Slope stability monitoring and early warning method and system

Abstract

The invention discloses a slope stability monitoring and early warning method and system. The method comprises the steps that slope treatment and monitoring equipment installation are carried out; monitoring data processing and CNN model input; carrying out convolutional neural network processing and risk classification; and according to the risk level result, generating an early warning instruction and sending the instruction to the monitoring terminal. According to the invention, through combination of the digital twin model and the CNN, multi-source data fusion and three-dimensional dynamic simulation are realized, and the monitoring precision and the early warning real-time performance are improved; rGB coding and imaging processing are adopted, monitoring data are visualized into images, and CNN feature extraction is facilitated; through comprehensive utilization and three-dimensional dynamic simulation of multi-source data, the system can more comprehensively and accurately predict the possibility of slope instability, early warning is given out in advance, adaptability is high, the system is suitable for complex geological conditions, reliability and accuracy of slope stability evaluation are remarkably improved, the slope instability risk is reduced to the maximum extent, and the slope stability evaluation efficiency is improved. And the safety and reliability of the project are ensured.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of highway traffic safety, and particularly relates to a slope stability monitoring and early warning method and system based on a digital twin model and a convolutional neural network. BACKGROUND

[0002] With the rapid development of infrastructure construction, slope stability monitoring and early warning have become the key to ensuring the safety of highways, bridges, tunnels and other projects.

[0003] At present, slope monitoring relies on manual observation and single sensor data collection, which has low monitoring accuracy, delayed early warning, and difficulty in capturing the dynamic changes inside the slope in a timely and accurate manner, resulting in a large uncertainty in instability risk assessment.

[0004] Existing monitoring systems mostly rely on two-dimensional models and do not fully consider the spatial diversity of three-dimensional geological conditions and the dynamic characteristics of time series data, which severely limits the accuracy of risk assessment and the effectiveness of management measures, and further affects the timeliness and reliability of early warning signals. SUMMARY

[0005] The application aims to provide a slope stability monitoring and early warning method and system based on a digital twin model and a convolutional neural network to solve the problems of low monitoring accuracy, delayed early warning and difficulty in adapting to complex geological conditions in the prior art.

[0006] To achieve the above-mentioned purpose, the application adopts the following technical solutions:

[0007] A slope stability monitoring and early warning method, comprising the following steps:

[0008] (1) Slope treatment and monitoring equipment installation:

[0009] Slope surface improvement: clean up vegetation, humus and other surface obstacles within the treatment range, and level the slope surface;

[0010] Anchor rod layout: spacing 1-2m, depth exceeding the sliding surface by more than 1m, drilling, hole cleaning, installation and cement grouting;

[0011] Laying of geotechnical net mat: fixed with U-shaped nails, depth not less than 30cm, overlapping thickness not less than 150mm;

[0012] Installation of sensors: including GNSS displacement meters, pore water pressure gauges and anchor rod stress gauges, ensuring close contact with the soil body;

[0013] Connection and debugging: connect the sensors to the monitoring terminal and perform preliminary debugging;

[0014] (2) Monitoring data processing and CNN model input:

[0015] Data acquisition: Extract historical and real-time monitoring data, including displacement, pore water pressure, and anchor stress;

[0016] Calculate the trend of change: based on the difference between real-time data and historical averages;

[0017] Normalization and RGB encoding: Mapping the trend of change to RGB color values;

[0018] Constructing an image matrix: Generate an N×N image, where each pixel represents the status of a monitoring point;

[0019] Image grayscale conversion and resizing: Convert to grayscale and resize to fit the CNN input;

[0020] Normalization: Adjusts pixel values ​​to the range of [0,1].

[0021] (3) Convolutional Neural Network Processing and Risk Classification: The processed image is input into the CNN model, and the slope risk level is output through convolutional layer, ReLU activation layer, pooling layer, fully connected layer and Softmax output layer;

[0022] Convolutional layers: extract local features of an image;

[0023] ReLU activation layer: performs nonlinear processing;

[0024] Pooling layer: performs downsampling to preserve key features;

[0025] Fully connected layer: maps feature vectors to risk levels;

[0026] Softmax output layer: Outputs the probability distribution of each risk level, and selects the highest probability as the final result;

[0027] (4) Based on the risk level results, an early warning instruction is automatically generated and sent to the monitoring terminal.

[0028] This invention also discloses a monitoring and early warning system for implementing the above-mentioned slope stability monitoring and early warning method, comprising:

[0029] Monitoring equipment includes GNSS displacement gauges, pore water pressure gauges, and anchor bolt stress gauges;

[0030] Monitoring terminal: used for data acquisition and transmission;

[0031] Monitoring server: integrates digital twin model and CNN model to realize data processing, risk analysis and early warning command generation;

[0032] The monitoring terminal is connected to the monitoring equipment and is used to receive sensor data and transmit it to the monitoring server;

[0033] The monitoring server is equipped with a digital twin model and a convolutional neural network model for real-time risk assessment and early warning.

[0034] The beneficial effects of this invention are:

[0035] 1. By combining digital twin models with CNNs, multi-source data fusion and three-dimensional dynamic simulation are achieved, improving monitoring accuracy and real-time early warning.

[0036] 2. By employing RGB encoding and image processing, the monitoring data is visualized as images, which facilitates CNN feature extraction;

[0037] 3. The system is highly adaptable and suitable for complex geological conditions, significantly improving the reliability and accuracy of slope stability assessment. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the monitoring and early warning system workflow of the present invention; Figure 2 This is a schematic diagram of a CNN-based slope ecological protection system. Figure 3 This is a schematic diagram of the monitoring terminal structure; Figure 4 This is a schematic diagram of the slope protection unit structure. Detailed Implementation

[0039] The present invention will be further described below with reference to the embodiments and accompanying drawings, but this is not intended to limit the scope of the invention.

[0040] A method for monitoring and early warning of slope stability, referring to Figure 1 It includes the following steps:

[0041] (1) Slope treatment and monitoring equipment installation:

[0042] First, slope treatment is carried out, including clearing vegetation and obstacles, and leveling the slope.

[0043] Anchor bolts were then installed at intervals of 1-2 meters, with a depth exceeding the sliding surface by more than 1 meter, and cement was poured in.

[0044] Lay geonet mats and fix them with U-shaped nails;

[0045] Next, install the GNSS displacement gauge, pore water pressure gauge, and anchor stress gauge to ensure that the equipment is in close contact with the soil.

[0046] Finally, connect the sensor to the monitoring terminal and debug it.

[0047] (2) Monitoring data processing and CNN model input:

[0048] Data acquisition: Historical and real-time monitoring data were extracted using GNSS displacement gauges, pore water pressure gauges, and anchor stress gauges, respectively.

[0049] The historical dataset contains M samples from N monitoring points at different time points, where N is no less than 64 and M is no less than 3000. The numerical indicators included are displacement, pore water pressure, and anchor stress. The real-time dataset records the corresponding values ​​from the M monitoring points at the current moment to reflect the real-time changes and dynamic evolution of the slope's condition.

[0050] Calculating the trend of change: For each monitoring point, the difference between real-time monitoring data and historical data is calculated, and then the trend of change for each monitoring point is derived. This is done using the following method:

[0051]

[0052] in, It represents the changing trend of the i-th monitoring point. This is the current real-time monitoring data. It is the average of historical data;

[0053] Normalization and RGB encoding: The calculated trend data is normalized, limiting its value range to [0, 1]. The normalized values ​​are then encoded into corresponding RGB values ​​according to the magnitude of change, with green indicating small changes, yellow indicating medium changes, and red indicating large changes. This generates an array containing N RGB values, each corresponding to the trend of a monitoring point. Subsequently, based on preset thresholds and risk level standards, the trend value of each monitoring point is matched with the corresponding risk level.

[0054]

[0055] in, and Using a preset risk threshold, all monitoring points will be assigned a corresponding risk level in this way, and the risk level will be combined with the RGB encoded image of the monitoring point to construct a training sample set;

[0056] Constructing an image matrix: The size of the training sample set should ensure coverage of all risk levels and various monitoring situations. Use the generated RGB values ​​to draw an N×N image matrix, where each pixel represents the status of a monitoring point.

[0057] Image grayscale conversion: Convert the drawn N×N RGB image matrix into a grayscale image, and calculate the grayscale value of each pixel using a weighted average method. The formula is:

[0058]

[0059] in, , , These represent the values ​​of the red, green, and blue channels, respectively. The calculated grayscale values ​​will be used for subsequent CNN feature extraction and analysis.

[0060] Image resizing: Bilinear interpolation is used to resize the image to fit the CNN input layer. Although rounding errors may be introduced when scaling grayscale values ​​to the range of 0 to 255, these errors are usually negligible.

[0061] The formula for calculating size adjustment is as follows:

[0062]

[0063] Among them, lowercase and The pixel positions of the original image. and Original image size, and The target image size;

[0064] Normalization: The resized image is normalized to adjust the pixel values ​​to the range of [0, 1]. The image data is then input into a CNN model for feature extraction and classification analysis.

[0065] (3) Convolutional Neural Network Processing and Risk Classification: The processed image is input into the CNN model, and the slope risk level is output through convolutional layer, ReLU activation layer, pooling layer, fully connected layer and Softmax output layer;

[0066] Convolutional layer processing: Normalized and grayscale image data are input into the convolutional layer of the CNN. Local features of the image are extracted through multiple convolutional kernels. These features include the displacement pattern of the slope surface, the change of pore water pressure, and the distribution of anchor stress.

[0067] The convolution operation is represented as:

[0068]

[0069] in, It is the first Each convolutional kernel is located at... Output characteristics at that location For the input image, For the first One convolutional kernel, This represents the convolution operation. For bias terms;

[0070] ReLU activation layer processing: The output of the convolutional layer is processed by the ReLU activation function. The function of ReLU is to set all negative values ​​to zero, while keeping positive values ​​unchanged.

[0071] Its mathematical expression is:

[0072] ;

[0073] Pooling layer processing: The activated feature map is input into the pooling layer for downsampling. Max pooling is typically used to reduce the size of the feature map, thereby reducing computational complexity while retaining important feature information. The pooling operation is represented as follows:

[0074]

[0075] in, The output value of the pooling layer reflects the most significant changing trend during slope monitoring.

[0076] Fully Connected Layer and Risk Classification: The feature map output from the pooling layer is flattened into a one-dimensional vector and then input into the fully connected layer. The fully connected layer maps the feature vector to the risk level classification result, as shown in the following formula:

[0077]

[0078] in, This is the weight matrix. For flattened eigenvectors, For bias vectors, To output the classification results;

[0079] Output Layer and Softmax Function: Building upon the fully connected layers, the output layer uses the Softmax function to normalize the classification results, converting them into probability distributions for each risk level. The formula is as follows:

[0080]

[0081] in, Indicates the first The predicted probability of a risk level. It is the first output of the fully connected layer. A score for each risk level. It is the total number of risk levels;

[0082] The Softmax function converts the output scores into probabilities between 0 and 1, with the sum of the probabilities for all categories being 1; ultimately, the highest probability value is selected. The final risk level of the slope monitoring is determined, and this result is used as the basis for system early warning.

[0083] (4) Based on the risk level results, an early warning instruction is automatically generated and sent to the monitoring terminal.

[0084] Reference Figures 2-4 Slope stability monitoring and early warning system, including:

[0085] Monitoring equipment includes GNSS displacement gauge 3, pore water pressure gauge 4, anchor stress gauge 5, etc., used to collect real-time monitoring data of the slope;

[0086] Monitoring terminal 2: used for data acquisition and transmission, and fixed to the mounting pole 7 through the protective box 6, with a lightning rod 8 and a solar power supply panel 9 installed on the top of the mounting pole 7;

[0087] Monitoring Server 1: Integrates digital twin model and CNN model to realize data processing, risk analysis and early warning command generation;

[0088] It also includes a data processing module, which processes the data collected by the sensors to generate historical and real-time datasets;

[0089] The CNN model module is used to receive processed data and perform data normalization, grayscale processing, convolution operations, pooling operations, fully connected operations, and classification prediction.

[0090] The early warning module compares the output of the CNN model with a preset threshold to generate a slope stability level, and automatically generates and issues an early warning instruction when the risk level reaches or exceeds the threshold.

[0091] And the slope protection unit 10, including anchor bolts 101, geonet mats 102, U-shaped nails 103, and slope protection vegetation 104; the anchor bolt stress gauge 5 is installed inside the anchor bolts 101, the geonet mats 102 are placed on the slope surface and fixed with U-shaped nails 103, and the slope protection vegetation 104 is densely planted in the three-dimensional grid space of the geonet mats 102.

[0092] Monitoring terminal 2 is connected to the monitoring equipment and is used to receive sensor data and transmit it to monitoring server 1;

[0093] The monitoring server is equipped with a digital twin model and a convolutional neural network model for real-time risk assessment and early warning.

[0094] This invention utilizes image analysis tools to identify abnormal patterns or hotspot areas in images, representing potential slope instability risks. The processed images and analysis results are uploaded to a slope monitoring server, and the images to be classified are input into a pre-trained CNN model. The model analyzes the input images and generates classification labels for the following day, predicting the slope stability level for that day. If the analysis results indicate that the risk level reaches or exceeds a preset threshold, the server will automatically generate and issue an early warning command based on a preset slope control strategy. Simultaneously, it will send control commands corresponding to the slope's stability risk level for the following day to the relevant terminals, ensuring preventative measures are taken before the risk occurs.

[0095] This invention is applicable to slope stability monitoring in highway, railway, and water conservancy projects, and has high practical value and promotion potential.

Claims

1. A method for monitoring and early warning of slope stability, characterized in that, It comprises the following steps: (1) slope treatment and monitoring equipment installation: the slope surface is treated, anchor rods, geotechnical mesh mats are laid, and GNSS displacement meters, pore water pressure gauges and anchor rod stress gauges are installed; (2) monitoring data processing and CNN model input: collecting historical and real-time monitoring data, calculating the change trend, normalizing and RGB encoding, generating an image matrix, and performing grayscale, size adjustment and normalization processing; (3) convolutional neural network processing and risk classification: input the processed image into the CNN model, pass through the convolutional layer, ReLU activation layer, pooling layer, fully connected layer and Softmax output layer, and output the slope risk level; (4) according to the risk level result, generate an early warning instruction and send it to the monitoring terminal.

2. The method of claim 1, wherein, In step (1), the anchor rod is laid at a spacing of 1-2 m and a depth of more than 1 m above the sliding surface; the geotechnical mesh mat is fixed with U-shaped nails, with a depth of not less than 30 cm and an overlapping thickness of not less than 150 mm.

3. The method of claim 1, wherein, In step (2), the change trend calculation formula is: ； where ΔXi is the change trend of the ith monitoring point, Xcurrent is the current monitoring data, and mean(Xhistory) is the average value of the historical data.

4. The method of claim 1, wherein, In step (2), the RGB encoding maps the change trend value to a color value: green for small changes, yellow for moderate changes, and red for large changes.

5. The method of claim 1, wherein, In step (2), the image grayscale is calculated using the formula: Gray=0.299×R+0.587×G+0.114×B; Wherein, R, G, B are red, green, blue channel values, respectively, is a calculated gray value.

6. The method of claim 1, wherein, In step (2), the image size adjustment uses the bilinear interpolation method, with the formula: ； where lower case and are pixel positions of the original image, and are the original image size, and are the target image size.

7. The method of claim 1, wherein, In step (3), the convolution operation formula of the CNN model is: ； wherein, is the output feature of the th convolution kernel at position , is the input image, is the th convolution kernel, denotes the convolution operation, is the bias term.

8. The method of claim 1, wherein, In step (3), the risk level is calculated by the Softmax function to obtain the probability distribution, with the formula: ； wherein, denotes the predicted probability of a risk class, is the score of the risk class output by the fully connected layer, is the total number of risk classes.

9. A slope stability monitoring and warning system for implementing the method according to any one of claims 1 to 8, characterized in that It comprises: Monitoring equipment: including GNSS displacement meters, pore water pressure gauges and anchor rod stress gauges; Monitoring terminal: for data acquisition and transmission; Monitoring server: for data processing, CNN model analysis and early warning instruction generation.

10. The system of claim 9, wherein, The monitoring terminal is connected with the monitoring equipment for receiving sensor data and transmitting to the monitoring server; The monitoring server is configured with a digital twin model and a convolutional neural network model for real-time risk assessment and early warning.

Images