Three-dimensional dynamic monitoring method of lidar in coal mine roadways

Abstract

This application relates to the field of coal mine safety monitoring technology and discloses a three-dimensional dynamic monitoring method for coal mine roadways using lidar. This method addresses the problems of lidar data interruption or unreliability under harsh operating conditions and poor monitoring continuity due to insufficient fusion of multi-source data in existing technologies. The key technical points are: simultaneously acquiring lidar surface data and optical strain gauge point data; constructing a digital twin model of the roadway; fusing and assimilating multi-source data using the Transformer mechanism to obtain initial roadway deformation data; determining the reliability of lidar data based on Bayesian estimation; when data is missing or unreliable, using a gating mechanism to combine optical strain gauge point data and the digital twin model to deduce the full-field deformation; and finally generating and outputting a three-dimensional dynamic deformation field. This invention is mainly used for real-time, continuous, and highly reliable deformation monitoring and safety early warning in coal mine roadways.

Description

Technical Field

[0001] This invention relates to the technical field of optical instruments for monitoring roadway deformation, and particularly to a three-dimensional dynamic monitoring method for coal mine roadways using lidar. Background Technology

[0002] In the field of coal mine safety production, roadway deformation monitoring is a key technical means to prevent accidents such as roof collapse and surrounding rock instability. With the development of optical monitoring technology, lidar (LiDAR) has become an important tool for roadway deformation monitoring due to its ability to acquire high-density three-dimensional point cloud data of the roadway surface non-contactly and efficiently. LiDAR can reconstruct the three-dimensional geometry of the roadway by emitting a laser beam and receiving the reflected signal, realizing the dynamic acquisition of deformation information across the entire field. In addition, optical strain gauges, based on fiber Bragg grating (FBG) technology, can accurately measure strain changes at local key points, and have advantages such as resistance to electromagnetic interference and good stability, making them suitable for the harsh environment of coal mines. However, existing monitoring systems have shortcomings in the collaborative utilization of multi-source data, making it difficult to ensure the continuity and reliability of data under complex working conditions.

[0003] Currently, common tunnel deformation monitoring solutions mainly rely on a single lidar system or a simple combination of multiple sensors. For example, some existing technologies use lidar to periodically scan the tunnel and extract deformation information through point cloud data processing algorithms. While these solutions can provide relatively accurate overall data under favorable environmental conditions, in extreme conditions in coal mines (such as high-concentration dust, water vapor interference, equipment failure, or severe surrounding rock deformation), lidar data is prone to interruptions, increased noise, or significantly reduced reliability, leading to monitoring blind spots or false alarms. Furthermore, single lidar systems lack the verification and supplementation of high-precision local data, making it difficult to provide reliable deformation assessments in critical areas (such as stress concentration points).

[0004] Other existing technologies attempt to combine lidar with other sensors such as optical strain gauges, but typically only perform parallel data acquisition and independent display, without establishing a deep multi-source data fusion and complementarity mechanism. For example, while optical strain gauges can provide high-precision strain data from sparsely distributed points, their spatial coverage is limited and cannot directly reflect the overall deformation distribution; while lidar data, although covering the entire field, may have insufficient local accuracy due to environmental interference. Existing technologies have failed to effectively solve the problems of time alignment, data correlation, weight allocation, and data assimilation under physical constraints between lidar and optical strain gauges. In particular, when lidar data is missing or unreliable, there is a lack of ability to extrapolate the overall deformation based on optical strain gauge points and tunnel mechanics models, resulting in insufficient data integrity of the monitoring system under extreme conditions and affecting the continuity of safety decisions.

[0005] The inventors of this application have discovered the following technical problem with the above-mentioned technology: how to construct a multi-source data fusion and complementarity mechanism that integrates lidar with other sparsely deployed optical strain gauges in the tunnel to ensure the integrity and high reliability of deformation monitoring data under any working condition. Summary of the Invention

[0006] This invention provides a three-dimensional dynamic monitoring method for coal mine roadways using lidar. The technical problem to be solved is: how to construct a multi-source data fusion and complementarity mechanism that integrates lidar and other sparsely deployed optical strain gauges in the roadway to ensure the integrity and high reliability of deformation monitoring data under any working condition.

[0007] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0008] A method for three-dimensional dynamic monitoring of coal mine roadways using lidar, the method comprising: step S10, simultaneously acquiring lidar surface data collected by lidar deployed in the coal mine roadway and optical strain gauge point data collected by multiple optical strain gauges deployed at key points of the coal mine roadway; step S20, constructing a digital twin model of the roadway based on the geometric and mechanical parameters of the coal mine roadway, the digital twin model including a roadway geometric model and a roadway mechanical model; step S30, based on the roadway digital twin model, using the Transformer mechanism to perform multi-source data fusion and data assimilation processing on the lidar surface data and the optical strain gauge point data to obtain initial roadway deformation data; step S40, based on... Based on Bayesian estimation theory, the credibility of the lidar surface data is determined; Step S50: When the lidar surface data is missing or the credibility value is lower than a preset threshold, the full-field deformation of the coal mine roadway is inferred using a gating mechanism based on the optical strain gauge point data and the roadway digital twin model to obtain inferred roadway deformation data; Step S60: When the lidar surface data is not missing and the credibility value is not lower than the preset threshold, the initial roadway deformation data is used as output data; Step S70: Based on the output data or the inferred roadway deformation data, a three-dimensional dynamic deformation field of the coal mine roadway is generated and output, wherein the three-dimensional dynamic deformation field represents the continuous deformation state of the coal mine roadway in time sequence.

[0009] The beneficial effects of this invention are as follows:

[0010] 1. This method constructs a digital twin model of the roadway by simultaneously acquiring surface data from lidar and point data from optical strain gauges. It then utilizes the Transformer mechanism for multi-source data fusion and assimilation, combined with Bayesian estimation to dynamically assess the reliability of lidar data. In cases of missing or unreliable data, a gating mechanism is used to extrapolate the full-field deformation, thereby generating a three-dimensional dynamic deformation field. This achieves continuous, complete, and highly reliable deformation monitoring data under any operating condition, significantly improving the monitoring accuracy and decision support capabilities for coal mine safety production.

[0011] 2. Improved integrity and continuity of monitoring data: By constructing a digital twin model of the tunnel that integrates geometric and mechanical properties, and utilizing the Transformer mechanism to deeply fuse and assimilate LiDAR surface data and optical strain gauge point data under physical constraints, this invention overcomes the limitations of a single data source. Especially when LiDAR data is missing or unreliable, it can intelligently deduce full-field deformation based on limited optical strain gauge point data and the tunnel mechanical model using a gating mechanism, thereby ensuring the integrity and continuity of monitoring data under any operating condition and effectively avoiding monitoring blind spots.

[0012] 3. Enhanced reliability and accuracy of monitoring results: This invention innovatively applies Bayesian estimation theory, using reliable optical strain gauge point data as prior information to dynamically evaluate the credibility of lidar surface data. This mechanism can automatically identify and quantify data uncertainties caused by environmental interference and other factors, providing a scientific basis for subsequent data processing strategies (direct adoption or initiation of extrapolation), thereby significantly improving the reliability and accuracy of the final output three-dimensional dynamic deformation field.

[0013] 4. Intelligent and Adaptive Monitoring Process: The technical solution of this invention constitutes a closed-loop intelligent monitoring system. It can not only adaptively switch operating modes when data is abnormal, but also periodically update the mechanical parameters of the digital twin model based on historical monitoring data using Transformer and gating mechanisms, enabling the model to continuously evolve and better reflect the actual long-term deformation behavior of the roadway. This self-learning and adaptive capability allows the monitoring system to maintain high accuracy over a long period, improving the system's intelligence level and long-term practicality. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0015] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. Identical components are indicated by the same reference numerals.

[0016] It should be noted that the terms “front,” “back,” “left,” “right,” “up,” and “down” used in the following description refer to the directions shown in the attached diagram, while the terms “inside” and “outside” refer to the directions toward or away from the geometric center of a specific component, respectively.

[0017] To make the content of this invention easier to understand, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings.

[0018] Invention Concept: This invention addresses the technical problems of insufficient multi-source data fusion and the susceptibility to interruption or unreliability of lidar data in harsh environments, leading to low data integrity and reliability in coal mine roadway deformation monitoring. It proposes a three-dimensional dynamic lidar monitoring method. This method constructs a digital twin model of the roadway by simultaneously acquiring lidar surface data and optical strain gauge point data. It then utilizes the Transformer mechanism for multi-source data fusion and assimilation, combined with Bayesian estimation to dynamically assess the reliability of lidar data. When data is missing or unreliable, a gating mechanism is used to extrapolate the full-field deformation, thereby generating a three-dimensional dynamic deformation field. This achieves continuous, complete, and highly reliable deformation monitoring data under any operating condition, significantly improving the monitoring accuracy and decision support capabilities for coal mine safety production.

[0019] A method for three-dimensional dynamic monitoring of coal mine roadways using lidar, the method comprising:

[0020] Step S10: Simultaneously acquire lidar surface data collected by lidar deployed in the coal mine roadway and optical strain gauge point data collected by multiple optical strain gauges deployed at key points in the coal mine roadway. The lidar surface data represents the overall deformation information of the coal mine roadway, while the optical strain gauge point data represents the local deformation information of the key points in the coal mine roadway. This step is the foundation for data acquisition in the entire monitoring system. The lidar, by emitting a laser beam and receiving reflected signals, acquires the spatial coordinates of a large number of points on the roadway surface, forming dense point cloud data, i.e., "lidar surface data," which comprehensively reflects the deformation of the entire roadway surface. The optical strain gauge, on the other hand, is a high-precision device for measuring local deformation. It monitors minute strain changes at specific locations using fiber Bragg grating technology, forming "optical strain gauge point data." The complementarity of these two data sources lies in the fact that lidar provides overall data, but may be affected by environmental interference, while optical strain gauges provide localized but extremely stable and reliable data. In the complex environment of a coal mine, this multi-source data acquisition effectively overcomes the limitations of a single data source, especially ensuring the continuity of monitoring under harsh conditions such as dust and water vapor. Step S20: Based on the geometric and mechanical parameters of the coal mine roadway, a digital twin model of the roadway is constructed. This digital twin model includes a geometric model and a mechanical model. The geometric model describes the three-dimensional spatial structure of the coal mine roadway, while the mechanical model describes its deformation behavior under stress conditions. The digital twin model is one of the core innovations of this invention; it is not a simple three-dimensional visualization model but a dynamic simulation system that integrates physical laws. The geometric model accurately reproduces the spatial morphology of the roadway using CAD technology, while the mechanical model, based on rock mechanics principles, describes the deformation behavior of the roadway under the influence of surrounding rock pressure, groundwater, and other factors. Key parameters to consider in constructing this model include the roadway's geometry (such as cross-sectional dimensions and curvature) and the physical properties of the rock mass (such as elastic modulus and Poisson's ratio). The digital twin model acts like a "virtual laboratory," predicting the roadway's deformation behavior under different working conditions and comparing it with actual measurement data for verification. Step S30: Based on the digital twin model of the tunnel, the Transformer mechanism is used to perform multi-source data fusion and assimilation processing on the lidar surface data and the optical strain gauge point data to obtain initial tunnel deformation data. The Transformer mechanism includes converting the lidar surface data and optical strain gauge point data into time-series data, calculating the correlation weights between the multi-source data through a self-attention mechanism, and performing weighted fusion based on the correlation weights. The data assimilation processing includes using the tunnel mechanical model to perform physical constraint optimization on the fused data. This step is the core innovation of this invention, realizing intelligent fusion of multi-source data. The Transformer mechanism originates from the field of natural language processing and is innovatively applied here to engineering monitoring data fusion.The principle is to treat data from different sources as "language sequences" and automatically learn the correlations between data points through a self-attention mechanism. For example, when a lidar indicator shows significant deformation in a certain area, the system will automatically focus on optical strain gauge data near that area and assign it higher weight. Data assimilation involves comparing the fusion result with the predictions of the digital twin model and using physical laws to correct possible measurement errors. This process can be represented by the following formula:

[0021]

[0022]

[0023] In the formula, The deformation value after fusion. For LiDAR data, For optical strain gauge data, For adaptive weighting coefficients, This is the weight matrix. and These are the feature vectors of the lidar and strain gauge data, respectively.

[0024] Step S40: Based on Bayesian estimation theory, determine the reliability of the lidar surface data. The Bayesian estimation theory uses the optical strain gauge point data as prior information to calculate the posterior probability distribution of the lidar surface data, and generates a reliability value for the lidar surface data based on the posterior probability distribution. This step innovatively applies Bayesian statistical methods to sensor data reliability assessment. The principle is that optical strain gauge data, due to its fixed installation location and minimal environmental interference, can serve as a "reliable reference" (prior information); while lidar data may be affected by dust, moisture, etc., resulting in errors (likelihood function). Using the Bayesian formula, the reliability (posterior probability) of the lidar data after considering the optical strain gauge reference can be calculated. The specific formula is:

[0025]

[0026] In the formula, This represents the posterior probability distribution of the lidar data. Let be the likelihood function. For the prior probability distribution, This is an evidence factor.

[0027] The confidence score is typically taken as the reciprocal of the variance of the posterior distribution; the smaller the variance, the more reliable the data. The advantage of this method is that it can dynamically assess the quality of LiDAR data, providing a basis for subsequent data missing handling.

[0028] Step S50: When the lidar surface data is missing or the confidence value is lower than a preset threshold, based on the optical strain gauge point data and the roadway digital twin model, a gating mechanism is used to deduce the full-field deformation of the coal mine roadway, obtaining deduced roadway deformation data. The gating mechanism includes an input gate, a forget gate, and an output gate, used to control the information flow between the optical strain gauge point data and the roadway mechanical model during the deduction process, and to simulate deformation propagation by iteratively updating the hidden state. This step is another core innovation of the invention, solving the industry problem of data interruption under extreme conditions. When lidar data is unreliable or missing, the system does not simply stop working, but instead uses the optical strain gauge point data and the digital twin model to deduce the full-field deformation. The gating mechanism originates from a gated recurrent unit (GRU) neural network and includes three key components: the input gate determines the degree of new information inclusion, the forget gate controls the retention ratio of historical information, and the output gate adjusts the final output. These gating units work together to simulate the spatial propagation law of roadway deformation.

[0029] In this way, the system can reasonably deduce the deformation state of the entire tunnel based on limited point data and the laws of rock mechanics.

[0030] Step S60: When the lidar surface data is not missing and the confidence value is not lower than the preset threshold, the initial roadway deformation data is used as the output data; this step defines the normal working mode of the system. When the lidar data quality and integrity meet the requirements (i.e., the confidence value is higher than the preset threshold, usually set to 0.6-0.8), the system directly uses the initial roadway deformation data obtained by fusion in step S30 as the final output. This design ensures that the most accurate original data is used under good conditions, avoiding information loss caused by unnecessary data processing. Threshold setting needs to balance sensitivity and stability: too high a threshold will lead to frequent switching to the inference mode, increasing the computational burden; too low a threshold may miss data quality issues. This invention achieves the scientific nature and adaptability of threshold judgment through Bayesian confidence evaluation. Step S70: Based on the output data or the inferred roadway deformation data, a three-dimensional dynamic deformation field of the coal mine roadway is generated and output, wherein the three-dimensional dynamic deformation field represents the continuous deformation state of the coal mine roadway in time. This step transforms the processed data into intuitive visualization results. The three-dimensional dynamic deformation field not only displays the deformation state at the current moment but also includes historical deformation trends, forming a continuous spatiotemporal monitoring view. In specific implementation, the system converts discrete point data or inferred data into a continuous gridded deformation field using an interpolation algorithm. Then, visualization techniques such as color mapping and vector arrows are applied to intuitively display the deformation magnitude and direction. The deformation field is typically updated at a rate of 1-5 frames per second, ensuring real-time performance while avoiding data overload. The key innovation lies in the close integration of the physical model and visualization, making the deformation field not only aesthetically pleasing but also consistent with the laws of rock mechanics, avoiding unreasonable deformation patterns that may appear in traditional visualizations. Furthermore, in step S10, the synchronous acquisition includes:

[0031] Step S101: The coal mine roadway is continuously scanned by the lidar to generate lidar surface data, which includes point cloud data and its timestamp. This sub-step details the specific process of lidar data acquisition. The lidar emits a laser beam by rotating its scanning head, calculates the distance by measuring the laser's round-trip time, and determines the spatial coordinates by combining angle information to form point cloud data. Each point contains three-dimensional coordinates (x, y, z) and reflection intensity, and the timestamp ensures the accuracy of the data's temporal sequence. In the coal mine environment, the lidar is usually installed above the roadway's central axis, with a scanning frequency set to 0.1-5Hz to balance data quality and system load. The density of point cloud data directly affects the accuracy of deformation monitoring. This invention optimizes scanning parameters to obtain a sufficiently dense point cloud (typically 100-1000 points per square meter) while ensuring real-time performance. Step S102: The strain signals of the key points are synchronously acquired by multiple optical strain gauges to generate optical strain gauge point data, which includes strain values, their position coordinates, and timestamps. This sub-step details the optical strain gauge data acquisition process. Optical strain gauges are based on fiber Bragg grating (FBG) technology. When the fiber is subjected to strain, the grating period changes, causing a shift in the reflected wavelength. By measuring this wavelength change, the strain value can be accurately calculated. Compared with traditional electrical strain gauges, optical strain gauges have advantages such as resistance to electromagnetic interference, corrosion resistance, and long-distance transmission, making them particularly suitable for the harsh environment of coal mines. The selection of key points needs to consider the weak points and stress concentration areas of the tunnel structure, typically including the top center, the middle of the side walls, the bottom center, and corners. Each strain gauge data point includes the strain value (unit: microstrain με), precise installation location coordinates, and a timestamp. Step S103: Time alignment processing is performed on the lidar surface data and the optical strain gauge point data to ensure that they are on the same time base. This sub-step solves the time synchronization problem of multi-source data. Since lidar and optical strain gauges may use different clock systems and sampling frequencies, direct use can lead to time misalignment. Time alignment processing typically employs two methods: hardware synchronization (triggered by the same clock source) and software synchronization (aligned through interpolation algorithms). This invention prioritizes hardware synchronization to ensure high accuracy. When hardware synchronization is not feasible, algorithms such as cubic spline interpolation are used for software synchronization to align data of different frequencies onto a unified time grid. The accuracy of time alignment directly affects the subsequent data fusion effect. This invention controls the time error within ±10ms to meet the monitoring requirements of coal mines. Further, in step S20, the construction of the roadway digital twin model includes:

[0032] Step S201: Obtain the design drawings and geological exploration data of the coal mine roadway, and extract the geometric parameters, including the roadway cross-sectional shape, dimensions, and spatial coordinates; this sub-step is the foundation for constructing the geometric model. The design drawings provide the theoretical shape and dimensions of the roadway, while the geological exploration data describes the distribution of the surrounding rock mass. The geometric parameters include not only basic cross-sectional shapes (such as rectangles, arches, horseshoes, etc.) and dimensions (width, height, radius of curvature, etc.), but also the spatial orientation of the roadway (such as inclination angle, curvature) and precise coordinate positions. These parameters are converted into a precise three-dimensional geometric model using CAD software, serving as the basic framework for the digital twin. Coal mine roadways are usually not perfectly regular geometric bodies and may have construction errors and local deformations; therefore, this invention also considers modeling the initial deformation to make the geometric model closer to the actual situation. Step S202: Based on the rock mass mechanical properties of the coal mine roadway, determine the mechanical parameters, including the elastic modulus, Poisson's ratio, and yield strength; this sub-step determines the key parameters of the mechanical model. Rock mass mechanical properties are typically obtained through field sampling and laboratory testing, including elastic modulus (representing the rock's resistance to deformation), Poisson's ratio (representing the ratio of lateral to longitudinal deformation), and yield strength (representing the stress value at which the rock begins to undergo plastic deformation). These parameters are crucial for accurately simulating tunnel deformation; for example, a larger elastic modulus results in smaller deformation under the same stress, and a larger Poisson's ratio indicates more pronounced lateral deformation. This invention also considers the heterogeneity and anisotropy of the rock mass, dividing the tunnel region into multiple sub-regions and assigning different mechanical parameters to each, thereby improving model accuracy. Step S203: Using the aforementioned geometric and mechanical parameters, the tunnel geometric model and the tunnel mechanical model are constructed using the finite element method, and integrated to form the tunnel digital twin model. This sub-step completes the final construction of the digital twin model. The finite element method discretizes the continuous tunnel structure into a large number of small elements (such as tetrahedrons or hexahedrons), applies mechanical equations to each element, and then combines them to solve the deformation behavior of the entire system. The geometric model defines the spatial location and connection relationships of the elements, while the mechanical model provides the material properties and mechanical equations of the elements. When integrated, the digital twin model can simulate the deformation response of a tunnel under various loads (such as surrounding rock pressure and groundwater pressure). The innovation of this invention lies in the close integration of the digital twin model with real-time monitoring data, making it not only a static simulation tool, but also continuously updated and optimized based on actual monitoring data.

[0033] Furthermore, in step S30, the multi-source data fusion and data assimilation processing using the Transformer mechanism includes:

[0034] Step S301: Convert the lidar surface data and the optical strain gauge point data into time-series data, wherein the time-series data includes deformation values ​​at multiple time steps; this sub-step completes the data preparation before Transformer processing. The time-series data organizes the spatially distributed monitoring data in chronological order, forming a "data matrix," where rows represent time steps and columns represent spatial locations. For lidar surface data, due to the large and irregular size of the point cloud data, it needs to be meshed first, discretizing the continuous space into fixed grid points; for optical strain gauge point data, because its location is fixed, it can be directly arranged in chronological order. This conversion gives data from different sources and with different structures a unified time-series representation, facilitating subsequent Transformer processing. The time-series length is typically set to 10-100 time steps to balance historical information and computational complexity. Step S302: The time-series data is processed using a Transformer encoder. A self-attention mechanism is used to calculate the correlation weight between the lidar surface data and the optical strain gauge point data. Based on these correlation weights, a weighted sum is performed to generate fused deformation data. This sub-step details the core processing procedure of the Transformer. The self-attention mechanism is a key component of the Transformer; it calculates the correlation weight between data points, expressed by the formula:

[0035]

[0036] In the formula, For querying the matrix, The key matrix, For value matrices, Let T be the dimension of the key vector, and T denote the transpose of the matrix.

[0037] In this invention, Q, K, and V are all obtained from the input data through linear transformation. The self-attention mechanism can automatically discover the correlation between data points at different locations and times. For example, when an optical strain gauge in a certain area shows abnormal deformation, the system will automatically increase the weight of the lidar data in the vicinity of that area. This correlation is not preset but is automatically learned through training data, and can adapt to different tunnel conditions and deformation patterns.

[0038] Step S303: The fused deformation data is assimilated using the tunnel mechanics model. By minimizing the error between the model prediction and the fused deformation data, the initial tunnel deformation data is optimized. This sub-step completes the data assimilation process, combining the fusion result with the physical model. Data assimilation is the process of combining observed data with model predictions. This invention employs a variational assimilation method, optimizing the following objective function:

[0039]

[0040] In the formula, Let be the state vector to be optimized. Background field (model prediction). For background error covariance, For the observation operator, For observation data, The observation error covariance is represented by T, which denotes the matrix transpose operation.

[0041] By minimizing the objective function, the system finds the optimal solution that conforms to both physical laws and closely reflects actual observations. This process ensures that the fusion result will not exhibit "pseudo-deformations" that violate the laws of rock mechanics, such as large deformations that spontaneously occur without external force. Furthermore, in step S40, the determination of the reliability of the lidar surface data includes:

[0042] Step S401: Using the optical strain gauge point data as the prior distribution, construct the likelihood function of the lidar surface data; this sub-step establishes the mathematical foundation for Bayesian reliability assessment. The prior distribution represents our initial understanding of the system state. Here, optical strain gauge data is used as the prior because of its high reliability but low spatial resolution. The likelihood function describes the probability of observing the current lidar data under a given real state, and is usually assumed to be a Gaussian distribution.

[0043]

[0044] In the formula, These are lidar observations. This represents the actual state. For the observation operator, This represents the standard deviation of the observation error.

[0045] variance of the likelihood function This reflects the uncertainty of lidar data, which is affected by environmental conditions (such as dust concentration). This invention, through historical data analysis, establishes a correlation between environmental conditions and... The relational model enables the likelihood function to dynamically adapt to different operating conditions. Step S402: Based on Bayes' theorem, calculate the posterior probability distribution of the lidar surface data; this sub-step applies Bayes' theorem to complete the core calculation of reliability. Bayes' theorem combines prior information and observation data to obtain the posterior distribution:

[0046]

[0047] In practical applications, the mean and variance of the posterior distribution are usually the focus. The mean represents the most likely true state, while the variance reflects uncertainty. This invention pays particular attention to the posterior variance, using it as a key indicator for reliability assessment. The smaller the posterior variance, the higher the consistency between the lidar data and prior information, and the more reliable the data; conversely, the higher the variance, the lower the reliability. Step S403: Extract the variance index from the posterior probability distribution as the reliability value, where the variance index represents the uncertainty of the lidar surface data. This sub-step completes the specific calculation of the reliability value. The posterior variance directly reflects the uncertainty of the data, but for ease of use, this invention converts it into a reliability value in the [0,1] interval:

[0048]

[0049] In the formula, For posterior variance, This is a scaling factor used to adjust the sensitivity of the confidence value.

[0050] This transformation ensures that the larger the variance, the lower the reliability, and exhibits a monotonic relationship within a reasonable range. A reliability value close to 1 indicates that the data is very reliable, while a value close to 0 indicates that the data is almost unreliable. The system automatically decides whether to switch to the simulation mode based on the reliability value, realizing intelligent management of the monitoring process. Furthermore, in step S50, the simulation of the full-field deformation of the coal mine roadway using the gating mechanism includes:

[0051] Step S501: When the lidar surface data is missing, the deformation of the key points is extracted from the optical strain gauge point data as input. This sub-step defines the triggering conditions and initial input of the simulation mode. When the lidar fails completely due to malfunction or extreme environment, the system starts the simulation based solely on the optical strain gauge point data. The deformation of the key points serves as the "seed data" for the simulation, and the system assumes that the measurements of these points are accurate and reliable. Since the number of optical strain gauges is limited (usually 5-10), while the tunnel surface contains tens of thousands of points, this simulation from sparse points to the entire field is extremely challenging. The innovation of this invention lies in combining a physical model and a data-driven method. It utilizes the laws of rock mechanics to constrain the simulation process and learns the actual deformation mode through a gating mechanism, avoiding the parameter inaccuracies that may exist in a purely physical model. Step S502: Using a gated cyclic unit network, with the deformation of the key points as the initial input, the input gate controls the addition of new information, the forget gate controls the retention of historical information, and the output gate controls the output of the hidden state to perform a full-field deformation simulation. This sub-step describes in detail the working principle of the gating mechanism. A gated recurrent unit (GRU) is a special type of recurrent neural network whose gating mechanism includes:

[0052] Input gate: determines the degree of influence of new observation data on the current state;

[0053] Forget Gate: Controls the proportion of historical information retained;

[0054] Output gate: Adjusts the deformation field of the final output;

[0055] These gating units are implemented using the sigmoid function, with output values ​​between [0,1] representing the "degree of openness". For example, when the system detects a sudden change in the deformation pattern, the input gates open more, allowing new information to quickly influence the deduction results; when the deformation pattern stabilizes, the forget gates retain more historical information, making the deduction results smoother. This mechanism enables the system to adapt to different deformation dynamics, without being overly sensitive to noise or missing the true deformation trend.

[0056] Step S503: Based on the tunnel mechanics model, perform physical constraint correction on the simulation results, and output the simulated tunnel deformation data by iteratively updating the hidden state of the gated cyclic unit network. This sub-step completes the physical rationality correction of the simulation results. Purely data-driven simulations may produce results that violate physical laws, such as large deformations that occur spontaneously without external force. This invention calculates the residuals by comparing the simulation results with the predictions of the tunnel mechanics model:

[0057]

[0058] In the formula, For the results of the deduction, For mechanical model prediction, For observation operators.

[0059] Then, the gradient descent method is used to adjust the GRU network parameters to minimize the residuals. This process ensures that the inference results conform to both actual observation trends and the basic laws of rock mechanics, avoiding unreasonable "pseudo-deformations." Iteratively updating the hidden states enables the system to continuously learn deformation patterns, improving the accuracy of long-term inferences. Furthermore, in step S503, the physical constraint correction includes:

[0060] Step S5031: Calculate the residuals between the deduction results and the predicted values ​​of the tunnel mechanics model at the key points; this sub-step defines the specific calculation method for physical constraint correction. The residual calculation focuses on the key points where the optical strain gauges are installed, because these points have both deduction results and actual observations, allowing for direct comparison. The residual formula is:

[0061]

[0062] In the formula, For the first The residuals at each key point For the results of the deduction, These are the predicted values ​​from the mechanical model.

[0063] The residuals reflect the inconsistencies between the extrapolated results and the physical model, serving as the basis for correction. The selection of key points is crucial, covering the main deformation patterns of the tunnel, such as roof subsidence and sidewall bulging. This invention also considers the spatial correlation of residuals, meaning that the residuals of adjacent key points often have similarities, which helps improve the accuracy of correction. Step S5032: Adjust the parameters of the gated recurrent unit network using gradient descent to minimize the residuals; this sub-step describes the optimization method for parameter adjustment. Gradient descent calculates the gradient of the loss function with respect to the network parameters, adjusts the parameters in the opposite direction of the gradient, and gradually reduces the residuals. The loss function is defined as:

[0064]

[0065] In the formula, For loss function, For the number of key points, For residuals, For network parameters, is the regularization coefficient.

[0066] Regularization term To prevent overfitting due to excessive parameter adjustment, the choice of learning rate is crucial: too high a rate may cause oscillations in the optimization process, while too low a rate results in slow convergence. This invention employs an adaptive learning rate strategy, dynamically adjusting the learning rate based on the residual size to ensure both rapid and stable optimization. Step S5033: Repeat steps S5031 and S5032 until the residual is less than the tolerance value. This sub-step completes the iterative process of physical constraint correction. Iterative correction ensures a high degree of consistency between the deduction results and the physical model; the tolerance value is typically set to 5%-10% of the average deformation of key points. The iteration process may require 3-10 iterations, depending on the degree of difference between the initial deduction and the physical model. To prevent infinite loops, the system also sets a maximum number of iterations (typically 20). This iterative correction mechanism allows the system to find the optimal balance between physical laws and actual observations, neither over-relying on potentially inaccurate physical models nor completely ignoring the constraints of physical laws. Further, in step S70, generating and outputting the three-dimensional dynamic deformation field includes:

[0067] Step S701: Convert the output data or the inferred roadway deformation data into gridded deformation data, wherein the gridded deformation data includes the deformation vector and time series of each grid point in the coal mine roadway; this sub-step completes the conversion from discrete data to a continuous field. Gridding is the process of interpolating point data (whether it is a lidar point cloud or the inferred result) onto a regular grid. Common methods include inverse distance weighting and Kriging interpolation. This invention adopts an adaptive grid based on roadway geometry, using a denser grid in key areas (such as the top and corners) and a sparser grid in flat areas, which ensures the accuracy of key areas while controlling the computational load. The deformation vector of each grid point contains displacement components (u, v, w) in three directions and time series information, forming a four-dimensional dataset (x, y, z, t). Step S702: Based on the gridded deformation data, generate a graphical representation of the three-dimensional dynamic deformation field through a visualization algorithm; this sub-step realizes an intuitive display of the deformation data. The visualization algorithm includes:

[0068] Deformation cloud map: The size of the deformation is represented by a color map, with red typically indicating a large deformation and blue indicating a small deformation.

[0069] Deformation vector: Arrows indicate the direction and magnitude of deformation.

[0070] Transformation animation: Showing the evolution of transformation over time.

[0071] Cross-sectional view: showing the deformation distribution on a specific cross-section.

[0072] The innovation of this invention lies in the close integration of physical deformation with the tunnel structure. For example, the original tunnel outline is preserved in the cloud map as a reference, making the degree of deformation immediately apparent. The system also provides multiple perspective switching and local magnification functions, facilitating technicians to focus on key areas. Step S703: Output the three-dimensional dynamic deformation field to the monitoring system for real-time display and early warning. This sub-step completes the final application of the monitoring results. The output interface typically adopts standard communication protocols (such as MQTT, OPCUA) to ensure compatibility with existing monitoring systems. The system not only provides visual display but also integrates intelligent early warning functions: when the deformation rate exceeds the threshold or the deformation mode becomes abnormal, different levels of early warning (such as yellow warning, red warning) are automatically triggered. The early warning threshold is set based on historical data and rock mechanics analysis, taking into account the specific conditions and safety margin of the tunnel. In addition, the system also provides a data export function for subsequent analysis and archiving. Further, it also includes step S80;

[0073] Step S80: Periodically update the digital twin model of the tunnel. The update is based on the historical three-dimensional dynamic deformation field, using the Transformer mechanism for feature extraction, and a gating mechanism to control the update rhythm of model parameters. This step is an extended innovation of the present invention, realizing the adaptive evolution of the digital twin model. In traditional monitoring systems, digital twin models are usually static and cannot adapt to long-term changes in tunnel conditions (such as rock creep and aging of support structures). The present invention, by periodically updating model parameters, enables the digital twin model to "learn" the actual behavior of the tunnel, improving the accuracy of long-term monitoring. The update process includes two key steps: first, using the Transformer mechanism to extract temporal features from historical deformation data to identify long-term trends in deformation patterns; then, using a gating mechanism to control the update rhythm of model parameters to avoid overfitting short-term fluctuations. This "learning-update" mechanism gives the system long-term adaptability, enabling it to cope with slow changes in tunnel conditions.

[0074] Step S80, in which the periodic update includes:

[0075] Step S801: Collect the three-dimensional dynamic deformation fields from historical periods as a training dataset;

[0076] This sub-step defines the data foundation for model updates. The training dataset contains a sufficiently long period of historical deformation data (typically 3-6 months), covering various operating conditions and deformation patterns. Data preprocessing includes outlier removal and normalization to ensure the quality of the training data. The time span of the historical data is crucial: too short a timeframe fails to capture long-term trends, while too long a timeframe may contain outdated deformation patterns. This invention employs a sliding time window strategy, prioritizing recent data while retaining some historical data to maintain the continuity of long-term trends.

[0077] Step S802: Use the Transformer encoder to extract temporal features from the training dataset to obtain feature vectors;

[0078] This sub-step details the feature extraction process. The Transformer encoder extracts key temporal features from historical deformation data, such as deformation trends, periodic changes, and abrupt changes, through a multi-layered self-attention mechanism. Compared to traditional feature extraction methods, Transformer can capture long-distance dependencies, such as recognizing complex patterns like "accelerated top deformation typically occurs 24 hours after significant sidewall deformation." The feature vector is a compressed representation of the original data, retaining the most important information while removing noise and redundancy.

[0079] Step S803: Input the feature vector into the gated recurrent unit network, output the corrected mechanical parameters through the gating mechanism, and update the roadway digital twin model.

[0080] This sub-step completes the adaptive update of the model parameters. The gating mechanism acts as a "regulator," controlling the magnitude and pace of parameter updates. For example, when the feature vectors show a stable deformation trend, the gating mechanism limits the magnitude of parameter updates to avoid overfitting short-term fluctuations; when the feature vectors show a significant trend change, the gating mechanism allows for larger updates, enabling the model to quickly adapt to the new situation. The specific update formula is:

[0081]

[0082] In the formula, For the updated parameters, For the original parameters, This is a gating signal (0≤g≤1). For feature vector-based The update volume.

[0083] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for three-dimensional dynamic monitoring of a coal mine roadway by laser radar, characterized in that, The method comprises the following steps: in step S10, laser radar surface data collected by a laser radar arranged in a coal mine tunnel and optical strain gauge point data collected by a plurality of optical strain gauges arranged at key points of the coal mine tunnel are synchronously acquired; in step S20, a tunnel digital twin model is constructed based on geometric parameters and mechanical parameters of the coal mine tunnel, the tunnel digital twin model comprising a tunnel geometric model and a tunnel mechanical model; in step S30, based on the tunnel digital twin model, multi-source data fusion and data assimilation processing are performed on the laser radar surface data and the optical strain gauge point data using a Transformer mechanism, to obtain initial tunnel deformation data; in step S40, the reliability of the laser radar surface data is judged based on a Bayesian estimation theory; in step S50, when the laser radar surface data is missing or the reliability value is lower than a preset threshold, full-field deformation of the coal mine tunnel is deduced based on the optical strain gauge point data and the tunnel digital twin model using a gating mechanism, to obtain deduced tunnel deformation data; in step S60, when the laser radar surface data is not missing and the reliability value is not lower than the preset threshold, the initial tunnel deformation data is taken as output data; in step S70, based on the output data or the deduced tunnel deformation data, a three-dimensional dynamic deformation field of the coal mine tunnel is generated and output, wherein the three-dimensional dynamic deformation field represents continuous deformation states of the coal mine tunnel over time; In step S30, the multi-source data fusion and data assimilation processing using the Transformer mechanism comprises the following steps: in step S301, the laser radar surface data and the optical strain gauge point data are respectively converted into time sequence data, wherein the time sequence data comprises deformation values at a plurality of time steps; in step S302, the time sequence data is processed using a Transformer encoder, the correlation weight between the laser radar surface data and the optical strain gauge point data is calculated through a self-attention mechanism, and weighted summation is performed based on the correlation weight, to generate fusion deformation data; in step S303, the fusion deformation data is subjected to data assimilation processing using the tunnel mechanical model, to minimize the error between model prediction and the fusion deformation data, and to optimize the initial tunnel deformation data; In step S40, the judgment of the reliability of the laser radar surface data comprises the following steps: in step S401, a likelihood function of the laser radar surface data is constructed taking the optical strain gauge point data as a prior distribution; in step S402, a posterior probability distribution of the laser radar surface data is calculated based on a Bayesian formula; in step S403, a variance index is extracted from the posterior probability distribution as the reliability value, wherein the variance index represents the uncertainty of the laser radar surface data. In step S50, the use of the gating mechanism to deduce the full-field deformation of the coal mine tunnel includes: step S501, when the laser radar surface data is missing, the deformation of the key point is extracted from the optical strain gauge point data as input; step S502, using a gated recurrent unit network, using the deformation of the key point as the initial input, and through the addition of new information by input gate control, the retention of historical information by forgetting gate control, and the output of hidden state by output gate control, the full-field deformation is deduced; step S503, based on the tunnel mechanics model, the results of the deduction are corrected, and the hidden state of the gated recurrent unit network is updated by iteration, and the deduced tunnel deformation data is output; In step S503, the physical constraint correction includes: step S5031, calculating the residual of the deduced results and the predicted value of the tunnel mechanics model at the key points; step S5032, using gradient descent method to adjust the parameters of the gated recurrent unit network to minimize the residual; step S5033, repeating steps S5031 and S5032 until the residual is less than the tolerance value.

2. The coal mine roadway laser radar three-dimensional dynamic monitoring method according to claim 1, characterized in that, In step S10, the synchronization acquisition includes: step S101, continuously scanning the coal mine tunnel by the laser radar to generate the laser radar surface data, wherein the laser radar surface data includes point cloud data and its timestamp; step S102, synchronously collecting the strain signals of the key points by the multiple optical strain gauges to generate the optical strain gauge point data, wherein the optical strain gauge point data includes strain value and its position coordinates and timestamp; step S103, time alignment processing is performed on the laser radar surface data and the optical strain gauge point data to ensure that the laser radar surface data and the optical strain gauge point data are under the same time reference.

3. The coal mine roadway laser radar three-dimensional dynamic monitoring method according to claim 1, characterized in that, In step S20, the construction of the tunnel digital twin model includes: step S201, obtaining the design drawings and geological exploration data of the coal mine tunnel, and extracting the geometric parameters, wherein the geometric parameters include the cross-sectional shape, size and spatial coordinates of the tunnel; step S202, based on the rock mass mechanical properties of the coal mine tunnel, determining the mechanical parameters, wherein the mechanical parameters include the elastic modulus, Poisson's ratio and yield strength; step S203, using the geometric parameters and the mechanical parameters, constructing the tunnel geometric model and the tunnel mechanics model by finite element method, and integrating to form the tunnel digital twin model.

4. The coal mine roadway laser radar three-dimensional dynamic monitoring method according to claim 1, characterized in that, In step S70, generating and outputting the three-dimensional dynamic deformation field includes: step S701, converting the output data or the deduced tunnel deformation data into grid deformation data, wherein the grid deformation data includes the deformation vector and time series of each grid point of the coal mine tunnel; step S702, based on the grid deformation data, generating a graphical representation of the three-dimensional dynamic deformation field by a visualization algorithm; step S703, outputting the three-dimensional dynamic deformation field to a monitoring system for real-time display and early warning.

5. The coal mine roadway laser radar three-dimensional dynamic monitoring method according to claim 1, characterized in that, The method further comprises: periodically updating the roadway digital twin model in step S80, wherein the updating is based on the historical three-dimensional dynamic deformation field, uses a Transformer mechanism for feature extraction, and controls the updating rhythm of model parameters through a gating mechanism.

6. The coal mine roadway laser radar three-dimensional dynamic monitoring method according to claim 5, characterized in that, In step S80, the periodic updating comprises: collecting the three-dimensional dynamic deformation field in a historical period as a training data set in step S801; performing time series feature extraction on the training data set using a Transformer encoder to obtain a feature vector in step S802; inputting the feature vector into a gated recurrent unit network to output corrected mechanical parameters through a gating mechanism, and updating the roadway digital twin model in step S803.

Images