Tunnel surrounding rock instability early warning method, device, equipment and medium

Abstract

The invention discloses a tunnel surrounding rock instability early warning method, device and equipment and a medium. The method comprises the following steps: acquiring a microseismic signal and surrounding rock surface image sequence synchronously acquired by sensors arranged along a tunnel; microseismic characteristic parameters representing internal damage of the surrounding rock are extracted from the microseismic signals; extracting visual characteristic parameters representing the surface damage of the surrounding rock from the image sequence; the micro-seismic characteristic parameters and the visual characteristic parameters are subjected to correlation fusion analysis, a fusion analysis result is obtained, and the fusion analysis result at least comprises surrounding rock stable state comprehensive indexes and risk area space information; and based on the fusion analysis result, carrying out risk grade determination and sending out an early warning. According to the method, the problem of insufficient early warning accuracy of a single monitoring means is solved, and earlier and more accurate judgment of the instability precursor of the tunnel surrounding rock is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel engineering safety monitoring, and in particular to a tunnel surrounding rock instability early warning method, device, equipment and medium. BACKGROUND

[0002] The safe construction and operation of tunnels and underground projects are heavily dependent on timely and accurate judgments of the stability of surrounding rocks. For a long time, engineering sites have mainly used manual inspection, displacement convergence meters and other contact monitoring methods. These methods are not only inefficient and have discrete data collection, but more importantly, they have significant lag, and problems are usually only discovered after macroscopic deformation of the surrounding rock is visible, at which point the best disposal opportunity has often been missed, with high safety risks.

[0003] In recent years, the application of advanced technologies such as microseismic monitoring has enabled real-time capture and quantitative analysis of internal rupture events in surrounding rocks, theoretically improving the perception of internal damage in rock masses. However, engineering practice has shown that the instability process of surrounding rocks is a complex result of the combined evolution of internal damage accumulation and surface morphology changes. Relying solely on microseismic parameters reflecting internal rupture sometimes makes it difficult to accurately characterize the overall stability of surrounding rocks and the deterministic trend of their development towards macroscopic instability. This inherent uncertainty makes it difficult for existing early warning methods to achieve breakthroughs in accuracy and reliability, and cannot provide sufficient reliable evidence for high-risk construction decisions. SUMMARY

[0004] The present application provides a tunnel surrounding rock instability early warning method, device, equipment and medium, aiming to solve the problem of insufficient accuracy and delayed early warning of tunnel surrounding rock instability in the prior art due to reliance on a single type of monitoring data.

[0005] In a first aspect, the present application provides a tunnel surrounding rock instability early warning method, comprising: acquiring microseismic signals and surrounding rock surface image sequences synchronously collected by sensors arranged along the tunnel; extracting microseismic feature parameters representing internal damage of the surrounding rock from the microseismic signals; extracting visual feature parameters representing surface damage of the surrounding rock from the image sequences; performing correlation fusion analysis on the microseismic feature parameters and the visual feature parameters to obtain a fusion analysis result, the fusion analysis result including at least a comprehensive index of the stability of the surrounding rock and spatial information of a risk area; based on the fusion analysis result, determining a risk level and issuing an early warning.

[0006] In a second aspect, the present application also provides a tunnel surrounding rock instability early warning device, comprising: a data acquisition module, configured to acquire microseismic signals and image sequences of a surrounding rock surface synchronously collected by sensors arranged along a tunnel; a microseismic data analysis module, configured to extract microseismic characteristic parameters representing internal damage of the surrounding rock from the microseismic signals; an image data analysis module, configured to extract visual characteristic parameters representing surface damage of the surrounding rock from the image sequences; a fusion evaluation module, configured to perform correlation fusion analysis on the microseismic characteristic parameters and the visual characteristic parameters to obtain a fusion analysis result, the fusion analysis result at least including a comprehensive index of a stability state of the surrounding rock and spatial information of a risk area; a warning module, configured to perform risk level determination and issue a warning based on the fusion analysis result.

[0007] In a third aspect, an embodiment of the present application further provides a computer device, which comprises a memory and a processor, the memory has a computer program stored thereon, and the processor implements the tunnel surrounding rock instability warning method when executing the computer program.

[0008] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program can implement the tunnel surrounding rock instability warning method when executed by a processor.

[0009] The present application effectively overcomes the inherent limitations of a single monitoring method by synchronously collecting and fusion analyzing two types of heterogeneous monitoring data, i.e., microseismic data and visual data; by performing correlation fusion analysis on internal damage features and surface damage features, the continuous evaluation of the internal-to-external damage evolution process of the tunnel surrounding rock can be realized, thereby significantly improving the accuracy and reliability of the instability warning and enabling the identification of risk precursors at an earlier stage; meanwhile, the quantitative comprehensive index and spatialized information generated by the fusion analysis also provide a reliable data basis for subsequent risk management and accurate decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0011] FIG. 1 is a flowchart of the tunnel surrounding rock instability warning method provided by an embodiment of the present application; FIG. 2 is a flowchart of data fusion evaluation provided by an embodiment of the present application; Figure 3 FIG. 3 is a flowchart of data spatio-temporal correlation provided by an embodiment of the present application; Figure 4 A flowchart of a fusion computing process provided by an embodiment of the present application is shown in FIG. 1. Figure 5 A flowchart of a model processing process provided by an embodiment of the present application is shown in FIG. 2. Figure 6 A flowchart of a model training process provided by an embodiment of the present application is shown in FIG. 3. Figure 7 A flowchart of a support decision scheme generation process provided by an embodiment of the present application is shown in FIG. 4. FIG. 8 is a schematic diagram of a tunnel surrounding rock instability early warning device provided by an embodiment of the present application. FIG. 9 is a schematic block diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0012] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of the present application.

[0013] It should be understood that, when used in the specification and the appended claims, the terms “comprise” and “include” indicate the presence of the described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0014] It should also be understood that the terms used in the present application specification are only for the purpose of describing specific embodiments and are not intended to limit the present application. As used in the present application specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0015] It should be further understood that the term “and / or” used in the present application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations thereof, and includes these combinations.

[0016] As used in the present specification and the accompanying claims, the term “if’ can be interpreted as “when” or “upon” or “in response to a determination” or “in response to a detection” depending on the context. Similarly, the phrase “if it is determined” or “if [the recited condition or event] is detected” can be interpreted as meaning “upon determining” or “in response to a determining” or “upon detecting [the recited condition or event]” or “in response to a detection of [the recited condition or event]” depending on the context.

[0017] As Figure 1 shown in the figure, the tunnel surrounding rock instability early warning method provided by the embodiment of the present application comprises steps S1-S5: S1, acquiring microseismic signals and surrounding rock surface image sequences synchronously collected by sensors arranged along the tunnel.

[0018] In this step, taking a water diversion tunnel project of a certain hydropower station as an example, the tunnel has a buried depth of about 350 meters, and the surrounding rock is mainly type III surrounding rock and locally type IV. To implement the early warning method, an integrated monitoring section is arranged every 15 meters along the tunnel axis within a range of 200 meters behind the tunnel excavation face. A set of microseismic sensor array (including 4 high-frequency accelerometers arranged on the tunnel wall and the vault) and a panoramic high-definition infrared camera are installed at each section. The microseismic sensor has a sampling frequency of 5000 Hz, continuously recording the vibration waveform generated by the internal rupture of the surrounding rock; the infrared camera shoots the surrounding rock surface image at a frequency of 2 seconds / frame, and automatically synchronizes the time stamp and spatial coordinates. All data are transmitted in real time to the ground monitoring center through industrial Ethernet and stored in a time series database. The synchronous acquisition realizes the complete capture of the internal to external damage information of the surrounding rock under the same space-time scale, providing a strictly aligned original data basis for subsequent multi-source data fusion.

[0019] S2, extracting microseismic characteristic parameters representing internal damage of the surrounding rock from the microseismic signals.

[0020] In this step, the microseismic signal stream transmitted in real time is processed. First, the wavelet threshold denoising method is used to filter out environmental noise such as construction machinery vibration, retaining the effective frequency band of 100-1500 Hz. The STA / LTA (short-term average / long-term average) algorithm is used to automatically detect microseismic events, and the Geiger positioning method is used to invert the source coordinates (X, Y, Z) and occurrence time of each event by using the P-wave and S-wave arrival time difference and combining the geometric arrangement of the sensor array. For each successfully located microseismic event, the microseismic energy release rate and event spatial aggregation degree are extracted.

[0021] Microseismic energy release rate: the source energy of a single event is calculated by integrating the square of the velocity waveform, and then the total energy released by the unit volume of surrounding rock in each analysis period (such as 10 minutes) is calculated, with the unit of J / m 3•min. This parameter directly reflects the severity of internal fracturing and the rate of damage accumulation within the rock mass.

[0022] Event spatial clustering: A spatial clustering analysis method based on kernel density estimation is used to calculate the distribution density of seismic source points in three-dimensional space. By defining a clustering index (0~1), the distribution density of events is quantified as to whether they are uniformly dispersed or highly concentrated in a certain spatial region. High clustering usually indicates the formation of potential shear slip surfaces or failure cores.

[0023] S3. Extract visual feature parameters characterizing the damage to the surrounding rock surface from the image sequence.

[0024] In this step, the image sequence transmitted by the infrared camera is processed automatically. First, image preprocessing is performed, including grayscale conversion, contrast enhancement, and perspective correction, to eliminate the effects of uneven lighting and shooting angle. Then, an improved U-Net deep learning model is used for semantic segmentation of the images, accurately identifying the crack pixel regions on the surrounding rock surface. Based on the segmentation results, the crack width expansion rate and crack network complexity are extracted.

[0025] Crack width propagation rate: The skeleton line of the same crack is tracked and matched in images at different times, and the pixel change in its average width is measured. Combined with the camera's calibration parameters, the pixel width is converted to physical width (millimeters), and the propagation rate per unit time (mm / h) is calculated. This parameter is a key indicator for determining whether a crack is in an active propagation phase.

[0026] Crack network complexity: The fractal dimension of the identified crack regions was calculated using box counting. A higher fractal dimension indicates a more complex crack morphology, more branches, and stronger interconnectivity. Simultaneously, the number of crack intersections per unit area was counted as an auxiliary indicator of network complexity.

[0027] This embodiment introduces computer vision and image metrics methods to transform surface cracks observed by the naked eye into quantifiable and traceable geometric and dynamic feature parameters, thereby achieving an objective and continuous assessment of surface damage to the surrounding rock.

[0028] S4. Perform correlation and fusion analysis on the microseismic characteristic parameters and the visual characteristic parameters to obtain the fusion analysis results. The fusion analysis results include at least the comprehensive index of surrounding rock stability and spatial information of risk areas.

[0029] In a specific embodiment, such as Figure 2 As shown, step S4 involves performing a correlation and fusion analysis on the microseismic feature parameters and the visual feature parameters to obtain the fusion analysis results, including steps S41-S44: S41, associate the microseismic events and visually identified cracks to multiple independent analysis regions in the same tunnel 3D model according to their spatial positions and occurrence times.

[0030] In specific embodiments, as shown in FIG. 4, the step S41 of associating the microseismic events and visually identified cracks to multiple independent analysis regions in the same tunnel 3D model according to their spatial positions and occurrence times comprises steps S411-S413: Figure 3 S411, perform spatial clustering of microseismic events according to their hypocenter coordinates to form multiple microseismic event clusters.

[0031] In this step, take a tunnel section as an example, which has dense microseismic activities and new cracks in the past period. First, load the high-precision BIM 3D model of this section as a spatial reference benchmark, and cluster the hypocenter points (3D coordinates) of all microseismic events near this section. Specifically, use the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm based on density, set the neighborhood search radius ε = 4.5 meters (determined according to the sensor layout distance and positioning accuracy), and the minimum point number MinPts = 4. The algorithm automatically aggregates microseismic events with a distance less than 4.5 meters and an event number not less than 4 into a microseismic event cluster. For example, in the rock mass 2-6 meters above the vault, the algorithm identifies a cluster C1 composed of 8 events; at the back of the left wall 3 meters away, another cluster C2 composed of 5 events is identified. Each cluster represents a spatially concentrated internal damage zone, indicating that the internal rupture of the rock mass at this place is changing from random distribution to ordered concentration, which may be a precursor of potential damage nucleus or slip surface. Through clustering, discrete point events are summarized into damage zones with clear spatial significance, greatly simplifying the complexity of subsequent analysis.

[0032] S412, determine the spatial projection range of each microseismic event cluster on the surface of the tunnel 3D model.

[0033] ​In this step, for each microseismic event cluster (such as cluster C1) generated in the previous step, the active range in the three-dimensional rock mass needs to be mapped onto the tunnel excavation contour surface to determine the possible affected surface area. The specific operation is as follows: calculate the three-dimensional convex hull or outer envelope cuboid of all hypocenter points of cluster C1, and obtain its geometric center. Project the center point vertically onto the lining inner surface of the tunnel BIM model to obtain the projection base point O. Then, calculate the maximum distance R_max of all hypocenter points in the cluster to the geometric center. Finally, generate a spatial projection range area P1 on the inner surface of the three-dimensional model with the projection base point O as the center and R_max + a safety buffer distance of 2.0 meters as the radius. This area is the signal projection area of internal damage on the surface of the tunnel wall, which provides an accurate spatial search frame for finding the corresponding surface cracks in the next step.

[0034] S413, associate the cracks appearing in the preset time window and whose spatial positions fall into or are adjacent to the spatial projection range to the same analysis region as the corresponding microseismic event cluster.

[0035] In this step, the association between internal and external damage is established according to the principle of spatial and temporal proximity. For cluster C1 and its projection range P1, a dynamic association time window is set, usually 20 minutes before and after the active time period of the microseismic cluster. Automatically retrieve all newly generated or significantly expanded cracks identified by step S3 within this time window. For each crack, calculate the average position of its skeleton line on the surface of the three-dimensional model. If the position falls within the range P1, or the distance to the boundary of P1 is less than 1 meter, it is defined as adjacent, then it is determined that the crack has a high spatial and temporal correlation with the microseismic cluster C1. For example, a longitudinal crack that appears on the vault and whose position falls within the range P1 and whose appearance time is 10 minutes after the activity of cluster C1, they are bound together to form an independent evaluation unit named analysis region A1. This step discards the traditional method of isolated analysis data, and by establishing the association between internal and external damage in the same space and close time, it constructs a physical analysis unit that can truly reflect the collaborative evolution process of rock damage.

[0036] S42, for each analysis region, extract first type features from the microseismic events associated with it, and extract second type features from the crack information associated with it; wherein the first type features include microseismic energy release rate and event spatial aggregation degree, and the second type features include crack width expansion rate and crack network complexity degree.

[0037] In this step, standardized feature quantification is performed on each analysis region (e.g., region A1). First, the first type of feature (internal damage feature) is extracted from the microseismic event cluster C1 associated with this region: 1. Microseismic energy release rate: Calculate the total source energy released by all events in cluster C1 within the most recent hour, and then divide it by the product of the equivalent volume of the cluster (estimated through the 3D convex hull of its source point cloud) and time (1 hour) to obtain the energy release rate per unit volume per unit time (unit: J / m). 3 ·h), this value directly quantifies the intensity of internal damage accumulation. 2. Event spatial clustering: Based on clustering, the standard deviations σ_x, σ_y, and σ_z of the source point coordinates of cluster C1 in three directions are further calculated. The reciprocal of their root mean square is taken and normalized to obtain a clustering index between 0 and 1. The closer the index is to 1, the more concentrated the event is in a narrow spatial range, indicating that a through-type rupture surface or slip zone may be forming. Secondly, the second type of feature (surface damage feature) is extracted from all cracks bound to this area: 1. Crack width propagation rate: The increment of the average width of these cracks in the current analysis period (e.g., the last 30 minutes) is calculated and divided by time to obtain the average propagation rate (unit: mm / h), which is used to determine whether the surface damage is in an active development period. 2. Crack network complexity: These cracks are regarded as a whole, and the fractal dimension (Df) of their distribution pattern is calculated using the box counting method. The higher the Df value, the more complex the crack morphology and the stronger the space filling ability. Simultaneously, the density of crack intersections per unit area is statistically analyzed. Finally, the fractal dimension and intersection density are weighted and fused to obtain a comprehensive complexity index. This step generates a standard feature vector for each analysis region, containing four quantitative indicators (energy release rate, aggregation degree, expansion rate, and complexity), providing structured data input for the subsequent fusion calculation.

[0038] S43. The first type of features and the second type of features from the same analysis area are fused and calculated to obtain the risk assessment value of each analysis area.

[0039] In a specific embodiment, such as Figure 4 As shown, step S43, fusing the first type of features and the second type of features from the same analysis region to obtain the risk assessment value for each analysis region, includes steps S431-S433: S431. Combine the first type of features and the second type of features into a multidimensional input vector.

[0040] In this step, data preprocessing and vector construction are performed on the four feature values ​​of the analysis region A1. First, the max-min normalization method is used to map each feature value to the [0,1] interval based on the statistical range of historical monitoring data to eliminate dimensional differences. For example, if the historical maximum value of energy release rate is E_max and the current value is E_curr, then the normalized value is (E_curr / E_max). Then, in a fixed order (internal features first, visual features second), the four normalized scalar values ​​are concatenated into a four-dimensional feature vector: V_A1 = [E_norm, C_norm, W_norm, F_norm]. This vector is a standardized mathematical expression that the fusion model can process, encapsulating all quantitative information about the damage state of region A1 from the inside out at a specific time.

[0041] S432. Input the multidimensional input vector into a trained machine learning fusion model.

[0042] In this step, the input interface for the trained machine learning fusion model is a four-dimensional feature vector. The vector V_A1 is transmitted to the machine learning fusion model via internal function calls or service interfaces.

[0043] S433. The machine learning fusion model calculates the correlation weights and nonlinear relationships between the first type of features and the second type of features in the multidimensional input vector, and outputs a risk assessment value.

[0044] In this step, the machine learning fusion model begins deep analysis of the input vector V_A1, dynamically evaluating the complex coupling relationship between microseismic features and visual features, rather than simply weighting them. For example, when the machine learning fusion model identifies a pattern where high energy release rate (E_norm) and high crack propagation rate (W_norm) occur simultaneously, it assigns a very high risk weight to this synchronously accelerating pattern through an internal mechanism. Conversely, if high clustering (C_norm) is accompanied by low complexity (F_norm), the machine learning fusion model may determine that it indicates the early stage of deep shear rupture, where the surface response has not yet fully manifested, thus outputting a medium-risk value. The machine learning fusion model generates a single scalar value at the output layer through a nonlinear transformation network, namely the real-time risk assessment value (RS_A1) of region A1, typically a value between 0 and 100. This value is a comprehensive quantitative estimate of the likelihood of near instability in region A1 by the machine learning fusion model. Its accuracy directly stems from learning from a large amount of historical instability and stability case data, and it can characterize complex early warning patterns.

[0045] In a specific embodiment, such as Figure 5As shown, in step S433, the machine learning fusion model is a neural network model. The machine learning fusion model calculates the correlation weights and nonlinear relationships between the first type of features and the second type of features in the multidimensional input vector, and outputs a risk assessment value, including steps S4331-S4334: S4331. The multidimensional input vector is input into the input layer of the neural network model, and the input layer maps the first type of features and the second type of features to different initial feature vectors respectively.

[0046] In this step, the four-dimensional vector V_A1 enters the input layer. The input layer doesn't pass the vector directly; instead, it involves a feature enhancement process. Specifically, through two independent linear transformation matrices, the two values ​​[E_norm, C_norm] representing the first type of feature are mapped to a 64-dimensional vector H_micro, and the two values ​​[W_norm, F_norm] representing the second type of feature are mapped to another 64-dimensional vector H_visual. This mapping enhances the low-dimensional engineered features to a higher-dimensional representation space, aiming to extract more abstract and discriminative information, laying the foundation for subsequent deep interactions.

[0047] S4332. In the hidden layer of the neural network model, the mutual attention weight between the initial feature vector corresponding to the first type of feature and the initial feature vector corresponding to the second type of feature is calculated through a cross-attention mechanism, and the two types of initial feature vectors are weighted, converged and fused according to the attention weight to generate a fused intermediate vector.

[0048] In this step, the cross-attention mechanism of the neural network model is activated. This mechanism allows the model to dynamically focus on the correlation between two types of features. Using H_micro as the query and H_visual as the key and value, a set of attention weights is calculated. This set of weights explicitly indicates which aspects of the visual features (width changes or morphological complexity) are more worthy of attention in the context of the current microseismic activity. The weights are then used to weight and sum H_visual to obtain a condensed visual information vector related to the microseismic features. Similarly, using H_visual as the query and H_micro as the key and value, H_micro is calculated and weighted in reverse. The vectors obtained from the two convergences are concatenated to form a 128-dimensional fused intermediate vector. This vector contains bimodal depth information after importance filtering and interaction.

[0049] S4333. The fused intermediate vector is transformed by at least one fully connected layer and a nonlinear activation function to obtain a deep fused feature vector.

[0050] In this step, the aforementioned fused intermediate vector is fed into a subnetwork consisting of three fully connected layers. Each layer performs a linear transformation (W*X + b) and a non-linear activation (ReLU). Through this stacking, the network can extract higher-level, more abstract risk patterns from the fused information layer by layer. For example, the first layer might learn damage rate combination features, and the second layer might learn spatiotemporal correlation features of damage patterns. After the final fully connected layer, a 32-dimensional deep fused feature vector is output. This deep fused feature vector contains the decisive abstract features used for the final risk assessment.

[0051] S4334. The deep fusion feature vector is linearly mapped through the output layer to output the risk assessment value.

[0052] In this step, a 32-dimensional deep fusion feature vector is input into the output layer. The output layer is a linear layer that maps this high-dimensional vector to a single scalar value. To obtain a risk assessment value in the range of 0-100, a sigmoid function is appended after the linear layer to limit the output to (0,1), and then multiplied by 100. The risk assessment value RS_A1 for the model output region A1 is 78. This value is the final quantitative output after the entire complex neural network comprehensively analyzes the surrounding rock condition, and is used for subsequent early warning decisions.

[0053] S44. Based on the risk assessment values ​​of each analysis area, generate the comprehensive index of the surrounding rock stability state, and determine the spatial range of the analysis area where the risk assessment value exceeds the preset threshold as the spatial information of the risk area.

[0054] In this step, the risk assessment values ​​of all analysis areas are spatially integrated and visualized. Using the Kriging spatial interpolation algorithm, the risk values ​​discretely distributed at the center points of each analysis area (e.g., RS_A1=78, RS_A2=65, RS_B1=45, etc.) are interpolated to generate a continuous and smooth risk cloud map of the surrounding rock stability state covering the entire monitored tunnel section surface. Different colors represent different risk levels, intuitively displaying the global risk distribution. The average and standard deviation of the risk values ​​in all areas are calculated to form the Comprehensive Stability Index (CSI), for example, CSI=68. Preset warning thresholds are applied (in this embodiment, RS≥70 is the orange warning threshold, and RS≥85 is the red alarm threshold). All areas are automatically scanned, and areas with risk values ​​exceeding 70 (e.g., area A1) are highlighted in the tunnel's 3D BIM model. Their spatial boundaries (coordinate set), projected area, and estimated volume are accurately calculated, and this information is structured and stored as a risk area spatial information data package. This step provides a quantitative overview of the overall security situation (CSI and cloud map), and accurately identifies the specific risk locations, scope, and level that require immediate attention and action.

[0055] S5. Based on the fusion analysis results, determine the risk level and issue an early warning.

[0056] In this step, the early warning decision engine automatically runs tiered early warnings based on the fusion analysis results (CSI and spatial information of risk areas). The judgment rules are as follows: if CSI ≥ 75, or if any area has a risk assessment value ≥ 85, it is judged as a red (critical) warning; if 70 ≤ CSI < 75, or if ≥ 2 areas have risk values ​​between 70 and 85, it is judged as an orange (warning); if 65 ≤ CSI < 70, or if only one area has a risk value between 70 and 85, it is judged as a yellow (attention); the rest are green (normal). When an orange or red warning is judged, a multi-channel alarm is immediately triggered: on the large screen of the monitoring center, the risk area (such as area A1) in the 3D tunnel model begins to flash red, and a detailed warning panel pops up; at the same time, the warning information (including level, precise station number, dominant risk characteristics, and suggested actions) is automatically pushed to the mobile terminals of key personnel such as project manager, safety director, and on-site shift leader via SMS and a dedicated safety APP; voice warnings are also played simultaneously in the tunnel. The early warning release mechanism in this embodiment ensures that risk information is transmitted in the most effective way at the first moment.

[0057] In a specific embodiment, such as Figure 6 As shown, the machine learning fusion model is obtained through training steps, which include steps A1-A2: A1. Construct a training dataset. Each data sample includes the first and second types of features of a historical analysis region within a specific time period, as well as the actual stability state label corresponding to that region.

[0058] In this step, the quality of the training data directly determines the early warning capability of the fusion model. A training dataset is constructed based on years of safety monitoring records for multiple water diversion tunnels of a large hydropower station. The specific process is as follows: First, a large number of past analysis area cases with clear subsequent results are extracted from the historical database. For example, Case 1: In the arch area at chainage K10+200 of Tunnel A in a certain month of a certain year, high micro-seismic activity and crack propagation were recorded for three consecutive days; subsequently, local rockfall occurred at this location (defined as an instability event). Case 2: In the sidewall area at chainage K15+500 of Tunnel B in a certain month of a certain year, a brief micro-seismic event was detected, followed by calm, with no new cracks on the surface, and subsequent stability. For each such historical analysis area, following a process completely consistent with real-time early warning, its feature vector within a specific time period before and after the event (such as 24 hours before instability, or any 24 hours of stable state) is reconstructed. This involves extracting four types of features within that time period: micro-seismic energy release rate, event spatial clustering, crack width propagation rate, and crack network complexity, and then normalizing them. The most crucial step is assigning a label to each sample indicating its actual stability state. Label determination integrates information from multiple sources: for unstable cases, the label is "1" (high risk); for stable cases, the label is "0" (low risk). Labels for some complex cases are determined collectively by multiple senior geological engineers based on subsequent detailed exploration reports (such as whether additional support was implemented and data on subsequent rock mass deformation). Ultimately, a balanced dataset containing thousands of samples is constructed, where each sample is a four-dimensional feature vector with a 0 / 1 label. This step transforms scattered engineering experiences and lessons learned from historical records into standardized, structured knowledge that can be learned by machine learning algorithms.

[0059] A2. Using the combination of the first type of features and the second type of features as input, and the actual stability state label as the target output, the selected machine learning algorithm is trained to obtain the machine learning fusion model.

[0060] In this step, the selected neural network algorithm is trained using the constructed dataset to learn the mapping relationship from features to risk. The specific training process is as follows: First, the entire dataset is randomly divided into a training set (70%), a validation set (15%), and a test set (15%). A neural network architecture based on the cross-attention mechanism is selected as the machine learning algorithm. During training, mini-batch gradient descent is used, driven by the training set data. In each iteration, the algorithm reads a batch of samples (e.g., 32), inputs a vector composed of four feature values ​​into the network for forward propagation, and obtains the network's predicted risk value (between 0 and 1). Then, the loss between the predicted risk value and the true label (0 or 1) is calculated, typically using the binary cross-entropy loss function. Next, the gradient of the loss with respect to the millions of parameters (weights and biases) of the network is calculated using the backpropagation algorithm, and the Adam optimizer is used to update these parameters based on the gradient, so that the predicted value continuously approaches the true label. During training, after each epoch (one iteration of the training set), the model's performance is evaluated on an independent validation set, monitoring metrics such as accuracy and recall to prevent overfitting. Training stops early when performance on the validation set no longer improves. Finally, the final model undergoes rigorous performance evaluation on a test set that was never used in training to ensure its generalization ability. The trained model, i.e., the machine learning fusion model, has its internal parameters pre-defined with knowledge of the complex correlation weights and nonlinear relationships between microseismic events and visual features, enabling it to output an accurate risk assessment value for new, unseen feature vectors of the analysis area. This step embeds data-driven learning capabilities into traditional engineering safety monitoring, enabling early warning decisions to learn and optimize from historical big data.

[0061] In a specific embodiment, such as Figure 7 As shown, the tunnel surrounding rock instability early warning method also includes the step of generating a support decision scheme, specifically steps S61-S63: S61. Based on the spatial information of the risk area, determine the three-dimensional boundary of the body to be supported in the three-dimensional model of the tunnel.

[0062] In this step, after a high-level warning is issued, the support decision support module is activated, reads the spatial information data package of the risk area, and obtains the spatial boundary coordinates of the high-risk area. Then, in the tunnel's 3D BIM model, based on these boundary coordinates and comprehensively considering the rock mass structure surface attitude and in-situ stress direction, a more regularly shaped 3D solid model of the rock mass to be supported, encompassing the risk area, is automatically generated through geometric Boolean operations. Key geometric parameters such as the volume, maximum span, and minimum overburden thickness of this solid model are calculated, providing dimensional basis for support design.

[0063] S62. Based on the dominant damage characteristic type within the analysis area corresponding to the spatial information of the risk area, match the corresponding damage mechanism from the preset damage mode library.

[0064] In this step, the feature vectors extracted from the high-risk area are analyzed retrospectively. By comparing the relative strength and combination patterns of microseismic features (energy release rate, concentration) and visual features (expansion rate, complexity), the dominant damage type is determined. For example, if the microseismic energy release rate is high and the spatial concentration is significant, while the surface crack propagation is relatively slow, it may be determined to be a failure mode dominated by internal shear fracturing; if the complexity of the surface crack network increases sharply while the microseismic activity is relatively calm, it may correspond to a shallow tensile stripping mode. The judgment results are matched with a pre-set rock mass failure mode knowledge base, which contains descriptions of the mechanical mechanisms, evolution laws, and precursor characteristics of dozens of typical surrounding rock failures, thereby determining the most likely failure mechanism for the area to be supported.

[0065] S63. Combining the three-dimensional boundary of the body to be supported and the failure mechanism, query the support strategy knowledge base to generate at least one support scheme that includes support type and design parameters.

[0066] In this step, the 3D model of the rock mass to be supported and the matching failure mechanism are used as joint query conditions to access a support strategy knowledge base built based on a large number of engineering cases, expert experience, and numerical simulation results. This knowledge base uses case-based reasoning technology to retrieve support schemes successfully applied in similar geological conditions and failure modes throughout history. For example, for a medium-risk area with a volume of approximately 50 cubic meters dominated by internal shear fracture, the knowledge base might recommend: Scheme 1: Radial system mortar anchor support. Anchor specifications: Φ22mm, HRB400 threaded steel, length L=3.0m, spacing 1.2m×1.2m in a staggered pattern, grouting pressure 0.5-0.8MPa. This is combined with a steel mesh (Φ6.5@150×150) and shotcrete C25, with an initial shotcrete thickness of 50mm and a secondary shotcrete to a total thickness of 120mm. The scheme will detail material requirements, construction sequence, and key quality control points. It can output 1-3 alternative solutions, along with a brief description of their respective advantages and disadvantages, providing engineering and technical personnel with rapid, scientific, and targeted decision support. It directly transforms the results of monitoring and early warning into actionable engineering measures, forming a complete closed loop for safety management.

[0067] like Figure 8 As shown, this embodiment of the invention also provides a tunnel surrounding rock instability early warning device, comprising: Data acquisition module 10 is used to acquire microseismic signals and surrounding rock surface image sequences synchronously collected by sensors deployed along the tunnel; Microseismic data analysis module 20 is used to extract micro-seismic data characterizing internal damage of the surrounding rock from the microseismic signals. Seismic characteristic parameters; Image data analysis module 30 is used to extract visual feature parameters characterizing damage to the surrounding rock surface from the image sequence; The fusion evaluation module 40 is used to perform correlation and fusion analysis on the microseismic characteristic parameters and the visual characteristic parameters to obtain fusion analysis results. The fusion analysis results include at least the comprehensive index of surrounding rock stability and spatial information of risk areas. The early warning module 50 is used to determine the risk level and issue an early warning based on the fusion analysis results. In a specific embodiment, the fusion assessment module 40 includes: The spatiotemporal correlation unit is used to associate microseismic events and visually identified cracks with multiple independent analysis areas in the same 3D model of the tunnel, based on their spatial location and occurrence time. The feature extraction unit is used to extract a first type of feature from the associated microseismic events and a second type of feature from the associated crack information for each analysis region. The first type of feature includes microseismic energy release rate and event spatial clustering degree, and the second type of feature includes crack width expansion rate and crack network complexity. The fusion calculation unit is used to fuse the first type of features and the second type of features from the same analysis area to obtain the risk assessment value of each analysis area. The index synthesis unit is used to generate the comprehensive index of the surrounding rock stability state based on the risk assessment value of each analysis area, and to determine the spatial range of the analysis area where the risk assessment value exceeds a preset threshold as the spatial information of the risk area.

[0068] In a specific embodiment, the fusion computing unit includes: The vector combination subunit is used to combine the first type of features and the second type of features into a multi-dimensional input vector; The data input subunit is used to input the multidimensional input vector into a trained machine learning fusion model; The model processing subunit is used by the machine learning fusion model to calculate the correlation weights and nonlinear relationships between the first type of features and the second type of features in the multidimensional input vector, and output a risk assessment value.

[0069] In a specific embodiment, the machine learning fusion model is a neural network model, and the model processing subunit is specifically used for: The multidimensional input vector is input into the input layer of the neural network model, and the input layer maps the first type of features and the second type of features to different initial feature vectors respectively. In the hidden layer of the neural network model, the mutual attention weight between the initial feature vector corresponding to the first type of feature and the initial feature vector corresponding to the second type of feature is calculated through a cross-attention mechanism. Based on the attention weight, the two types of initial feature vectors are weighted, converged and fused to generate a fused intermediate vector. The fused intermediate vector is transformed by at least one fully connected layer and a nonlinear activation function to obtain a deep fused feature vector; The deep fusion feature vector is linearly mapped through the output layer to output the risk assessment value.

[0070] In a specific embodiment, the device further includes a model training module, which is used to perform training steps to obtain the machine learning fusion model, including: The dataset construction unit is used to build the training dataset. Each data sample includes the first and second types of features of a historical analysis region within a specific time period, as well as the actual stability state label of that region. The model training unit is used to train a selected machine learning algorithm with the combination of the first type of features and the second type of features as input and the actual stability state label as the target output, so as to obtain the machine learning fusion model.

[0071] In a specific embodiment, the spatiotemporal correlation unit is specifically used for: Spatial clustering of microseismic events is performed based on the source coordinates of the microseismic events to form multiple microseismic event clusters; For each cluster of microseismic events, its spatial projection range is determined on the surface of the three-dimensional model of the tunnel; Cracks that appear within a preset time window and whose spatial location falls within or is adjacent to the spatial projection range will be associated with the corresponding microseismic event clusters in the same analysis area.

[0072] In a specific embodiment, the device further includes a decision generation module, which is used to generate a support decision scheme, specifically including: The boundary determination unit is used to determine the three-dimensional boundary of the body to be supported in the three-dimensional model of the tunnel based on the spatial information of the risk area. The pattern matching unit is used to match the corresponding damage mechanism from a preset damage mode library based on the dominant damage feature type in the analysis area corresponding to the spatial information of the risk area. The scheme generation unit is used to combine the three-dimensional boundary of the body to be supported and the failure mechanism, query the support strategy knowledge base, and generate at least one support scheme that includes support type and design parameters.

[0073] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned tunnel surrounding rock instability early warning device and each unit can be referred to the corresponding description in the foregoing method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0074] The aforementioned tunnel surrounding rock instability early warning device can be implemented as a computer program, which can, for example... Figure 9 It runs on the computer device shown.

[0075] Please see Figure 9 , Figure 9 This is a schematic block diagram of a computer device provided in an embodiment of this application. The computer device 500 can be a terminal or a server. The terminal can be an electronic device with communication functions, such as a smartphone, tablet, laptop, desktop computer, personal digital assistant, or wearable device. The server can be a standalone server or a server cluster composed of multiple servers.

[0076] See Figure 9 The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a system bus 501. The memory may include a non-volatile storage medium 503 and internal memory 504.

[0077] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. When the computer program 5032 is executed, it causes the processor 502 to execute a method for early warning of tunnel surrounding rock instability.

[0078] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0079] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a tunnel surrounding rock instability early warning method.

[0080] This network interface 505 is used for network communication with other devices. Those skilled in the art will understand that... Figure 9 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device 500 to which the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0081] The processor 502 is used to run a computer program 5032 stored in the memory to perform the following steps: S1. Acquire the sequence of microseismic signals and surrounding rock surface images synchronously collected by sensors deployed along the tunnel; S2. Extract microseismic characteristic parameters that characterize the internal damage of the surrounding rock from the microseismic signals; S3. Extract visual feature parameters characterizing the damage to the surrounding rock surface from the image sequence; S4. Perform correlation and fusion analysis on the microseismic characteristic parameters and the visual characteristic parameters to obtain the fusion analysis results. The fusion analysis results include at least the comprehensive index of surrounding rock stability and spatial information of risk areas. S5. Based on the fusion analysis results, determine the risk level and issue an early warning.

[0082] It should be understood that in the embodiments of this application, the processor 502 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0083] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program may be stored in a storage medium, which is a computer-readable storage medium. The computer program is executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0084] Therefore, the present invention also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program. When executed by a processor, the computer program causes the processor to perform the following steps: S1. Acquire the sequence of microseismic signals and surrounding rock surface images synchronously collected by sensors deployed along the tunnel; S2. Extract microseismic characteristic parameters that characterize the internal damage of the surrounding rock from the microseismic signals; S3. Extract visual feature parameters characterizing the damage to the surrounding rock surface from the image sequence; S4. Perform correlation and fusion analysis on the microseismic characteristic parameters and the visual characteristic parameters to obtain the fusion analysis results. The fusion analysis results include at least the comprehensive index of surrounding rock stability and spatial information of risk areas. S5. Based on the fusion analysis results, determine the risk level and issue an early warning.

[0085] The storage medium is a physical, non-transient storage medium, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk, or any other physical storage medium capable of storing program code.

[0086] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0087] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0088] The steps in the method of this invention can be adjusted, merged, or reduced in order according to actual needs. The units in the device of this invention can be merged, divided, or reduced according to actual needs. Furthermore, the functional units in the various embodiments of this invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0089] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0090] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0091] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Since these modifications and variations fall within the scope of the claims and their equivalents, this invention also intends to include these modifications and variations.

[0092] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for early warning of tunnel surrounding rock instability, characterized in that, include: Acquire microseismic signals and surrounding rock surface image sequences synchronously collected by sensors deployed along the tunnel; Microseismic characteristic parameters characterizing internal damage to the surrounding rock are extracted from the microseismic signals; Visual feature parameters characterizing surface damage of the surrounding rock are extracted from the image sequence; The microseismic characteristic parameters and the visual characteristic parameters are correlated and fused to obtain the fusion analysis results. The fusion analysis results include at least the comprehensive index of surrounding rock stability and spatial information of risk areas. Based on the results of the fusion analysis, a risk level is determined and an early warning is issued.

2. The tunnel surrounding rock instability early warning method according to claim 1, characterized in that, The correlation and fusion analysis of the microseismic feature parameters and the visual feature parameters to obtain the fusion analysis results includes: Microseismic events and visually identified cracks are correlated with multiple independent analysis areas in the same 3D model of the tunnel based on their spatial location and occurrence time. For each analysis region, a first type of feature is extracted from the associated microseismic events, and a second type of feature is extracted from the associated crack information. The first type of feature includes the microseismic energy release rate and the spatial clustering of events, while the second type of feature includes the crack width expansion rate and the degree of crack network complexity. The first type of features and the second type of features from the same analysis area are fused and calculated to obtain the risk assessment value for each analysis area; Based on the risk assessment values ​​of each analysis area, a comprehensive index of the surrounding rock stability is generated, and the spatial range of the analysis area where the risk assessment value exceeds a preset threshold is determined as the spatial information of the risk area.

3. The tunnel surrounding rock instability early warning method according to claim 2, characterized in that, The step of fusing the first type of features and the second type of features from the same analysis region to obtain the risk assessment value for each analysis region includes: The first type of features and the second type of features are combined into a multi-dimensional input vector; The multidimensional input vector is then fed into a trained machine learning fusion model; The machine learning fusion model calculates the correlation weights and nonlinear relationships between the first type of features and the second type of features in the multidimensional input vector, and outputs a risk assessment value.

4. The tunnel surrounding rock instability early warning method according to claim 3, characterized in that, The machine learning fusion model is a neural network model. This model calculates the association weights and nonlinear relationships between the first and second types of features in the multidimensional input vector, and outputs a risk assessment value, including: The multidimensional input vector is input into the input layer of the neural network model, and the input layer maps the first type of features and the second type of features to different initial feature vectors respectively. In the hidden layer of the neural network model, the mutual attention weight between the initial feature vector corresponding to the first type of feature and the initial feature vector corresponding to the second type of feature is calculated through a cross-attention mechanism. Based on the attention weight, the two types of initial feature vectors are weighted, converged and fused to generate a fused intermediate vector. The fused intermediate vector is transformed by at least one fully connected layer and a nonlinear activation function to obtain a deep fused feature vector; The deep fusion feature vector is linearly mapped through the output layer to output the risk assessment value.

5. The tunnel surrounding rock instability early warning method according to claim 3 or 4, characterized in that, The machine learning fusion model is obtained through a training step, which includes: Construct a training dataset, in which each data sample includes the first and second types of features of a historical analysis region within a specific time period, as well as the actual stability state label of that region; Using the combination of the first type of features and the second type of features as input, and the actual stability state label as the target output, the selected machine learning algorithm is trained to obtain the machine learning fusion model.

6. The tunnel surrounding rock instability early warning method according to claim 2, characterized in that, The process of associating microseismic events with visually identified cracks, based on their spatial location and occurrence time, to multiple independent analysis regions within the same 3D tunnel model includes: Spatial clustering of microseismic events is performed based on the source coordinates of the microseismic events to form multiple microseismic event clusters; For each cluster of microseismic events, its spatial projection range is determined on the surface of the three-dimensional model of the tunnel; Cracks that appear within a preset time window and whose spatial location falls within or is adjacent to the spatial projection range will be associated with the corresponding microseismic event clusters in the same analysis area.

7. The tunnel surrounding rock instability early warning method according to claim 1, characterized in that, Also includes: The steps for generating a support decision plan are as follows: Based on the spatial information of the risk area, the three-dimensional boundary of the body to be supported is determined in the three-dimensional model of the tunnel; Based on the dominant damage characteristic type within the analysis area corresponding to the spatial information of the risk area, the corresponding damage mechanism is matched from the pre-set damage mode library; By combining the three-dimensional boundary of the body to be supported and the failure mechanism, the support strategy knowledge base is queried to generate at least one support scheme that includes support type and design parameters.

8. A tunnel surrounding rock instability early warning device, characterized in that, include: The data acquisition module is used to acquire microseismic signals and surrounding rock surface image sequences synchronously collected by sensors deployed along the tunnel. The microseismic data analysis module is used to extract microseismic characteristic parameters that characterize the internal damage of the surrounding rock from the microseismic signals. An image data analysis module is used to extract visual feature parameters characterizing damage to the surrounding rock surface from the image sequence; The fusion assessment module is used to perform correlation and fusion analysis on the microseismic characteristic parameters and the visual characteristic parameters to obtain fusion analysis results. The fusion analysis results include at least the comprehensive index of surrounding rock stability and spatial information of risk areas. The early warning module is used to determine the risk level and issue an early warning based on the fusion analysis results.

9. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the tunnel surrounding rock instability early warning method as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, can implement the tunnel surrounding rock instability early warning method as described in any one of claims 1 to 7.

Images