuracy, recall, and IoU. The prediction results were visualized by comparing the predicted fracture trajectories with the actual fracture trajectories to qualitatively analyze the model performance.

[0107] 8. Closing operation filling:

[0108] After crack identification, the `close_skeleton_contour` function is used to perform a closing operation on the image, filling small holes and connecting small blocks to enhance the continuity of the crack traces. Closing is a morphological operation used to fill small holes inside foreground objects and connect adjacent foreground objects. Closing involves first performing dilation and then erosion. Mathematically, closing can be represented as:

[0109]

[0110] In the formula: A is the original image; ⊕ represents the dilation operation; C is the structuring element, which is usually a matrix that defines the shape and size of the dilation and erosion operations; B represents the erosion operation; B is the image after the closing operation.

[0111] 9. Contour Extraction:

[0112] The `Find_contours` function is used to extract contours from the processed image. Contour extraction is the process of identifying object boundaries in an image. Mathematically, a contour can be considered as a local maximum of the image gradient (a change in image intensity). The formula for contour extraction can be expressed as:

[0113] Contours=Find_contours(I,level)

[0114] In the formula: I is the processed image; level is the threshold used to determine which gradient values ​​are considered part of the contour; Contours is the set of extracted contours. The Find_contours function returns the coordinates of all contours in the image.

[0115] 10. Skeleton Extraction:

[0116] The Skeletonize function is used to skeletonize the image, extracting the skeleton of the crack traces. Skeleton extraction simplifies binary objects in an image (usually the foreground in a binary image) into a skeleton of one pixel width. The mathematical principle of skeleton extraction is based on iteratively removing pixels on the object's boundaries until the object completely disappears or reaches a single pixel width. The Zhang-Suen algorithm is a popular skeleton extraction algorithm, and its iterative process can be represented as:

[0117] S = {p∈A | Thinning(p) = 1}

[0118] In the formula: A is the original binary image; S is the skeleton image; Thinning(p) is a function that determines whether to remove p based on pixel p and its neighborhood.

[0119] 11. Crack trace feature extraction:

[0120] The length of the crack trace is calculated using the Calculate_crack_length function.

[0121] The width of the crack trace is calculated using the Calculate_crack_width function.

[0122] The direction of the crack trace is estimated using the Estimate_crack_direction function.

[0123] The area of ​​the crack trace is extracted using the Extract_crack_areas function.

[0124] S4. Obtain grouting construction data. Input the uniaxial compressive strength of the rock mass, the fissure quantification parameters, and the grouting construction data into the gradient boosting tree model to obtain the unit ash consumption. Evaluate the grouting effect of the fissure surrounding rock through the unit ash consumption.

[0125] As an innovative implementation method, the uniaxial compressive strength of the rock mass, fracture quantification parameters, and grouting construction data are integrated into a third dataset. A gradient boosting tree model is trained using this third dataset. This model is based on decision trees as base learners and integrates multiple decision trees through a boosting method.

[0126] The grouting construction data includes pre-grouting permeability, grouting hole sequence, grouting pressure, and elevation parameters.

[0127] Specifically, the parameters for constructing the grouting effect evaluation dataset include: obtaining corresponding uniaxial compressive strength, fracture trace length, width, dip angle, area, pre-grouting permeability, grouting pressure, hole sequence, and elevation parameters by summarizing drilling parameters, surrounding rock images, and grouting construction record tables through machine learning and deep learning methods. The output parameter is the unit ash consumption. The established dataset is randomly divided into training set and validation set according to a certain proportion.

[0128] Specifically, the gradient boosting tree model employs a stepwise optimization approach, starting with the initial model F0(x) and iteratively training the decision tree to gradually fit the residuals. In the m-th step, model F... m (x) is the model F from the previous step. m-1 (x) Add the newly trained decision tree h m (x) multiplied by the learning rate γ m The result, namely

[0129] F m (x)=F m-1 (x)+γ m h m (x)

[0130] After M iterations, the final model is

[0131]

[0132] In each iteration, the parameters θ of the next decision tree are determined by minimizing the loss function L. m ,Right now

[0133]

[0134] Where L is the mean square error.

[0135] Furthermore, the performance of the optimized ensemble learning model in evaluating grouting effects was analyzed using performance evaluation metrics, including the coefficient of determination (R²). 2 ), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAE).

[0136]

[0137] This embodiment provides a systematic and repeatable method for making underground engineering rock mass information transparent and evaluating the effect of rock mass grouting, which can effectively improve the accuracy and real-time performance of rock mass feature prediction.

[0138] This embodiment extracts and evaluates fracture trace information at the working face of a rock tunnel, improving the accuracy and efficiency of fracture detection in different rock masses and complex environments. By combining drilling parameters, it makes the rock mass information clear and transparent from the inside out. Through the implementation of this embodiment, fracture characteristics at the working face of a rock tunnel can be accurately and efficiently identified and analyzed without human intervention, providing a scientific basis for optimizing grouting construction techniques.

[0139] Example 2

[0140] Based on the same inventive concept, this embodiment also provides a system for evaluating the grouting effect of fractured surrounding rock in underground caverns, including:

[0141] The data acquisition module is used to acquire drilling measurement parameters and surrounding rock image data during the underground cavern construction and exploration phase.

[0142] The first prediction module is used to input the drilling measurement parameters into the random forest model to obtain the uniaxial compressive strength of the rock mass;

[0143] The second prediction module is used to input the surrounding rock image data into the improved Unet model to obtain a binary image of the fracture; the improved Unet model integrates the CBAM attention mechanism and the KAN module; and then obtains the fracture quantization parameters through the fracture geometric feature quantization ensemble algorithm.

[0144] The third prediction module is used to acquire grouting construction data. It inputs the uniaxial compressive strength of the rock mass, the fissure quantification parameters, and the grouting construction data into the gradient boosting tree model to obtain the unit ash consumption. The grouting effect of the fissure surrounding rock is evaluated by the unit ash consumption.

[0145] The grouting effect evaluation system for fractured surrounding rock in underground caverns provided in this embodiment has all the advantages of the grouting effect evaluation method for fractured surrounding rock in underground caverns provided in Embodiment 1.

[0146] Example 3

[0147] This embodiment also discloses a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0148] Example 4

[0149] This embodiment also discloses a computer program product, including a computer program that, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0150] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for evaluating the grouting effect of fractured surrounding rock in underground caverns, characterized in that, Includes the following steps: Obtain drilling measurement parameters and surrounding rock image data during the underground cavern construction and exploration phase; The drilling measurement parameters are input into the random forest model to obtain the uniaxial compressive strength of the rock mass; The surrounding rock image data is input into the improved Unet model to obtain a binary fracture image. The improved Unet model integrates the CBAM attention mechanism and the KAN module. Then, the fracture geometric feature quantization processing is performed on the fracture binary image to obtain fracture quantization parameters. The original surrounding rock image and its corresponding binary fracture image are integrated into a second dataset, and the improved Unet model is trained using the second dataset. Obtain grouting construction data, input the uniaxial compressive strength of the rock mass, the fissure quantification parameters, and the grouting construction data into the gradient boosting tree model to obtain the unit ash consumption, and evaluate the grouting effect of the fissure surrounding rock through the unit ash consumption.

2. The method according to claim 1, characterized in that, The drilling measurement parameters include drilling speed, rotational speed, and rotational torque, and the surrounding rock image data includes the original surrounding rock image and its corresponding binary fracture image.

3. The method according to claim 1, characterized in that, The drilling measurement parameters and the uniaxial compressive strength of the rock mass are integrated into a first dataset; The uniaxial compressive strength of the rock mass, the quantitative parameters of fractures, and the grouting construction data are integrated into a third dataset.

4. The method according to claim 3, characterized in that, A random forest model is trained using the first dataset. The random forest model is based on decision trees as base learners and integrates multiple decision trees through the Bagging method. The gradient boosting tree model is trained using the third dataset. The gradient boosting tree model is based on decision trees as base learners and integrates multiple decision trees through the boosting method.

5. The method according to claim 3, characterized in that, The improved Unet model adds a CBAM module after each convolutional block and integrates a KAN module to identify crack features; The improved Unet model uses a loss function that combines binary cross-entropy loss and Dice loss, and is trained using an optimizer.

6. The method according to claim 1, characterized in that, After inputting the surrounding rock image data into the improved Unet model, the following steps are also included: The binary image of the crack is subjected to crack geometric feature quantization processing to obtain crack quantization parameters, wherein the crack quantization parameters include crack length, width, area and main direction; The quantization of the crack geometric features includes the use of morphological closing operations, contour extraction, and skeleton extraction algorithms.

7. The method according to claim 1, characterized in that, The grouting construction data includes pre-grouting permeability, grouting hole sequence, grouting pressure, and elevation parameters.

8. A system for evaluating the grouting effect of fractured surrounding rock in underground caverns, characterized in that, include: The data acquisition module is used to acquire drilling measurement parameters and surrounding rock image data during the underground cavern construction and exploration phase. The first prediction module is used to input the drilling measurement parameters into the random forest model to obtain the uniaxial compressive strength of the rock mass; The second prediction module is used to input the surrounding rock image data into the improved Unet model to obtain a binary fracture image. The improved Unet model integrates the CBAM attention mechanism and the KAN module. Then, the fracture geometric feature quantization processing is performed on the fracture binary image to obtain fracture quantization parameters. The original surrounding rock image and its corresponding binary fracture image are integrated into a second dataset, and the improved Unet model is trained using the second dataset. The third prediction module is used to acquire grouting construction data. It inputs the uniaxial compressive strength of the rock mass, the fissure quantification parameters, and the grouting construction data into the gradient boosting tree model to obtain the unit ash consumption. The grouting effect of the fissure surrounding rock is evaluated by the unit ash consumption.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1-7.

Images