Multi-dimensional dynamic deformation monitoring method in pumping and storage tunnel excavation process

Abstract

The invention relates to the technical field of constructional engineering, in particular to a multi-dimensional dynamic deformation monitoring method in the excavation process of a pumped storage tunnel. According to the method, data processing is carried out on a pumping and storage tunnel image and corresponding three-dimensional point cloud data, integration is carried out, and a geological sketch and a deformation monitoring report are generated; a convolutional neural network is adopted to identify a geological sketch in a pumping and storage tunnel image, and the training process of the network comprises the following steps: automatically extracting key features in the pumping and storage tunnel image by using a convolutional layer in the convolutional neural network, and adding DCNv2, DSConv and CBAM modules on the basis of VGGNet and yolov8; a large number of marked image data sets are used for training, and model parameters are optimized through a back propagation algorithm to minimize a loss function. According to the invention, high-precision photographing and laser scanning of the rock wall surface are realized, a geological sketch is formed, real-time deformation monitoring is carried out on the top arch and the side wall of the tunnel, and the construction safety and quality are ensured.

Description

Technical Field

[0001] The present invention relates to the field of construction engineering technology, and in particular to a multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel. Background Art

[0002] With the increasing construction of pumped-storage power plants, the safety and efficiency of tunnel excavation projects have become a key concern. During tunnel excavation, accurate assessment of the geological conditions of the rock face and effective monitoring of tunnel structure deformation are crucial.

[0003] Traditional monitoring methods typically rely on manual inspection, which is not only inefficient but also difficult to achieve continuous monitoring. Furthermore, manual inspections pose safety risks, especially in complex geological conditions. Therefore, developing a system that can automatically and continuously monitor geological changes and structural deformation during tunnel excavation is of great significance. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel. The technical solution adopted is as follows:

[0005] Obtain pumped storage tunnel images and corresponding 3D point cloud data;

[0006] The pumped storage tunnel image and the corresponding three-dimensional point cloud data are processed and integrated to generate a geological sketch and deformation monitoring report. A convolutional neural network is used to automatically identify the geological sketch in the pumped storage tunnel image. The convolutional neural network training process includes: normalizing and denoising the pumped storage tunnel image; using the convolutional layer in the convolutional neural network to automatically extract key features in the pumped storage tunnel image, the key features including edges and textures; adding DCNv2, DSConv, and CBAM modules based on VGGNet and yolov8; using a large number of annotated image datasets for training, and optimizing model parameters through a backpropagation algorithm to minimize the loss function;

[0007] The geological sketch includes the geological characteristics and crack distribution of the rock wall of the pumping tunnel; the deformation monitoring report includes the deformation conditions and deformation rate of the top arch and side walls of the pumping tunnel; based on the geological sketch and the deformation monitoring report, the stress and deformation conditions of the rock are analyzed; the stress and deformation conditions of the rock include the geological conditions of the rock wall, the deformation conditions of the top arch and the side walls.

[0008] Furthermore, the pumped storage tunnel image is used to generate a geological sketch, and the three-dimensional point cloud data is used to generate a rock deformation monitoring report.

[0009] Furthermore, the geological features of the rock wall of the pumped storage tunnel in the geological sketch include loose rock.

[0010] Furthermore, the convolutional neural network is used to automatically identify geological sketches in pumped storage tunnel images. The training process of the convolutional neural network also includes a validation and testing phase: using an independent data set to evaluate the performance of the model;

[0011] During the training phase, the amount of data is increased, data enhancement technology is introduced, and transfer learning methods are used; integrated learning methods such as Bagging and Boosting are adopted to combine the prediction results of multiple models.

[0012] Furthermore, the calculation formula of the output eigenvalue y(p0) of DCNv2 is as follows:

[0013]

[0014] Where Δp n , Δm n are the learnable offset and modulation scalar of the nth position respectively; p0 is the input feature value of DCNv2; x(p0+p n +Δp n ) is the input feature map x at position p0+p n +Δp n The value at p n ∈R is the position offset p of all the positions belonging to the set R n .

[0015] Furthermore, the ARIMA model is used to predict the deformation trend that may occur in the future. Its basic form is:

[0016] X t =c+φ1X t-1 +…+φ p X t-p +θ1∈ t-1 +…+θ q ∈ t-q +∈ t ;

[0017] Among them, X t is the observed value at time point t; c is a constant term; φ i and θ j are the autoregressive and moving average parameters, ∈ t It is white noise;

[0018] By fitting the historical displacement data, the model parameters are determined, and the displacement changes in the future time period are predicted using the model parameters.

[0019] The embodiments of the present invention have at least the following beneficial effects:

[0020] High-precision monitoring: Laser and visual scanners and mobile sensors enable high-precision photography and laser scanning of rock faces, as well as high-precision deformation monitoring of the top arch and side walls.

[0021] Real-time monitoring: The sensor can move continuously, collect data in real time, and generate deformation monitoring reports to ensure safety during construction.

[0022] Multi-dimensional monitoring: By installing sensors on the top arch and side walls, multi-dimensional deformation monitoring is achieved to ensure the breadth and accuracy of the monitoring range.

[0023] Flexible adjustment: The sensor can adjust its position according to the excavation progress, expand the monitoring range, and adapt to the construction needs of different stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0025] Figure 1 A flowchart of a method for multi-dimensional dynamic deformation monitoring during excavation of a pumped storage tunnel provided by one embodiment of the present invention;

[0026] Figure 2 A schematic diagram of a pumped storage tunnel provided by one embodiment of the present invention;

[0027] Figure 3 Another schematic diagram of a pumped storage tunnel provided by one embodiment of the present invention;

[0028] Figure 4 An embodiment of the present invention provides Figure 3 Another visual diagram of the pumped storage tunnel shown;

[0029] Figure 5 Another schematic diagram of a pumped storage tunnel provided in accordance with an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the following, in conjunction with the accompanying drawings and preferred embodiments, describes in detail the specific implementation method, structure, features and effects of a multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel proposed by the present invention. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics of one or more embodiments may be combined in any suitable form.

[0031] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0032] The following describes in detail a specific scheme of a multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel provided by the present invention with reference to the accompanying drawings.

[0033] See also Figure 1 , which shows a flowchart of a multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel provided by one embodiment of the present invention, the method comprising the following steps:

[0034] Step S100: Acquire a pumped storage tunnel image and corresponding three-dimensional point cloud data.

[0035] Tracks are installed on the roof and sidewalls of the pumped storage tunnel to ensure stability and safety. High-strength materials, such as basalt, should be used to support the weight of the scanner and sensors. Specifically, a straight track is laid on the top of the pumped storage tunnel using high-strength materials to ensure it can withstand the weight of the laser and visual scanners and their associated components. The track length should be determined based on the actual tunnel length and monitoring requirements, and the straightness error must not exceed ±5mm / 10m to ensure scanning accuracy. After installation, the track is calibrated using a high-precision level to ensure that the track meets the required levelness.

[0036] A track-mounted laser and visual scanner is installed on the track. This scanner can move along the track to photograph and laser scan the rock face of the pumped storage tunnel. The scanner should be equipped with a high-resolution camera and laser scanning module to obtain clear images and accurate scan data. The scanner transmits the collected data to a data processing unit, which generates a geological sketch and deformation monitoring report. The data processing unit executes step S200.

[0037] In this embodiment of the present invention, a device equipped with a high-resolution camera and a laser scanning module can be selected, such as a laser and visual scanner with a 30-megapixel high-resolution camera and a laser scanning module with an accuracy of ±1mm. The scanner is mounted on a track and can slide freely on the track to cover the entire area to be measured. The scanner's operating mode is set to timed scanning, with a complete scan performed once every hour. The data collected by the scanner includes image information and 3D point cloud data.

[0038] As construction progresses, sensors can accurately locate new geological layers as the pumped storage tunnel reaches them. A relative coordinate system is established, ensuring effective comparison and analysis of sensor data at different locations. Mobile measurements of horizontal and vertical deformation provide real-time recording of tunnel deformation.

[0039] As the tunnel advances through new geological formations, the laser and vision scanners are repositioned to ensure coverage of the newly excavated area. For example, similar mobile sensors are installed on the sidewalls. Sidewall sensors also use laser and vision technology, but may have different lens focal lengths to accommodate close-up imaging. Furthermore, to accommodate changing geological conditions, such as new tunnel widths and depths, the track layout may need to be adjusted or new rails added to ensure the scanners can reach all key observation points.

[0040] After tunnel excavation, continuous deformation monitoring is performed. The scanner moves along the track at preset intervals, performing a complete scan each time it moves to a new location. For example, a scan is performed every four hours to ensure data continuity and timeliness.

[0041] After each scan, the collected image data and 3D point cloud data are transmitted to the data processing unit. The image data is used to generate geological sketches, while the 3D point cloud data is used to calculate the displacement changes of the rock surface.

[0042] By comparing the results of multiple scans, we can analyze the specific changes of rocks over time.

[0043] See also Figure 2 , Figure 2 A schematic diagram of a pumped storage tunnel. Figure 2 The lower portion of the middle pumped storage tunnel has not yet been excavated. Slide rails, laser scanners, and visual scanners have been installed at the top of the tunnel to detect rock joints and analyze lithology. These scanners also record horizontal and vertical displacements.

[0044] See also Figure 3 , Figure 3 This is another schematic diagram of the pumped storage tunnel. Figure 3The lower part of the pumped storage tunnel has been excavated, and the laser and visual scanners are periodically sliding on the track for identification. The deformation is analyzed through the results of several scans, thereby analyzing the stress and deformation of the rock. Figure 4 , Figure 4 for Figure 3 Another visual schematic diagram of the pumped storage tunnel is shown, used to demonstrate the position information of the slide rail and the scanner.

[0045] See also Figure 5 , Figure 5 This is another schematic diagram of a pumped storage tunnel, showing a special case where the lower part of the pumped storage tunnel has been excavated. In order to adapt to different geological environments, such as tunnel width and depth, it may be necessary to adjust the track layout or add new slide rails to ensure that the scanner can reach all key observation points. Figure 5 In the figure, A is the pumped storage tunnel under excavation, B is the laid track, and C is the movable track-mounted sensor.

[0046] Step S200: Process and integrate the pumped storage tunnel image and the corresponding three-dimensional point cloud data to generate a geological sketch and a deformation monitoring report; use a convolutional neural network to automatically identify the geological sketch in the pumped storage tunnel image, and the training process of the convolutional neural network includes: normalizing and denoising the pumped storage tunnel image; using the convolution layer in the convolutional neural network to automatically extract key features in the pumped storage tunnel image, and the key features include: edges and textures; adding DCNv2, DSConv, and CBAM modules on the basis of VGGNet and yolov8; using a large number of annotated image data sets for training, and optimizing model parameters through the back propagation algorithm to minimize the loss function.

[0047] The geological sketch includes the geological characteristics and crack distribution of the rock wall of the pumping tunnel; the deformation monitoring report includes the deformation conditions, deformation rate and other information of the top arch and side walls of the pumping tunnel; based on the geological sketch and the deformation monitoring report, the stress and deformation conditions of the rock are analyzed; the stress and deformation conditions of the rock include the geological conditions of the rock wall, the deformation conditions of the top arch and the side walls.

[0048] Scanners and sensors transmit collected data wirelessly or wired to a data processing unit, which then processes the data. High-bandwidth, low-latency communication technologies should be used to ensure real-time data integrity. The data processing unit should possess robust data processing capabilities, enabling it to rapidly process large amounts of data and generate real-time monitoring reports.

[0049] The data processing unit receives data from the scanner and sensors, executes the operations in step S200, and performs preliminary processing, such as noise filtering and data correction. It then integrates the processed data to generate a geological sketch and deformation monitoring report. The geological sketch includes information such as the geological characteristics of the rock face and crack distribution of the pumped-storage tunnel; the deformation monitoring report includes information such as the deformation status and deformation rate of the tunnel's crown and sidewalls. The data processing unit possesses real-time analysis capabilities, enabling it to promptly identify potential safety hazards, such as abnormal deformation and crack expansion, and generate alarms.

[0050] Pumped storage tunnel images are used to generate geological sketches, and computer vision technology is used to identify rock joints, cracks and other features; three-dimensional point cloud data is used to generate rock deformation monitoring reports and analyze the deformation of the rock wall surface.

[0051] The calculation formula for deformation is as follows:

[0052] Horizontal displacement change Δx: Δx = x new -x old ;

[0053] Vertical displacement change Δy: Δy = y new -y old ;

[0054] Among them, x new Indicates the horizontal coordinate of the new position; y new Indicates the vertical coordinate of the new position; x old Indicates the horizontal coordinate of the old position coordinate; y old The horizontal coordinate representing the old position coordinate;

[0055] The deformation rate can be obtained by dividing the displacement change in consecutive time intervals by the time difference, that is, v is the deformation rate; Δd ​​is the displacement change; and Δt is the time difference. The displacement change can be either horizontal or vertical.

[0056] If abnormal deformation or crack expansion exceeding the warning line is detected based on preset safety thresholds, an alarm system will immediately trigger, alerting construction personnel to take appropriate measures. Specifically, the alarm system will be triggered when the deformation rate exceeds the preset deformation safety threshold; similarly, the alarm system will be triggered when the change in horizontal or vertical displacement exceeds the preset crack safety threshold. The preset deformation and crack safety thresholds can be set by the implementer based on actual experience and site conditions.

[0057] As tunnel excavation progresses, the existing monitoring system continues to function and needs to be appropriately adjusted according to the new geological conditions and construction progress, and the data processing steps are also updated accordingly.

[0058] Machine learning algorithms such as convolutional neural networks (CNNs) are used to automatically identify potential hazardous factors in pumped storage tunnel images, such as cracks and loose rock blocks. Therefore, it can also be understood as using convolutional neural networks to automatically identify geological sketches in pumped storage tunnel images, and improving recognition accuracy through deep learning models.

[0059] The model training process includes: data preprocessing, feature extraction, model construction, training phase, verification and testing phases;

[0060] (1) Data preprocessing stage: The pumped storage tunnel images are normalized and denoised to improve the generalization ability of the model.

[0061] (2) Feature extraction stage: Use the convolutional layer in the convolutional neural network to automatically extract the key features in the pumped tunnel image, including edges, textures, etc.

[0062] (3) Model construction phase: DCNv2, DSConv, and CBAM modules are added based on VGGNet and yolov8.

[0063] The calculation formula for the output eigenvalue y(p0) of DCNv2 is as follows:

[0064]

[0065] Where Δp n , Δm n are the learnable offset and modulation scalar of the nth position respectively; p0 is the input feature value of DCNv2; x(p0+p n +Δp n ) is the input feature map x at position p0+p n +Δp n The value at p n ∈R is the position offset p of all the positions belonging to the set R n Δp n Improved the fitting ability for irregular shaped objects, Δm n Irrelevant background information is suppressed to distinguish whether the introduced area is the area of ​​interest in this article, so that the model can reduce the interference of irrelevant content during feature learning.

[0066] Due to the dynamic structure, DSConv can cover two-dimensional changes (x-axis and y-axis) within the range of 9×9. It better adapts to slender tubular structures and thus perceives key features more effectively. Consider a convolution kernel of size 9. Taking the x-axis direction as an example, the specific position of each grid in K is expressed as: K i+c =(x i+c ,y i+c), where c = {0, 1, 2, 3, 4} represents the horizontal distance from the center grid. The selection of each grid position in the convolution kernel K is an accumulation process. i Initially, the position of the grid away from the center depends on the position of the previous grid: K i+1 Relative to K i The offset is added. Therefore, the offset needs to be accumulated Δ = {δ | δ∈ [-1, 1]} to ensure that the convolution kernel conforms to the linear morphological structure. Since the offset Δ is usually a fraction, each convolution position is based on its previous position as a reference, and the swing direction is freely selected, thus ensuring the continuity of the perception while being free to choose. The final formula is as follows: K i+c =(x i +Δx i+c ,y i +Δy i+c ).

[0067] (4) Training phase: A large number of annotated image datasets are used for training, and the model parameters are optimized through the back propagation algorithm to minimize the loss function; the image dataset contains annotated historical pumping tunnel images.

[0068] (5) Validation and testing phase: Use independent datasets to evaluate the performance of the model to ensure its accuracy and robustness on new data.

[0069] The recognition accuracy of the model is improved by increasing the amount of data, introducing data enhancement technology, and using transfer learning methods. Data enhancement technology includes rotation, scaling, and flipping.

[0070] Ensemble learning methods, such as Bagging and Boosting, are used to combine the prediction results of multiple models to further improve the stability and accuracy of recognition.

[0071] By analyzing historical data, we can predict the deformation trend that may occur in the future. We can also use time series analysis methods such as the Autoregressive Integrated Moving Average Model (ARIMA Model) to predict the deformation trend that may occur in the future.

[0072] The basic form of the ARIMA model is:

[0073] X t =c+φ1X t-1 +…+φ p X t-p +θ1∈ t-1 +…+θ q ∈ t-q +∈ t ;

[0074] Among them, Xt is the observed value at time point t; c is a constant term; φ i and θ j are the autoregressive and moving average parameters, ∈ t It is white noise;

[0075] By fitting the historical displacement data, the model parameters are determined, and the displacement changes in the future time period are predicted using the model parameters.

[0076] In addition to the ARIMA model, other time series forecasting methods can also be considered, such as the SARIMA model (seasonal ARIMA), LSTM model (long short-term memory network), etc., to adapt to different types of deformed data.

[0077] By calculating the deformation rate at different time periods, a deformation rate change curve can be drawn, thereby intuitively showing the changing trend of the rock stress state.

[0078] The data processing unit features built-in intelligent reporting software. This intelligent reporting function automatically generates a comprehensive report based on a template, including geological sketches, deformation monitoring results, and risk assessments, and automatically fills in relevant forms. The report not only includes a description of the current status but also provides historical data comparison charts, helping project managers fully understand the tunnel's evolving status. Monitoring reports should include information such as geological sketches, deformation monitoring data, and potential safety hazards.

[0079] By comparing scans at different time points, we can not only visually understand the detailed changes in the rock surface over time, but also deeply analyze the deformation patterns of the rock after being subjected to external forces. Key indicators such as deformation rate analyzed from this data are important for identifying potential safety hazards, thereby enabling the implementation of timely and effective preventive measures to ensure construction safety.

[0080] The monitoring report includes a comparative analysis of deformation at the same location over several consecutive days, allowing analysis of rock stress and deformation. It should be noted that the monitoring report here refers to the resulting geological sketch and deformation monitoring report. This report specifically includes the geological conditions of the rock face, deformation of the crown and sidewalls, and potential safety hazards. The report should be updated regularly to ensure that construction personnel are kept up to date with the latest conditions in the pumped storage tunnel.

[0081] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0082] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel, characterized in that: The method comprises the following steps: Obtain pumped storage tunnel images and corresponding 3D point cloud data; The pumped storage tunnel image and the corresponding three-dimensional point cloud data are processed and integrated to generate a geological sketch and deformation monitoring report. A convolutional neural network is used to automatically identify the geological sketch in the pumped storage tunnel image. The convolutional neural network training process includes: normalizing and denoising the pumped storage tunnel image; using the convolutional layer in the convolutional neural network to automatically extract key features in the pumped storage tunnel image, the key features including edges and textures; adding DCNv2, DSConv, and CBAM modules based on VGGNet and yolov8; using a large number of annotated image datasets for training, and optimizing model parameters through a backpropagation algorithm to minimize the loss function; Among them, the geological sketch includes: the geological characteristics and crack distribution of the rock wall surface of the pumping tunnel; the deformation monitoring report includes: the deformation conditions and deformation rate of the top arch and side walls of the pumping tunnel; among them, the stress and deformation conditions of the rock include: the geological conditions of the rock wall surface, and the deformation conditions of the top arch and side walls.

2. The multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel according to claim 1 is characterized in that: The pumped storage tunnel image is used to generate a geological sketch, and the three-dimensional point cloud data is used to generate a rock deformation monitoring report.

3. The multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel according to claim 1 is characterized in that: The geological features of the rock wall of the pumping tunnel in the geological sketch include loose rock.

4. The multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel according to claim 3 is characterized in that: The convolutional neural network is used to automatically identify geological sketches in pumped storage tunnel images. The training process of the convolutional neural network also includes a validation and testing phase: using an independent data set to evaluate the performance of the model; During the training phase, the amount of data is increased, data enhancement technology is introduced, and transfer learning methods are used; integrated learning methods such as Bagging and Boosting are adopted to combine the prediction results of multiple models.

5. The multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel according to claim 1 is characterized in that: The calculation formula for the output eigenvalue y(p0) of DCNv2 is as follows: Where Δp n , Δm n are the learnable offset and modulation scalar of the nth position respectively; p0 is the input feature value of DCNv2; x(p0+p n +Δp n ) is the input feature map x at position p0+p n +Δp n The value at p n ∈R is the position offset p of all the positions belonging to the set R n .

6. The multi-dimensional dynamic deformation monitoring method during the excavation of a pumped storage tunnel according to claim 1 is characterized in that: The ARIMA model is used to predict the deformation trend that may occur in the future. The basic form of the ARIMA model is: X t =c+φ1X t-1 +…+φ p X t-p +θ1∈ t-1 +…+θ q ∈ t-q +∈ t ; Among them, X t is the observed value at time point t; c is a constant term; φ i and θ j are the autoregressive and moving average parameters, ∈ t It is white noise; By fitting the historical displacement data, the model parameters are determined, and the displacement changes in the future time period are predicted using the model parameters.

Images