Roadway floor deformation monitoring and analyzing system

Abstract

The present application relates to the technical field of data monitoring, in particular to a roadway floor deformation monitoring and analyzing system, comprising: a data acquisition unit, which is used for synchronously acquiring real-time horizontal displacement data and real-time ground stress data of the roadway floor at each monitoring time through multiple groups of displacement sensors arranged on the roadway floor and stress monitoring devices arranged below the roadway floor; and a state determining unit, which is used for determining real-time vertical settlement and deformation state determining results of the roadway floor according to the real-time horizontal displacement data and the real-time ground stress data. The present application accurately maps the real-time horizontal displacement and the vertical settlement to a three-dimensional space model constructed by laser scanning through spatial alignment calibration, generates continuous three-dimensional deformation distribution characteristics, can clearly identify the position, range, shape and gradient change of the deformation area, makes up for the blind area of single-point monitoring, and provides reliable spatial basis for accurately positioning the hidden danger area and formulating targeted treatment scheme.

Description

Technical Field

[0001] This invention relates to the field of data monitoring technology, specifically to a tunnel floor deformation monitoring and analysis system. Background Technology

[0002] Currently, traditional systems typically rely on only a few discrete displacement sensors or single-point convergence meters deployed on the surface of the base plate. The deformation data obtained is limited to a finite number of measuring points and cannot reflect the overall deformation field of the base plate. For areas outside the measuring points, especially the blank areas between measuring points, the deformation status is completely unknown, resulting in obvious monitoring blind spots. When abnormal deformation of the base plate occurs off-point, the system may completely miss it, leading to the failure to detect hidden dangers in a timely manner. Furthermore, traditional methods cannot provide the spatial morphology of the deformation area, such as the range of the settlement channel, deformation gradient, and tilt direction, making it difficult to accurately locate the hidden danger area. During treatment, large-scale reinforcement can only be carried out based on experience, which is both costly and may miss critical parts.

[0003] Furthermore, traditional systems often rely solely on a single threshold set for displacement or displacement rate to determine hazard levels, failing to comprehensively consider the coupled effects of multiple factors on floor stability, such as the strength of the floor strata themselves, the convergence and constraint of the surrounding rocks on both sides, and groundwater seepage pressure. For example, if the displacement data does not exceed the limit but the convergence of the surrounding rocks intensifies or the rock strata strength is low, traditional systems may still classify it as safe, potentially ignoring the risk of structural instability. Conversely, they are prone to false alarms when displacement fluctuates due to temporary factors. At the same time, traditional systems rarely analyze the linkage between seepage pressure and deformation, and cannot provide timely warnings for water-induced weakened floor heaves, resulting in a lack of scientific basis for safety decisions and a high risk of underreporting or over-responding. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: a tunnel floor deformation monitoring and analysis system, comprising:

[0005] The data acquisition unit is used to simultaneously collect real-time horizontal displacement data and real-time ground stress data of the roadway floor at each monitoring moment through multiple sets of displacement sensors installed on the roadway floor and stress monitoring devices installed below the roadway floor.

[0006] The condition determination unit is used to determine the real-time vertical settlement and deformation condition determination results of the roadway floor based on real-time horizontal displacement data and real-time ground stress data.

[0007] The spatial alignment unit is used to acquire historical three-dimensional scanning data of the roadway roof and sidewalls collected by the laser scanning equipment deployed on the same roadway monitoring station when the deformation state determination result is abnormal deformation. Based on the historical three-dimensional scanning data and real-time horizontal displacement data, spatial alignment calibration is performed to obtain spatial alignment transformation parameters.

[0008] The feature distribution unit is used to determine the three-dimensional deformation distribution characteristics of the tunnel floor based on the spatial alignment transformation parameters, real-time horizontal displacement data, and real-time vertical settlement.

[0009] The safety assessment unit is used to obtain the rock stratum thickness parameters, floor bearing strength parameters, and convergence deformation of the surrounding rock on both sides of the roadway. Based on the three-dimensional deformation distribution characteristics, rock stratum thickness parameters, floor bearing strength parameters, and convergence deformation, the safety risk assessment level of the roadway floor is determined.

[0010] Preferably, based on real-time horizontal displacement data and real-time ground stress data, the real-time vertical settlement and deformation state determination results of the roadway floor are determined, including:

[0011] Extract the maximum and minimum horizontal displacement values ​​from the real-time horizontal displacement data and calculate the horizontal displacement difference.

[0012] The horizontal displacement difference and real-time ground stress data are input into the pre-trained settlement prediction model, which outputs the real-time vertical settlement. The settlement prediction model is trained using historical horizontal displacement difference data, historical ground stress data, and historical actual vertical settlement data.

[0013] Real-time horizontal displacement data and real-time vertical settlement are input into a pre-trained deformation recognition model. The deformation recognition model includes a temporal feature extraction network and a deformation determination network. The temporal feature extraction network extracts displacement trend feature vectors and change rate feature vectors, and the deformation determination network outputs predicted deformation results. The deformation state determination result is determined based on the predicted deformation results.

[0014] Preferably, spatial alignment calibration is performed based on historical 3D scan data and real-time horizontal displacement data to obtain spatial alignment transformation parameters, including:

[0015] The historical deformation and displacement data of the roadway floor collected by displacement sensors were acquired, and the historical three-dimensional scanning data of the roadway roof and sidewalls collected by laser scanning equipment were acquired.

[0016] Initial spatial alignment is performed based on historical deformation displacement data and historical 3D scan data to obtain initial alignment parameters;

[0017] The initial alignment parameters, displacement calibration matrix of the displacement sensor, historical deformation displacement data, and historical 3D scan data are input into the untrained deformation association model to obtain the predicted deformation mapping result corresponding to the historical 3D scan data; wherein, the predicted deformation mapping result includes the deformation prediction value of each scan point in the historical 3D scan data;

[0018] Based on the predicted deformation mapping results, the initial alignment parameters in the deformation association model are iteratively optimized until the iteration termination condition is met, thus obtaining the spatial alignment transformation parameters in the trained deformation association model.

[0019] Preferably, based on the predicted deformation mapping results, the initial alignment parameters in the deformation association model are iteratively optimized, including:

[0020] Obtain historical deformation mapping data corresponding to historical 3D scanning data in the deformation association model, and determine the spatial deviation cost based on historical deformation mapping data and historical deformation displacement data; wherein, historical deformation mapping data represents the mapping result of historical 3D scanning data in the displacement coordinate system of displacement sensor;

[0021] The deformation fitting loss value is determined based on the predicted deformation mapping results and the actual deformation measurement results corresponding to the historical 3D scanning data.

[0022] Based on the spatial deviation cost and the deformation fitting loss value, the initial alignment parameters in the deformation correlation model are iteratively optimized; wherein, the iteration termination conditions include the convergence of the spatial deviation cost value and / or the convergence of the deformation fitting loss value.

[0023] Preferably, the three-dimensional deformation distribution characteristics of the tunnel floor are determined based on spatial alignment transformation parameters, real-time horizontal displacement data, and real-time vertical settlement, including:

[0024] Based on the spatial alignment transformation parameters, the real-time horizontal displacement data and real-time vertical settlement are mapped onto the three-dimensional spatial model of the tunnel constructed by the laser scanning equipment to obtain three-dimensional deformation mapping data.

[0025] Based on the three-dimensional deformation mapping data, the three-dimensional deformation distribution characteristics of the tunnel floor are determined.

[0026] Preferably, obtaining the rock stratum thickness parameters of the tunnel floor, the bearing strength parameters of the floor, and the convergence deformation of the surrounding rock on both sides of the tunnel includes:

[0027] Historical 3D scanning data of the tunnel collected by laser scanning equipment and historical stress data of the tunnel sidewall collected by pressure sensors installed on the tunnel sidewall are obtained; wherein, the laser scanning equipment and the pressure sensors are set on the same tunnel monitoring station;

[0028] Historical 3D scan data and historical stress data are input into a pre-trained surrounding rock analysis model to obtain surrounding rock parameter data. Among them, the surrounding rock parameter data includes the surrounding rock attribute parameters corresponding to each scan point in the historical 3D scan data, and the surrounding rock attribute parameters include rock strata, floor and sidewalls.

[0029] Based on the surrounding rock parameter data, determine the rock stratum thickness parameters of the tunnel floor, the bearing strength parameters of the floor, and the convergence deformation of the surrounding rock on both sides of the tunnel.

[0030] Preferably, the surrounding rock analysis model includes a scanning feature extraction network, a stress feature extraction network, and a parameter fusion output network;

[0031] Correspondingly, historical 3D scanning data and historical stress data are input into the pre-trained surrounding rock analysis model to obtain surrounding rock parameter data;

[0032] Spatial structure feature vectors are obtained by extracting features from historical 3D scanning data using a scanning feature extraction network.

[0033] Mechanical feature vectors are obtained by extracting features from historical stress data using a stress feature extraction network.

[0034] The parameter fusion output network performs feature fusion on the spatial structure feature vector and the mechanical feature vector to obtain a comprehensive feature vector. Based on the comprehensive feature vector, the surrounding rock parameter data is output.

[0035] Preferably, the system further includes:

[0036] Based on the hydrological monitoring equipment deployed around the tunnel floor to collect seepage pressure data of the tunnel floor, the hydrological monitoring equipment is installed at a predetermined depth on the surface of the tunnel floor or on the side wall of the tunnel floor.

[0037] If the seepage pressure data exceeds the set seepage pressure threshold, a seepage risk warning will be triggered and a comprehensive analysis will be conducted in conjunction with the safety risk assessment level.

[0038] If the seepage pressure data is less than or equal to the set seepage pressure threshold, the analysis will be based solely on the safety risk assessment level.

[0039] The monitoring and analysis results of the tunnel floor are presented on the remote monitoring terminal interface. The monitoring and analysis results include at least one of the following: deformation level category at each monitoring time, real-time horizontal displacement data, real-time vertical settlement, safety risk assessment level, or seepage risk early warning information.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] (1) This invention uses spatial alignment calibration to accurately map real-time horizontal displacement and vertical settlement to a three-dimensional spatial model constructed by laser scanning, generating continuous three-dimensional deformation distribution features. It can clearly identify the location, range, shape and gradient changes of the deformation area, make up for the blind spots of single-point monitoring, and provide a reliable spatial basis for accurately locating hidden danger areas and formulating targeted treatment plans.

[0042] (2) This invention uses displacement and geostress data to determine the deformation state, and combines the rock layer thickness, bearing strength of each layer and convergence deformation of both sides obtained by surrounding rock analysis. At the same time, it links seepage pressure monitoring to achieve a comprehensive evaluation of the stability of the bottom plate affected by geological conditions, surrounding rock deformation and hydrological factors. When seepage exceeds the threshold, an early warning is automatically triggered and linked with the risk level analysis. It can distinguish the risk response strategies under different working conditions, avoid false alarms or omissions of single indicators, and thus provide more scientific and timely support for safe decision-making in roadway mining. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the overall system architecture in one embodiment of the present invention;

[0044] Figure 2 This is a schematic flowchart of the deformation state determination steps in one embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram of the steps for spatial alignment transformation parameters in one embodiment of the present invention.

[0046] In the diagram: 1. Data acquisition unit; 2. Status determination unit; 3. Spatial alignment unit; 4. Feature distribution unit; 5. Security assessment unit. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Example 1, please refer to Figure 1 - Figure 3 This invention provides a technical solution: a tunnel floor deformation monitoring and analysis system, comprising:

[0049] Data acquisition unit 1 is used to simultaneously acquire real-time horizontal displacement data and real-time ground stress data of the roadway floor at each monitoring moment through multiple sets of displacement sensors installed on the roadway floor and stress monitoring devices installed below the roadway floor.

[0050] State determination unit 2 is used to determine the real-time vertical settlement and deformation state determination results of the roadway floor based on real-time horizontal displacement data and real-time ground stress data.

[0051] The spatial alignment unit 3 is used to acquire historical three-dimensional scanning data of the roadway roof and sidewalls collected by the laser scanning equipment deployed on the same roadway monitoring station when the deformation state determination result is abnormal deformation. Based on the historical three-dimensional scanning data and real-time horizontal displacement data, spatial alignment calibration is performed to obtain spatial alignment transformation parameters.

[0052] Feature distribution unit 4 is used to determine the three-dimensional deformation distribution characteristics of the tunnel floor based on spatial alignment transformation parameters, real-time horizontal displacement data and real-time vertical settlement.

[0053] Safety assessment unit 5 is used to obtain the rock stratum thickness parameters, floor bearing strength parameters, and convergence deformation of the surrounding rock on both sides of the roadway. Based on the three-dimensional deformation distribution characteristics, rock stratum thickness parameters, floor bearing strength parameters, and convergence deformation, the safety risk assessment level of the roadway floor is determined.

[0054] It should be noted that the underground mining tunnel of the metal mine is about 600 meters deep and 5.2 meters wide. At the center line of the tunnel floor and 0.8 meters on both sides, a set of inductive displacement sensors is installed every 10 meters along the tunnel direction, for a total of 5 monitoring sections. Each set contains 3 horizontal monitoring points to continuously record the horizontal displacement of the floor rock mass along the tunnel in both the transverse and longitudinal directions. At the same time, vibrating wire geostress gauges are installed 3 meters, 5 meters and 8 meters below the floor of each monitoring section to synchronously collect six-component geostress data at a sampling frequency of once per minute. During a routine monitoring, the horizontal displacement rate at the 3rd section suddenly jumped from 0.3 mm per day to 1.7 mm per day. At the same time, the direction of the maximum principal stress at this point was deflected by an angle of 12 degrees, and the minimum principal stress value decreased by about 0.8 MPa.

[0055] After data preprocessing, based on the elastic mechanical relationship between the displacement field and the stress field, and combined with the Poisson's ratio and elastic modulus of the floor strata, the real-time vertical settlement at each measuring point on the floor surface at this section was calculated. The maximum settlement value appeared in the center of the floor, reaching 4.2 mm. The measured horizontal strain increment was compared with the preset critical strain threshold of 0.002, and combined with the stress unloading coefficient, it was determined that the area had entered the yielding stage, and the result was determined to be abnormal deformation. The critical strain threshold of 0.002 was determined based on the lithological characteristics of the roadway floor strata and indoor mechanical tests.

[0056] Indoor test calibration: Before or in the early stage of tunnel construction, collect representative rock core samples of the floor, such as silty mudstone, fine sandstone, mudstone, etc., and carry out uniaxial / triaxial compression tests to obtain the yield strain of each rock layer and the strain value corresponding to the end point of the elastic segment in the stress-strain curve; 0.002 is the typical value of the yield strain of the main rock layer of the tunnel floor.

[0057] Engineering experience correction: Based on long-term monitoring data of similar mine roadways, the indoor test values ​​are corrected to reflect the influence of actual conditions such as field scale effect and joints and fissures.

[0058] The stress unloading factor is used to quantify the degree of weakening of the base rock mass strength by stress release. Its value is determined as follows:

[0059] The unloading coefficient is calculated as: Current measured minimum principal stress / Minimum principal stress in the initial geostress field. When this coefficient is less than 1, it indicates that the floor rock mass is in an unloaded state and its strength has decreased. The initial geostress field is obtained through geostress testing before tunnel excavation. After construction, real-time geostress data is continuously collected through a stress monitoring device under the floor. The unloading coefficient is obtained by dividing the real-time minimum principal stress by the initial value. When the horizontal strain increment is ≥0.002 and the stress unloading coefficient is lower than a certain lower threshold, the floor is judged to have entered the yielding stage.

[0060] Then, historical point cloud data of the tunnel roof and sidewalls obtained by the pulsed three-dimensional laser scanner installed at the same monitoring station in the past 72 hours were retrieved. The point cloud data was spatially aligned and calibrated with the horizontal displacement data measured by the floor displacement sensor at the current moment. The calibration process used a reference prism fixed in the tunnel as a reference and solved the rotation matrix and translation vector by the iterative nearest point algorithm to obtain the spatial alignment transformation parameters, so that the floor displacement coordinates and the roof and sidewall point cloud coordinates are unified to the same spatial reference system.

[0061] Using this transformation parameter, real-time horizontal displacement data and vertical settlement are mapped to a three-dimensional spatial grid. Interpolation generates a three-dimensional deformation field of the entire cross-section of the floor slab. The abnormal area is elliptical, with the major axis extending longitudinally along the roadway for about 2.6 meters and the minor axis spreading laterally for about 1.9 meters. The maximum subsidence point is biased towards the left side of the roadway for about 0.3 meters, and there is a horizontal displacement zone at a depth of 3 meters in the floor slab.

[0062] The rock strata thickness parameters obtained from the borehole cores at this section were collected. The bottom plate consists of an alternating layer structure of 0.8 m thick silty mudstone, 2.1 m thick fine sandstone, and 1.4 m thick mudstone. The bearing strength parameters of the bottom plate were obtained from indoor uniaxial compression tests, which showed that the uniaxial compressive strength of silty mudstone was 28 MPa, fine sandstone was 45 MPa, and mudstone was 12 MPa. The convergence deformation of the surrounding rock on both sides of the tunnel was obtained by multi-point displacement gauges installed on the sidewalls. The cumulative convergence of the left sidewall was 15 mm, and the cumulative convergence of the right sidewall was 13 mm. The convergence rate of both sideswalls increased by 0.3 times in the past week.

[0063] The maximum settlement, deformation gradient, and fault displacement from the three-dimensional deformation distribution characteristics, combined with the bearing capacity weakening coefficient of the rock strata assemblage and the weakening effect of the convergence of the two sides on the constraint of the arch foot of the foundation, are input into the trained risk assessment model; whereby the value of the bearing capacity weakening coefficient is based on the mechanical properties of the foundation rock strata assemblage and the theory of progressive failure of the surrounding rock:

[0064] The base plate is typically composed of multiple layers of rock, such as 0.8m of silty mudstone, 2.1m of fine sandstone, and 1.4m of mudstone. Each layer has different compressive strengths due to its different lithology. The bearing capacity weakening coefficient comprehensively reflects the interlayer synergistic weakening effect of the multi-layer structure under overall load. The ultimate bearing capacity is directly proportional to the cohesion of the rock mass and decreases with the decrease of the internal friction angle. When the reduction coefficient of the progressive failure parameter of the surrounding rock is less than 1, the ultimate bearing capacity decreases at a decreasing rate with the reduction coefficient. Through numerical simulation (such as FLAC3D), the strength reduction analysis of each layer of the base plate is performed. The sudden displacement of the characteristic point of the base plate is used as the failure criterion to solve the stability safety factor and potential fracture surface of each layer. Finally, the overall bearing capacity weakening coefficient is comprehensively determined, which is usually between 0.6 and 0.9.

[0065] The risk assessment model is generated using a supervised learning method.

[0066] Sample construction: Historical monitoring data are used as samples. Each sample contains input features, such as the maximum settlement, deformation gradient, and displacement in the three-dimensional deformation distribution features, the bearing capacity weakening coefficient, the weakening effect parameters of the convergence of the two sides on the constraint of the bottom arch foot, and output labels. The safety risk level is marked by experts according to the actual bottom drum damage situation, ranging from level 1 to level 5.

[0067] Model selection: Use multi-class neural network or random forest algorithms; the neural network structure can be designed as: input layer → fully connected layer → Dropout layer → fully connected layer → output layer, outputting the probability distribution of 5 risk levels;

[0068] Training and Validation: The samples are divided into training and test sets in a 7:3 ratio. Cross-validation is used to adjust hyperparameters such as learning rate, batch size, and regularization coefficient. Classification accuracy and F1-score are used as evaluation metrics until the model achieves satisfactory prediction accuracy on the test set.

[0069] The risk assessment model classifies safety risks into five levels, with level 1 being stable and level 5 being extremely dangerous. In this case, the output result was level 4, i.e., high risk. Specifically, tension cracks appeared in the local area of ​​the floor, and the deep mudstone layer showed a tendency to soften and plastically flow, which could lead to floor heave instability within 3 to 5 days. Based on this, the mine's technical department immediately took reinforcement measures, including constructing anchor-injection combined support for the floor in the abnormal area, reducing the intensity of blasting operations in nearby mining areas, and increasing the monitoring frequency to once every 10 seconds. The subsequent three-day tracking data showed that the deformation rate gradually returned to the safe range, verifying the rationality of the assessment results.

[0070] In an optional embodiment, the real-time vertical settlement and deformation state determination results of the roadway floor are determined based on real-time horizontal displacement data and real-time ground stress data, including:

[0071] B1. Extract the maximum and minimum horizontal displacement values ​​from the real-time horizontal displacement data and calculate the horizontal displacement difference.

[0072] B2. Input the horizontal displacement difference and real-time ground stress data into the pre-trained settlement prediction model and output the real-time vertical settlement; wherein, the settlement prediction model is obtained by training through historical horizontal displacement difference data, historical ground stress data and historical actual vertical settlement data.

[0073] B3. Input the real-time horizontal displacement data and real-time vertical settlement into the pre-trained deformation recognition model. The deformation recognition model includes a temporal feature extraction network and a deformation determination network. The temporal feature extraction network extracts displacement trend feature vectors and change rate feature vectors, and the deformation determination network outputs predicted deformation results. The deformation state determination results are determined based on the predicted deformation results.

[0074] It should be noted that in the third monitoring section of a mine roadway, five horizontal displacement sensors, numbered 1 to 5, were evenly spaced on the bottom surface. At a certain synchronous acquisition time, the horizontal displacement values ​​measured at each point were 1.2 mm, 1.8 mm, 2.5 mm, 2.0 mm, and 1.5 mm, respectively. The maximum horizontal displacement value of 2.5 mm and the minimum horizontal displacement value of 1.2 mm were extracted from this set of real-time horizontal displacement data, and the difference between the two was calculated to obtain a horizontal displacement difference of 1.3 mm. At the same time, the geostress gauge buried under the bottom plate of this section measured the real-time maximum principal stress as 18.6 MPa, the minimum principal stress as 7.2 MPa, and the intermediate principal stress as 11.4 MPa.

[0075] The horizontal displacement difference of 1.3 mm and the three principal stress values ​​mentioned above are combined into a feature vector and input into the pre-trained settlement prediction model.

[0076] Settlement prediction models employ gradient boosting regression trees, such as XGBoost, CatBoost, or LightGBM, as detailed below:

[0077] Input features: Horizontal displacement difference (maximum horizontal displacement value - minimum horizontal displacement value) + real-time geostress data (maximum principal stress, minimum principal stress, intermediate principal stress, etc.), which constitute the feature vector;

[0078] Output label: Vertical settlement measured by a precision level;

[0079] Sample size: 2,000 historical samples accumulated from the mine over the past two years;

[0080] Algorithm selection: XGBoost regressor or CatBoost / LightGBM is used. Core hyperparameters include:

[0081] n_estimators (number of trees): 300~500;

[0082] max_depth (maximum depth of the tree): 3~8;

[0083] learning_rate: 0.01~0.1;

[0084] reg_lambda (L2 regularization coefficient): 1~10;

[0085] subsample ratio: 0.7~0.9;

[0086] Hyperparameter optimization: Use grid search or Bayesian optimization, and use the mean square error (MSE) of K-fold cross-validation (K=5 or 10) as the selection criterion to find the parameter combination with the highest accuracy;

[0087] Training process: The 2000 samples are divided into training and validation sets in an 8:2 ratio. The training set is used for model fitting, and the validation set is used for early stopping to prevent overfitting.

[0088] The model performance was evaluated using the coefficient of determination (R²) and mean squared error (MSE).

[0089] The closer R² is to 1, the better the fit.

[0090] The smaller the MSE, the higher the prediction accuracy;

[0091] When the validation set MSE no longer decreases for several consecutive rounds, stop training and save the model;

[0092] In real-time monitoring, the current horizontal displacement difference and ground stress data are input into the trained model, and the real-time vertical settlement is calculated and output.

[0093] The settlement prediction model was trained based on 2,000 sets of historical samples accumulated in the mine over the past two years. The historical samples included different horizontal displacement differences, different combinations of ground stress, and their corresponding vertical settlement measured by a precision level. The model adopted a gradient boosting regression tree structure and determined the optimal hyperparameters through cross-validation. After inputting the current features, the model forward calculated and output the real-time vertical settlement of 3.7 mm.

[0094] The real-time horizontal displacement data sequence of 60 consecutive monitoring moments, together with the vertical settlement sequence corresponding to each moment obtained just now, are input into the pre-trained deformation recognition model. The deformation recognition model is composed of a temporal feature extraction network and a deformation judgment network connected in series. The temporal feature extraction network adopts a two-layer bidirectional long short-term memory network with 128 hidden units in each layer, an input sequence length of 60, a feature dimension of 6, and includes 5 horizontal displacement measurement points and 1 vertical settlement measurement.

[0095] The network performs forward and backward recursive operations on the sequence, outputting a displacement trend feature vector and a rate of change feature vector at the last time step. The trend feature vector is 256-dimensional, and the rate feature vector is 128-dimensional. The deformation determination network consists of three fully connected layers. The first layer receives the concatenated 384-dimensional features and activates them using ReLU. The second layer compresses the features to 64 dimensions, and the third layer outputs 2-dimensional logical values, corresponding to the unnormalized scores of normal and abnormal deformations, respectively. By inferring the sequence of the cross-section at the current time step, the abnormal deformation score is found to be 0.87, and the normal deformation score is 0.13. After softmax mapping, the abnormal probability reaches 87 percentage points.

[0096] According to the preset judgment threshold, when the abnormal probability exceeds 65 percentage points, it is judged as an abnormal deformation. In this case, the abnormal probability of 87 percentage points far exceeds the threshold, so the deformation status judgment result is abnormal deformation. This judgment result triggers an alarm and automatically stores all the above input data and intermediate calculation results into the database for subsequent analysis.

[0097] In an optional embodiment, spatial alignment calibration is performed based on historical 3D scan data and real-time horizontal displacement data to obtain spatial alignment transformation parameters, including:

[0098] C1. Obtain historical deformation displacement data of the roadway floor collected by displacement sensors, and obtain historical three-dimensional scanning data of the roadway roof and sidewalls collected by laser scanning equipment.

[0099] C2. Perform initial spatial alignment based on historical deformation displacement data and historical 3D scan data to obtain initial alignment parameters;

[0100] C3. Input the initial alignment parameters, the displacement calibration matrix of the displacement sensor, the historical deformation displacement data, and the historical 3D scan data into the untrained deformation association model to obtain the predicted deformation mapping result corresponding to the historical 3D scan data; wherein, the predicted deformation mapping result includes the deformation prediction value of each scan point in the historical 3D scan data;

[0101] C4. Based on the predicted deformation mapping results, iteratively optimize the initial alignment parameters in the deformation association model until the iteration termination condition is met, thus obtaining the spatial alignment transformation parameters in the trained deformation association model.

[0102] It should be noted that at the No. 2 monitoring station of a certain mine roadway, the historical deformation and displacement data of the floor plate collected by the displacement sensor of the station in the past 30 days were retrieved from the database. The data includes displacement records at 10-minute intervals in both the transverse and longitudinal directions of the roadway, totaling 4,320 time points. At the same time, the historical three-dimensional scan data obtained by the laser scanning equipment installed on the arch during the same period were retrieved. The scanning frequency was once every 2 hours, and a total of 360 point clouds were obtained. Each point cloud contains about 500,000 scan points, covering the curved surface of the roof and the left and right sidewalls.

[0103] The historical deformation displacement data and historical 3D scanning data were initially spatially aligned. The initial spatial alignment used three target spheres fixed on the permanent reference point of the roadway floor as a common reference. The coordinate changes of the target spheres recorded by the displacement sensor and the geometric center extraction results of the target spheres in the laser point cloud were used to calculate the initial spatial alignment parameters by least squares fitting. These parameters included two angular offsets of 0.05 degrees and -0.03 degrees of rotation around the roadway axis, and three translations of 12.7 mm, -5.2 mm and 3.8 mm of longitudinal, transverse and vertical directions along the roadway, as well as a scale factor of 0.9997.

[0104] After obtaining the initial alignment parameters, they are input together with the displacement calibration matrix of the displacement sensor itself into the deformation association model that has not yet been trained. This model adopts a graph neural network-based architecture, with the initial layer containing three spatial graph convolutional layers, each with a receptive field radius of 0.5 meters. The displacement calibration matrix is ​​a 5x5 diagonal matrix with diagonal elements of 1.02, 0.98, 1.01, 1.03, and 0.97, used to correct measurement gain errors caused by differences in installation angles at each measuring point. The model uses each scanning point in the historical 3D scanning data as a graph node, with the node features being the three... The model uses 3D coordinates, with edge weights determined by the spatial distance between scan points. It receives historical deformation displacement data as boundary constraints and calculates the predicted deformation value for each scan point through forward propagation. For example, for a scan point on the top plate with original coordinates X = -1.2 meters, Y = 0.0 meters, and Z = 3.5 meters, the model's first iteration outputs a predicted deformation value of -0.6 millimeters vertically, 0.3 millimeters horizontally, and -0.1 millimeters vertically. All 500,000 scan points in the point cloud receive their respective predicted deformation values, forming the predicted deformation mapping result.

[0105] The objective function is defined as the mean square error between the predicted deformation mapping result and the displacement field interpolated at the actual displacement sensor measurement points, plus a regularization term to constrain the smooth change of the alignment parameters. In each iteration, the gradient of the rotation angle, translation amount, and scale coefficient in the initial alignment parameters is calculated based on the current prediction error. An adaptive moment estimation optimizer is used, with a learning rate of 0.001 and a batch size of 16 point clouds. After the first iteration, the total error is 3.42 square millimeters. After the second iteration, the error decreases to 2.18 square millimeters, the fifth iteration to 1.05 square millimeters, the tenth iteration to 0.63 square millimeters, the twentieth iteration to 0.31 square millimeters, and the thirty-fifth iteration to 0.19 square millimeters. When the iteration reaches the 48th iteration, the error change is less than the set termination threshold of 0.01 square millimeters for three consecutive iterations, satisfying the iteration termination condition.

[0106] At this point, the iteration is stopped, and the final alignment parameters are saved as the spatial alignment transformation parameters in the trained deformation association model. These parameters are rotation angles of 0.04 degrees and -0.02 degrees, translation amounts of 13.1 mm, -5.6 mm, and 4.2 mm, and a scale factor of 1.0001. These parameters are then used in subsequent real-time monitoring spatial alignment operations to ensure that the displacement sensor data and laser scanning data are in the same coordinate system, providing an accurate benchmark for subsequent extraction of three-dimensional deformation distribution features.

[0107] In an optional embodiment, the initial alignment parameters in the deformation association model are iteratively optimized based on the predicted deformation mapping results, including:

[0108] Obtain historical deformation mapping data corresponding to historical 3D scanning data in the deformation association model, and determine the spatial deviation cost based on historical deformation mapping data and historical deformation displacement data; wherein, historical deformation mapping data represents the mapping result of historical 3D scanning data in the displacement coordinate system of displacement sensor;

[0109] The deformation fitting loss value is determined based on the predicted deformation mapping results and the actual deformation measurement results corresponding to the historical 3D scanning data.

[0110] Based on the spatial deviation cost and the deformation fitting loss value, the initial alignment parameters in the deformation correlation model are iteratively optimized; wherein, the iteration termination conditions include the convergence of the spatial deviation cost value and / or the convergence of the deformation fitting loss value.

[0111] It should be noted that at the No. 5 monitoring station in a mine roadway, a forward propagation of the deformation association model has been completed, obtaining the predicted deformation value corresponding to each scan point in the historical 3D scan data, i.e., the predicted deformation mapping result. At the same time, the model also contains intermediate outputs, which are the sets of mapped coordinates of each scan point in the historical 3D scan data after being transformed by the current alignment parameters and projected onto the displacement coordinate system where the displacement sensor is located. This set is called historical deformation mapping data. The mapping data contains 360 sets of point clouds, each set containing approximately 500,000 mapping points, and each mapping point has its 3D coordinates in the displacement coordinate system.

[0112] The historical deformation mapping data was acquired and compared with the historical deformation displacement data actually recorded by the displacement sensor at the same time. The displacement sensor has 5 measuring points, and each measuring point recorded displacement values ​​at 4320 times over a historical 30-day period. In the displacement coordinate system, the mapped coordinates of each scanning point were spatially matched with the position of the nearest displacement sensor measuring point, the Euclidean distance between them was calculated, and multiplied by the confidence weight coefficient of the displacement sensor in that direction. The average value of all matching results of all 360 sets of point clouds was calculated to obtain the spatial deviation cost. In the first iteration, the cost value was 2.87 mm. The specific calculation method was as follows: for each time point, the square root of the sum of the squares of the lateral deviation, longitudinal deviation, and vertical deviation between the mapped point and the adjacent measuring point was calculated, and then the average was calculated over all times and all measuring points.

[0113] Using an independent high-precision total station, 18 fixed monitoring markers on the roof and sidewalls of the tunnel were periodically measured manually. The actual three-dimensional displacement changes of these markers were used as the actual deformation measurement results. During each scan, the predicted deformation values ​​corresponding to these marker positions in the predicted deformation mapping results were extracted and subtracted from the actual deformation values ​​measured by the total station. The mean absolute error of all markers at all times was calculated to obtain the deformation fitting loss value. The loss value was 0.42 mm in the first iteration.

[0114] The spatial deviation cost of 2.87 mm and the deformation fitting loss of 0.42 mm were weighted and summed, with weights set to 0.6 and 0.4 respectively, resulting in a comprehensive loss of 1.89 mm. Subsequently, gradient descent was used to adjust the initial alignment parameters, which included a rotation angle of 0.03 degrees around the X-axis, a rotation angle of -0.01 degrees around the Y-axis, a translation of 10.5 mm laterally, -4.3 mm longitudinally, and 3.2 mm vertically, and a scale factor of 0.9998. After the first iteration, the comprehensive loss decreased to 1.54 mm; after the second iteration, it decreased to 1.21 mm; after the fifth iteration, it was 0.76 mm; after the tenth iteration, it was 0.43 mm; after the fifteenth iteration, it was 0.28 mm; and after the eighteenth iteration, it was 0.19 mm.

[0115] During the iteration process, the changes in spatial deviation cost and deformation fitting loss value are monitored simultaneously. The iteration termination condition is set as follows: the change in spatial deviation cost value is less than 0.02 mm for two consecutive iterations, or the change in deformation fitting loss value is less than 0.01 mm for two consecutive iterations. In the 21st iteration, the spatial deviation cost value is 0.051 mm, which is 0.012 mm from 0.063 mm in the 20th iteration, and is still less than 0.02 mm. The deformation fitting loss value in the 21st iteration is 0.008 mm, which is 0.003 mm from 0.011 mm in the 20th iteration, and is less than 0.01 mm, satisfying the convergence condition of deformation fitting loss value. Therefore, the iteration is terminated after the 21st iteration, and the current alignment parameters are output as the spatial alignment transformation parameters in the trained deformation association model. The final parameters are rotation angles of 0.04 degrees and -0.02 degrees, translation amounts of 12.8 mm, -5.1 mm, and 4.0 mm, and a scale factor of 1.0000.

[0116] The entire iterative optimization process takes into account both the geometric consistency of spatial alignment and the accuracy of deformation prediction, ensuring that the final parameters can reduce coordinate system deviation and improve deformation prediction accuracy. When this set of parameters is used for spatial alignment and deformation mapping calculations in subsequent real-time monitoring, the experimental results show that the overall error has decreased significantly compared with the initial state, verifying the effectiveness of the joint optimization strategy.

[0117] In an optional embodiment, the three-dimensional deformation distribution characteristics of the tunnel floor are determined based on spatial alignment transformation parameters, real-time horizontal displacement data, and real-time vertical settlement, including:

[0118] Based on the spatial alignment transformation parameters, the real-time horizontal displacement data and real-time vertical settlement are mapped onto the three-dimensional spatial model of the tunnel constructed by the laser scanning equipment to obtain three-dimensional deformation mapping data.

[0119] Based on the three-dimensional deformation mapping data, the three-dimensional deformation distribution characteristics of the tunnel floor are determined.

[0120] It should be noted that the spatial alignment transformation parameters for the fourth monitoring section of the mine roadway have been trained and saved. The specific parameters are: rotation angle around the roadway strike axis of 0.04 degrees, rotation angle around the vertical axis of -0.02 degrees, lateral translation of 12.8 mm, longitudinal translation of -5.1 mm, vertical translation of 4.0 mm, and scale factor of 1.0001. Five displacement sensor points, numbered A to E, are installed on the bottom plate of this section. At a given real-time monitoring moment, the real-time horizontal displacement data for each point includes both lateral and longitudinal displacements. The vertical settlement has been determined by the settlement prediction model. Synchronous output; Measuring point A: lateral displacement 1.2 mm, longitudinal displacement 0.3 mm, vertical settlement 3.7 mm; Measuring point B: lateral displacement 0.8 mm, longitudinal displacement -0.1 mm, vertical settlement 4.2 mm; Measuring point C: lateral displacement 1.5 mm, longitudinal displacement 0.5 mm, vertical settlement 3.9 mm; Measuring point D: lateral displacement 0.9 mm, longitudinal displacement -0.2 mm, vertical settlement 4.5 mm; Measuring point E: lateral displacement 0.4 mm, longitudinal displacement 0.1 mm, vertical settlement 3.2 mm.

[0121] The original three-dimensional coordinates of these measuring points in their own local coordinate system, i.e., the spatial position of each measuring point at installation time, are added to the aforementioned real-time displacement to obtain the actual three-dimensional coordinates of each measuring point at the current moment. Then, spatial alignment transformation parameters are applied to rotate, translate, and scale the actual coordinates of each measuring point, transforming them into the overall three-dimensional spatial model of the tunnel constructed by the laser scanning equipment. After transformation, the new coordinates of measuring point A are X = -1.8 meters, Y = 0.2 meters, and Z = 2.9 meters; the new coordinates of measuring point B are... X equals -0.9 meters, Y equals 0.1 meters, and Z equals 2.8 meters; the new coordinates of measuring point C are X equals 0.0 meters, Y equals 0.3 meters, and Z equals 2.7 meters; the new coordinates of measuring point D are X equals 0.9 meters, Y equals -0.1 meters, and Z equals 2.6 meters; the new coordinates of measuring point E are X equals 1.8 meters, Y equals 0.0 meters, and Z equals 2.5 meters; these transformed point coordinates, together with their corresponding three displacement components, namely the lateral, longitudinal, and vertical deformation values, constitute the initial three-dimensional deformation mapping data;

[0122] Since only 5 discrete measuring points are insufficient to describe the continuous deformation field of the entire floor surface, a dense triangular mesh of the floor area in the 3D spatial model of the tunnel was used. The mesh spacing was set to 0.1 meters, the entire floor area had a transverse width of 5.2 meters and a longitudinal length of 10 meters, generating approximately 5400 mesh nodes. For each mesh node, the three nearest measuring points were selected, and an inverse distance weighted interpolation method was used, with the weight inversely proportional to the square of the distance. The interpolation yielded the transverse, longitudinal, and vertical deformation values ​​at that node. During the interpolation process, the surface elevation of the floor area in the laser scanning point cloud was also introduced as a terrain constraint to ensure that the interpolation results were consistent with the actual geometry of the floor. Finally, the 3D deformation values ​​at all mesh nodes were obtained, forming complete 3D deformation mapping data, which was stored in point cloud format. Each node contained six fields: X, Y, Z coordinates and corresponding transverse displacement, longitudinal displacement, and vertical settlement.

[0123] Based on the three-dimensional deformation mapping data, the three-dimensional deformation distribution characteristics of the tunnel floor were determined. The spatial distribution of vertical settlement was extracted. The maximum settlement value occurred at grid coordinates X = -0.3 m and Y = 0.1 m, with a settlement of 4.7 mm. This location corresponds to about 0.3 m to the left side of the center of the floor, which coincides with the vicinity of sensor measuring point D. In the longitudinal direction, the settlement gradually increases and then decreases from the upstream end to the downstream end, forming an arched settlement trough. The longitudinal influence range of the settlement trough is about 6.2 m, and the lateral influence range is about 4.0 m. In terms of lateral deformation gradient, the lateral displacement of the left side points inward into the tunnel, with a value of 1.3 mm, while the lateral displacement of the right side points outward, with a value of 0.8 mm, indicating that the floor has a tendency to be squeezed to both sides. The maximum lateral strain occurs 0.5 m from the left side, with a strain value of about 0.0008. The longitudinal displacement distribution shows that the two ends of the floor are squeezed towards the middle. The longitudinal displacement at the upstream end is -0.4 mm, the longitudinal displacement at the downstream end is +0.3 mm, and the longitudinal displacement in the middle is close to zero, indicating the existence of a longitudinal compression zone.

[0124] The total displacement vector of each grid node was calculated, i.e., the combined displacement in the lateral, longitudinal, and vertical directions. It was found that the area with the largest total displacement was concentrated within a radius of 0.8 meters around the maximum settlement point, with the total displacement value ranging from 5.0 mm to 5.8 mm. The deformation contour map showed that the vertical settlement contour lines were elliptical, with a major-to-minor axis ratio of approximately 1.4:1, and the major axis along the direction of the tunnel. The lateral inclination rate of the floor slab, i.e., the settlement difference per meter of width, was 1.2 mm per meter on the left side and 0.9 mm per meter on the right side, showing asymmetry and indicating that the floor slab deformation had a biased pressure characteristic. At the same time, subtracting the three-dimensional deformation from the original floor slab surface revealed the unevenness of the floor slab surface. A significant concave surface appeared near the maximum settlement point, with a concave depth of 4.7 mm and an area of ​​approximately 3.1 square meters.

[0125] These three-dimensional deformation distribution features are summarized, including the maximum settlement value and its location, the range of the settlement trough, the transverse and longitudinal strain gradients, the maximum total displacement, the inclination rate, and the area of ​​the concave region, and used as the basic input for subsequent safety risk assessment. The entire mapping and feature extraction process is completed based on spatial alignment transformation parameters, ensuring accurate consistency between displacement sensor data and laser scanning three-dimensional spatial model.

[0126] In an optional embodiment, obtaining the rock stratum thickness parameters of the tunnel floor, the bearing strength parameters of the floor, and the convergence deformation of the surrounding rock on both sides of the tunnel includes:

[0127] Historical 3D scanning data of the tunnel collected by laser scanning equipment and historical stress data of the tunnel sidewall collected by pressure sensors installed on the tunnel sidewall are obtained; wherein, the laser scanning equipment and the pressure sensors are set on the same tunnel monitoring station;

[0128] Historical 3D scan data and historical stress data are input into a pre-trained surrounding rock analysis model to obtain surrounding rock parameter data. Among them, the surrounding rock parameter data includes the surrounding rock attribute parameters corresponding to each scan point in the historical 3D scan data, and the surrounding rock attribute parameters include rock strata, floor and sidewalls.

[0129] Based on the surrounding rock parameter data, determine the rock stratum thickness parameters of the tunnel floor, the bearing strength parameters of the floor, and the convergence deformation of the surrounding rock on both sides of the tunnel.

[0130] It should be noted that the No. 1 monitoring station in the mine roadway is equipped with both a laser scanning device for the arch and vibrating wire pressure sensors embedded in the middle of the side walls on both sides. The historical three-dimensional scanning data of the station over the past 20 days was retrieved, totaling 240 point clouds. Each point cloud covers the roof and the left and right side walls, with a scanning point density of 400 points per square meter, for a total of approximately 800,000 points. At the same time, the historical stress data collected by the pressure sensors within the same time window was retrieved. Three sensors were deployed on each of the left and right side walls, located at heights of 1 meter, 2 meters, and 3 meters respectively, with a sampling interval of 5 minutes. A total of 5,760 stress readings were obtained, including normal stress perpendicular to the side wall and shear stress parallel to the side wall.

[0131] These 240 sets of point cloud data were paired with corresponding stress data according to time series. Each set of input included the point cloud at a certain moment and the average stress within 2 minutes before and after that moment, forming a total of 240 training samples. The above samples were then input into a pre-trained surrounding rock analysis model. This model is based on a hybrid architecture of three-dimensional convolutional neural network and graph attention network. The input layer accepts point cloud coordinates and stress features, and after multiple nonlinear transformations, outputs the surrounding rock attribute parameters for each scan point. The attribute parameters are three types of probability values, representing the probability of the scan point belonging to the rock stratum, the floor, and the sidewall, respectively.

[0132] The model outputs three probability values ​​for each scan point. For example, a scan point on the roof has a probability of 0.92 belonging to rock strata, 0.05 belonging to the floor, and 0.03 belonging to the sidewall, thus identifying it as a rock strata. A scan point on the floor has a probability of 0.88 belonging to the floor, 0.10 belonging to rock strata, and 0.02 belonging to the sidewall, thus identifying it as the floor. A scan point on the left sidewall has a probability of 0.90 belonging to the sidewall, 0.07 belonging to rock strata, and 0.03 belonging to the floor, thus identifying it as the sidewall. All scan points in the entire point cloud obtain corresponding surrounding rock attribute labels.

[0133] Based on these attribute labels, the point cloud is semantically segmented to divide the tunnel space into rock strata, floor, and sidewall zones. In the floor zone, the system extracts all scan points marked as floor and uses the elevation coordinates and density distribution of these points, combined with the original reflection intensity data corresponding to the scan points, to calculate the location of each layer interface below the floor. By taking a cross-section every 0.2 meters along the tunnel direction and counting the abrupt changes in elevation of the floor scan points at each cross-section, three obvious interfaces are identified, located at 0.8 meters, 2.9 meters, and 4.3 meters below the floor surface, respectively. Thus, the rock strata thickness parameters are determined as follows: first layer 0.8 meters, second layer 2.1 meters, third layer 1.4 meters, with a total cumulative thickness of 4.3 meters.

[0134] For the bearing strength parameter of the base plate, the predicted uniaxial compressive strength value corresponding to each scanning point of the base plate is additionally retrieved from the surrounding rock analysis model. This predicted value is an implicit feature output by the model after stress data inversion and indoor test calibration.

[0135] The predicted uniaxial compressive strength is not directly output from the surrounding rock analysis model, but is obtained indirectly through stress data inversion combined with laboratory test calibration.

[0136] In the early stages of tunnel construction, rock core samples were collected at different depths of the floor slab to conduct indoor uniaxial compression tests and obtain the actual uniaxial compressive strength of each rock layer. Simultaneously, geostress monitoring data at the same location and time were recorded by a stress monitoring device below the floor slab. A large number of paired data (geostress parameters → uniaxial compressive strength) were used as training samples.

[0137] The model uses a BP neural network or GA-BP neural network, with geostress parameters such as maximum principal stress, minimum principal stress, and intermediate principal stress as inputs, and outputs uniaxial compressive strength. Training allows the model to learn the nonlinear mapping relationship between geostress and rock strength. Optimization methods such as particle swarm optimization (PSO) are employed to further improve inversion accuracy. After training, real-time geostress data is input into the inversion model to predict the uniaxial compressive strength corresponding to each scan point on the current base plate. The inversion model is periodically calibrated and corrected using new indoor test results to ensure that the predicted values ​​always remain consistent with actual field conditions.

[0138] The surrounding rock analysis model (3D convolutional neural network + graph attention network) outputs surrounding rock attribute labels for each scan point, such as rock layer / bottom plate / sidewall; for scan points marked as bottom plate, the system additionally retrieves the predicted uniaxial compressive strength value corresponding to that point from the above inversion model; finally, the average strength of each layer is statistically analyzed according to depth and output as the bottom plate bearing strength parameter.

[0139] The predicted strength values ​​of all base plate scanning points are statistically analyzed according to depth layers. The average strength of the first layer of silty mudstone is 28.2 MPa with a standard deviation of 1.3 MPa; the average strength of the second layer of fine sandstone is 45.1 MPa with a standard deviation of 2.0 MPa; and the average strength of the third layer of mudstone is 12.0 MPa with a standard deviation of 0.9 MPa. The system takes the average value of each layer as the base plate bearing capacity parameter, namely 28.2 MPa, 45.1 MPa and 12.0 MPa.

[0140] To analyze the convergence deformation of the surrounding rock on both sides of the tunnel, all scanning points marked as sidewalls were analyzed. Fixed reference lines were selected on the left and right sidewalls, located 1.5 meters above the floor and extending along the tunnel direction. For each set of historical scanning data, the horizontal distance between all scanning points on the left sidewall reference line and the tunnel centerline was extracted, and the average distance was calculated. The same method was applied to the right sidewall. The average distance at the initial moment was used as the initial value, and the average distance at each subsequent moment was subtracted from it to obtain the horizontal displacement of both sidewalls. Calculations showed that in the past 20 days, the average distance of the left sidewall changed from the initial value of 2.60 meters to 2.585 meters, a decrease of 15 millimeters, indicating a convergence of 15 millimeters into the tunnel; the average distance of the right sidewall changed from the initial value of 2.60 meters to 2.587 meters, a decrease of 13 millimeters, indicating a convergence of 13 millimeters.

[0141] Simultaneously, the convergence rate over the most recent week was extracted. The convergence amount of the left sidewall in the last week was 3.9 mm, with an average rate of 0.56 mm per day, which is about 0.3 times higher than the previous week's 0.43 mm per day. The convergence amount of the right sidewall in the last week was 3.2 mm, with an average rate of 0.46 mm per day, which is about 0.31 times higher than the previous week's 0.35 mm per day. These convergence deformation data, together with the difference in convergence values ​​of the left and right sidewalls at various heights, are used as parameters for the convergence deformation of the surrounding rock on both sides of the roadway.

[0142] The system outputs complete surrounding rock parameter data, including the thickness of the bottom slab layers (0.8 m, 2.1 m, 1.4 m), the bearing strength of each layer (28.2 MPa, 45.1 MPa, 12.0 MPa), the cumulative convergence of the left sidewall (15 mm) and the right sidewall (13 mm), the rate of increase over the past week (0.3 times), and the convergence differences at different heights of the left and right sidewalls. For example, at a height of 1 m, the left sidewall converges 16 mm and the right sidewall converges 14 mm; at a height of 2 m, the left sidewall converges 15 mm and the right sidewall converges 13 mm; at a height of 3 m, the left sidewall converges 13 mm and the right sidewall converges 11 mm. All parameters are stored and used for subsequent safety risk assessments.

[0143] In an optional embodiment, the surrounding rock analysis model includes a scanning feature extraction network, a stress feature extraction network, and a parameter fusion output network;

[0144] Correspondingly, historical 3D scanning data and historical stress data are input into the pre-trained surrounding rock analysis model to obtain surrounding rock parameter data;

[0145] Spatial structure feature vectors are obtained by extracting features from historical 3D scanning data using a scanning feature extraction network.

[0146] Mechanical feature vectors are obtained by extracting features from historical stress data using a stress feature extraction network.

[0147] The parameter fusion output network performs feature fusion on the spatial structure feature vector and the mechanical feature vector to obtain a comprehensive feature vector. Based on the comprehensive feature vector, the surrounding rock parameter data is output.

[0148] It should be noted that at the No. 3 monitoring station in the mine roadway, 120 sets of three-dimensional scanning point clouds and synchronous stress data collected in the past 24 hours were selected. The scanning feature extraction network adopts a sparse three-dimensional convolutional neural network. The input point cloud coordinates and reflection intensity are processed through 4 layers of convolution and pooling to output a 512-dimensional spatial structure feature vector, representing the geometric shape and texture features of the roadway roof, floor and sidewalls. The stress feature extraction network adopts a one-dimensional temporal convolutional network. The input normal stress and shear stress temporal data of 3 measuring points on each of the left and right sidewalls are 120. After 3 layers of convolution and global average pooling, a 256-dimensional mechanical feature vector is output, reflecting the stress state and changing trend of the surrounding rock.

[0149] The parameter fusion output network first concatenates the two feature vectors mentioned above into a 768-dimensional array, then passes them through two fully connected layers. The first layer compresses the array to 384 dimensions, and the second layer outputs three probability values ​​for each scan point, corresponding to the probability of belonging to the rock layer, the base, and the sidewall, respectively. For example, a point on the top plate outputs probabilities of 0.91, 0.06, and 0.03, and is identified as a rock layer; a point on the base plate outputs probabilities of 0.05, 0.87, and 0.08, and is identified as the base. During the fusion process, an attention mechanism is used to automatically assign weights to spatial and mechanical features, with weight coefficients of 0.62 and 0.38, respectively. Finally, the network outputs the surrounding rock attribute labels for all scan points and simultaneously inverts the strength parameters of each region. The average strength of the first layer of the base plate is 28.2 MPa, the second layer is 45.1 MPa, and the third layer is 12.0 MPa. The cumulative convergence amounts of the left and right sidewalls are 15 mm and 13 mm, respectively.

[0150] In an alternative embodiment, the system further includes:

[0151] Based on the hydrological monitoring equipment deployed around the tunnel floor to collect seepage pressure data of the tunnel floor, the hydrological monitoring equipment is installed at a predetermined depth on the surface of the tunnel floor or on the side wall of the tunnel floor.

[0152] If the seepage pressure data exceeds the set seepage pressure threshold, a seepage risk warning will be triggered and a comprehensive analysis will be conducted in conjunction with the safety risk assessment level.

[0153] If the seepage pressure data is less than or equal to the set seepage pressure threshold, the analysis will be based solely on the safety risk assessment level.

[0154] The monitoring and analysis results of the tunnel floor are presented on the remote monitoring terminal interface. The monitoring and analysis results include at least one of the following: deformation level category at each monitoring time, real-time horizontal displacement data, real-time vertical settlement, safety risk assessment level, or seepage risk early warning information.

[0155] It should be noted that four piezometers are installed around the floor of the No. 2 monitoring station in the mine roadway. Two are buried 0.3 meters below the floor surface, and two are buried 1.0 meter deep in the sidewall boreholes, with a sampling interval of 1 minute. The seepage pressure threshold is set at 0.20 MPa. At a certain moment, the measured seepage pressure is 0.27 MPa, exceeding the threshold, immediately triggering a seepage risk warning. This warning is then linked to the previously assessed safety risk level of 4 (high risk) for comprehensive analysis, and the overall judgment is that the risk under the influence of seepage has risen to level 4+, requiring priority treatment. At another moment, the measured seepage pressure is 0.12 MPa, not exceeding the threshold, so only the safety risk assessment level of 2 (low risk) is analyzed, and no warning is triggered. The remote monitoring terminal interface displays the deformation level category at each monitoring moment in real time. Currently, it is an abnormal deformation, with a maximum real-time horizontal displacement of 1.7 mm and a real-time vertical settlement of 3.7 mm, a safety risk assessment level of 4, and also displays whether the seepage risk warning information has been triggered or not.

[0156] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A tunnel floor deformation monitoring and analysis system, characterized in that, include: The data acquisition unit is used to simultaneously collect real-time horizontal displacement data and real-time ground stress data of the roadway floor at each monitoring moment through multiple sets of displacement sensors installed on the roadway floor and stress monitoring devices installed below the roadway floor. The state determination unit is used to determine the real-time vertical settlement and deformation state determination results of the roadway floor based on real-time horizontal displacement data and real-time ground stress data. The spatial alignment unit is used to acquire historical three-dimensional scanning data of the roadway roof and sidewalls collected by the laser scanning equipment deployed on the same roadway monitoring station when the deformation state determination result is abnormal deformation. Based on the historical three-dimensional scanning data and real-time horizontal displacement data, spatial alignment calibration is performed to obtain spatial alignment transformation parameters. The feature distribution unit is used to determine the three-dimensional deformation distribution characteristics of the tunnel floor based on the spatial alignment transformation parameters, real-time horizontal displacement data, and real-time vertical settlement. The safety assessment unit is used to obtain the rock stratum thickness parameters, floor bearing strength parameters, and convergence deformation of the surrounding rock on both sides of the roadway. Based on the three-dimensional deformation distribution characteristics, rock stratum thickness parameters, floor bearing strength parameters, and convergence deformation, the safety risk assessment level of the roadway floor is determined.

2. The tunnel floor deformation monitoring and analysis system according to claim 1, characterized in that, Based on real-time horizontal displacement data and real-time ground stress data, the real-time vertical settlement and deformation state of the tunnel floor are determined, including: Extract the maximum and minimum horizontal displacement values ​​from the real-time horizontal displacement data and calculate the horizontal displacement difference. The horizontal displacement difference and real-time ground stress data are input into the pre-trained settlement prediction model, which outputs the real-time vertical settlement. The settlement prediction model is trained using historical horizontal displacement difference data, historical ground stress data, and historical actual vertical settlement data. Real-time horizontal displacement data and real-time vertical settlement are input into a pre-trained deformation recognition model. The deformation recognition model includes a temporal feature extraction network and a deformation determination network. The temporal feature extraction network extracts displacement trend feature vectors and change rate feature vectors, and the deformation determination network outputs predicted deformation results. The deformation state determination result is determined based on the predicted deformation results.

3. The tunnel floor deformation monitoring and analysis system according to claim 2, characterized in that, Spatial alignment calibration is performed based on historical 3D scan data and real-time horizontal displacement data to obtain spatial alignment transformation parameters, including: The historical deformation and displacement data of the roadway floor collected by displacement sensors were acquired, and the historical three-dimensional scanning data of the roadway roof and sidewalls collected by laser scanning equipment were acquired. Initial spatial alignment is performed based on historical deformation displacement data and historical 3D scan data to obtain initial alignment parameters; The initial alignment parameters, the displacement calibration matrix of the displacement sensor, historical deformation displacement data, and historical 3D scan data are input into the untrained deformation association model to obtain the predicted deformation mapping result corresponding to the historical 3D scan data; wherein, the predicted deformation mapping result includes the deformation prediction value of each scan point in the historical 3D scan data; Based on the predicted deformation mapping results, the initial alignment parameters in the deformation association model are iteratively optimized until the iteration termination condition is met, thus obtaining the spatial alignment transformation parameters in the trained deformation association model.

4. The tunnel floor deformation monitoring and analysis system according to claim 3, characterized in that, Based on the predicted deformation mapping results, the initial alignment parameters in the deformation correlation model are iteratively optimized, including: Obtain historical deformation mapping data corresponding to historical 3D scanning data in the deformation association model, and determine the spatial deviation cost based on historical deformation mapping data and historical deformation displacement data; wherein, historical deformation mapping data represents the mapping result of historical 3D scanning data in the displacement coordinate system of displacement sensor; The deformation fitting loss value is determined based on the predicted deformation mapping results and the actual deformation measurement results corresponding to the historical 3D scanning data. Based on the spatial deviation cost and the deformation fitting loss value, the initial alignment parameters in the deformation correlation model are iteratively optimized; wherein, the iteration termination conditions include the convergence of the spatial deviation cost value and / or the convergence of the deformation fitting loss value.

5. The tunnel floor deformation monitoring and analysis system according to claim 4, characterized in that, Based on spatial alignment transformation parameters, real-time horizontal displacement data, and real-time vertical settlement, the three-dimensional deformation distribution characteristics of the tunnel floor are determined, including: Based on the spatial alignment transformation parameters, the real-time horizontal displacement data and real-time vertical settlement are mapped onto the three-dimensional spatial model of the tunnel constructed by the laser scanning equipment to obtain three-dimensional deformation mapping data. Based on the three-dimensional deformation mapping data, the three-dimensional deformation distribution characteristics of the tunnel floor are determined.

6. The tunnel floor deformation monitoring and analysis system according to claim 5, characterized in that, Obtain the rock stratum thickness parameters of the tunnel floor, the bearing strength parameters of the floor, and the convergence deformation of the surrounding rock on both sides of the tunnel, including: Historical 3D scanning data of the tunnel collected by laser scanning equipment and historical stress data of the tunnel sidewall collected by pressure sensors installed on the tunnel sidewall are obtained; wherein, the laser scanning equipment and the pressure sensors are set on the same tunnel monitoring station; Historical 3D scan data and historical stress data are input into a pre-trained surrounding rock analysis model to obtain surrounding rock parameter data. Among them, the surrounding rock parameter data includes the surrounding rock attribute parameters corresponding to each scan point in the historical 3D scan data, and the surrounding rock attribute parameters include rock strata, floor and sidewalls. Based on the surrounding rock parameter data, determine the rock stratum thickness parameters of the tunnel floor, the bearing strength parameters of the floor, and the convergence deformation of the surrounding rock on both sides of the tunnel.

7. The tunnel floor deformation monitoring and analysis system according to claim 6, characterized in that, The surrounding rock analysis model includes a scanning feature extraction network, a stress feature extraction network, and a parameter fusion output network; Correspondingly, historical 3D scanning data and historical stress data are input into the pre-trained surrounding rock analysis model to obtain surrounding rock parameter data; Spatial structure feature vectors are obtained by extracting features from historical 3D scanning data using a scanning feature extraction network. Mechanical feature vectors are obtained by extracting features from historical stress data using a stress feature extraction network. The parameter fusion output network performs feature fusion on the spatial structure feature vector and the mechanical feature vector to obtain a comprehensive feature vector. Based on the comprehensive feature vector, the surrounding rock parameter data is output.

8. The tunnel floor deformation monitoring and analysis system according to claim 7, characterized in that, The system also includes: Based on the hydrological monitoring equipment deployed around the tunnel floor to collect seepage pressure data of the tunnel floor, the hydrological monitoring equipment is installed at a predetermined depth on the surface of the tunnel floor or on the side wall of the tunnel floor. If the seepage pressure data exceeds the set seepage pressure threshold, a seepage risk warning will be triggered and a comprehensive analysis will be conducted in conjunction with the safety risk assessment level. If the seepage pressure data is less than or equal to the set seepage pressure threshold, the analysis will be based solely on the safety risk assessment level. The monitoring and analysis results of the tunnel floor are presented on the remote monitoring terminal interface. The monitoring and analysis results include at least one of the following: deformation level category at each monitoring time, real-time horizontal displacement data, real-time vertical settlement, safety risk assessment level, or seepage risk early warning information.

Images