An intelligent monitoring method and system for surrounding rock stability based on artificial intelligence

Abstract

This invention relates to the field of underground engineering monitoring technology, and discloses an intelligent monitoring method and system for surrounding rock stability based on artificial intelligence. It utilizes miniature sensors to collect initial index data in real time, performs noise reduction on the initial index data, removes abnormal and missing data through data cleaning, and obtains preprocessed data after normalization. Spatial features are extracted by constructing a sensor spatial topology map using a graph convolutional network, and temporal features are extracted by capturing the long-term dependency of deformation rate using a multi-head attention mechanism. After fusion processing, a fused feature vector is obtained. The fused feature vector is input into a prediction model to obtain the predicted deformation value and stability assessment result of the surrounding rock. When the stability state of the surrounding rock reaches instability, a risk warning signal is automatically issued. This invention improves the accuracy and reliability of surrounding rock stability prediction, providing strong technical support for the stability management of surrounding rock in underground engineering.

Description

Technical Field

[0001] This invention relates to the field of underground engineering monitoring technology, specifically to an intelligent monitoring method and system for surrounding rock stability based on artificial intelligence. Background Technology

[0002] Underground engineering plays an important role in infrastructure construction, such as tunnels and mine roadways. The stability of the surrounding rock is a key factor in the safety of underground engineering, and its monitoring is crucial to ensuring the safety of construction and operation. Traditional methods for monitoring the stability of the surrounding rock in underground engineering mainly rely on manual inspections and measurements from single sensors, which have problems such as low monitoring efficiency, poor real-time performance, and limited data processing capabilities. Summary of the Invention

[0003] To address the aforementioned problems, this invention presents an intelligent monitoring method and system for surrounding rock stability based on artificial intelligence.

[0004] The first aspect of this invention provides an intelligent monitoring method and system for surrounding rock stability based on artificial intelligence, the method comprising the following steps: Initial index data are collected in real time using miniature sensors, including at least the spray layer thickness, anchor bolt stress, and surrounding rock deformation. A noise reduction autoencoder is used to denoise the initial index data. Abnormal and missing data are removed by data cleaning, and the data is obtained after normalization. Based on the preprocessed data, a spatial topology graph of the sensor is constructed through a graph convolutional network to extract spatial features. A multi-head attention mechanism is used to capture the long-term dependence of deformation rate to extract temporal features. After fusion processing, a fused feature vector is obtained. By inputting the fused feature vector into the prediction model, the deformation prediction value and stability assessment result of the surrounding rock are obtained. When the stability of the surrounding rock reaches an unstable state, a risk warning signal is automatically issued to remind relevant personnel to take corresponding support and reinforcement measures.

[0005] Optionally, in a first implementation of the first aspect of the present invention, the step of using a noise-reducing autoencoder to denoise the initial index data, removing abnormal and missing data through data cleaning, and obtaining preprocessed data after normalization includes: The initial index data is input into the noise reduction autoencoder. Controllable noise is injected into the initial index data through Gaussian noise perturbation, random masking, and noise intensity control. The noise-reduced data is generated through the encoding-decoding process. A sliding window is constructed to identify outliers in the denoised data through unsupervised learning. If the outlier is a single point anomaly, the weighted average of the mean values ​​before and after the outlier and the mean values ​​of the three adjacent sensors at the same time is used for repair. If the outlier is an anomaly of multiple consecutive points, spatial interpolation of the nearby sensors is used as a substitute. For short-term missing data, cubic spline interpolation is used to complete the missing data. For long-term missing data, historical data is used to build a time series prediction model to predict and complete the missing data. The cleaned index data is normalized using the min-max normalization method to obtain preprocessed data.

[0006] Optionally, in a second implementation of the first aspect of the present invention, the step of constructing a sensor spatial topology graph using a graph convolutional network to extract spatial features based on preprocessed data, extracting temporal features by capturing the long-term dependency of deformation rate using a multi-head attention mechanism, and obtaining a fused feature vector after fusion processing includes: Based on the deployment location and network connectivity of the sensors, a sensor spatial topology map is constructed. Based on the sensor spatial topology map, a convolutional network is used to extract spatial features to obtain spatial features. The deformation rate sequence is calculated and input into a multi-head attention mechanism. Multiple parallel attention heads simultaneously focus on the dependencies at different time intervals. Each attention head calculates a weighted sum of Q, K, V and attention weights to capture the long-term dependency features of the deformation rate at different time scales. The spatial feature vector output by the graph convolutional network and the temporal feature vector output by the multi-head attention mechanism are concatenated along their dimensions to obtain a fused feature vector.

[0007] Optionally, in a third implementation of the first aspect of the present invention, the step of extracting spatial features using a graph convolutional network based on the sensor spatial topology map to obtain spatial features includes: The input layer organizes the preprocessed data into a node feature matrix, calculates the adjacency matrix and degree matrix based on the sensor spatial topology graph according to the sensor location, and adds self-loops; Graph convolutional layers normalize, propagate, linearly transform, and activate the data from the input layer. By aggregating the features of nodes and their neighboring nodes, they learn the spatial dependencies between sensors and obtain the final spatial features through multi-layer graph convolution operations.

[0008] Optionally, in the fourth implementation of the first aspect of the present invention, the prediction model adopts a dual-channel modeling architecture, the data-driven channel adopts a spatiotemporal Transformer network, and the physics engine channel is simulated based on the finite element method.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, the step of inputting the fused feature vector into the prediction model to obtain the deformation prediction value and stability assessment result of the surrounding rock includes: Positional encoding is added to the fused feature vector, and the encoded fused feature vector is input into the encoder layer of the spatiotemporal Transformer network. The multi-head self-attention module captures the spatial and temporal dependencies of the features. The decoder layer generates the predicted value of the surrounding rock deformation based on the feature vector output by the encoder, thus obtaining the sequence of predicted value of the surrounding rock deformation of the data-driven channel. Obtain a pre-established three-dimensional finite element model of the surrounding rock, set excavation boundary conditions and support conditions according to the current engineering status, calculate the stress, strain and displacement distribution of the surrounding rock under the current working conditions based on the finite element method, obtain the deformation results of the surrounding rock in physical simulation, and extract the deformation prediction value corresponding to the sensor position from the simulation results. A weighted fusion method is used to fuse the results of the data-driven channel and the physics engine channel to obtain the final predicted value of the surrounding rock deformation. The final predicted value of surrounding rock deformation is compared with the preset stability threshold to classify and evaluate the stability status of the surrounding rock, thus obtaining the stability evaluation result.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, the step of comparing the final predicted value of the surrounding rock deformation with a preset stability threshold to classify and evaluate the stability state of the surrounding rock and obtain a stability evaluation result includes: When the predicted value of the surrounding rock deformation is less than the stability threshold, the surrounding rock is determined to be in a stable state. When the predicted value of the surrounding rock deformation is between the stability threshold and the instability threshold, it is determined to be in a basically stable state. When the predicted value of surrounding rock deformation is greater than or equal to the instability threshold, it is determined to be in an unstable state.

[0011] A second aspect of this invention provides an intelligent monitoring method and system for surrounding rock stability based on artificial intelligence, the system comprising: The data acquisition module is used to collect initial index data in real time using micro sensors. The initial index data includes at least the spray layer thickness, anchor bolt stress, and surrounding rock deformation. The data processing module is used to denoise the initial index data using a noise reduction autoencoder, remove abnormal and missing data through data cleaning, and obtain the preprocessed data after normalization. The feature extraction module is used to extract spatial features based on the preprocessed data by constructing a sensor spatial topology map through a graph convolutional network, and to extract temporal features by capturing the long-term dependence of deformation rate using a multi-head attention mechanism. After fusion processing, a fused feature vector is obtained. The prediction module is used to input the fused feature vector into the prediction model to obtain the deformation prediction value and stability assessment result of the surrounding rock; The early warning module is used to automatically issue a risk warning signal when the stability of the surrounding rock reaches an unstable state, reminding relevant personnel to take corresponding support and reinforcement measures.

[0012] A third aspect of the present invention provides an intelligent monitoring device for surrounding rock stability based on artificial intelligence, the intelligent monitoring device for surrounding rock stability based on artificial intelligence includes a memory and at least one processor, the memory storing instructions; the at least one processor calls the instructions in the memory to cause the intelligent monitoring device for surrounding rock stability based on artificial intelligence to perform the various steps of the intelligent monitoring method for surrounding rock stability based on artificial intelligence as described in any of the preceding claims.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the artificial intelligence-based intelligent monitoring method for surrounding rock stability as described in any of the preceding claims.

[0014] The technical solution provided by this invention utilizes miniature sensors to collect initial index data in real time. A noise-reducing autoencoder is used to denoise the initial index data, and abnormal and missing data are removed through data cleaning. After normalization, preprocessed data is obtained. Based on the preprocessed data, a spatial topology map of the sensor is constructed using a graph convolutional network to extract spatial features. A multi-head attention mechanism is used to capture the long-term dependency of deformation rates and extract temporal features. After fusion processing, a fused feature vector is obtained. The fused feature vector is input into a prediction model to obtain the predicted deformation value and stability assessment result of the surrounding rock. When the stability state of the surrounding rock reaches instability, a risk warning signal is automatically issued to remind relevant personnel to take corresponding support and reinforcement measures. This invention achieves efficient processing and accurate analysis of monitoring data, automatically extracts key features from the data, improves the accuracy and reliability of surrounding rock stability prediction, and enhances the accuracy and effectiveness of risk warning, providing strong technical support for the stability management of surrounding rock in underground engineering. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 A flowchart illustrating an intelligent monitoring method for surrounding rock stability based on artificial intelligence, provided in an embodiment of the present invention. Figure 2A schematic diagram of the structure of an intelligent monitoring system for surrounding rock stability based on artificial intelligence provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an intelligent monitoring device for surrounding rock stability based on artificial intelligence, provided in an embodiment of the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The flowchart of the intelligent monitoring method for surrounding rock stability based on artificial intelligence provided in this embodiment of the invention includes the following steps: Step 101: Use micro sensors to collect initial index data in real time, including at least the spray layer thickness, anchor bolt stress and surrounding rock deformation. In this embodiment, at engineering sites such as tunnels or mines, a grid-based deployment method is adopted based on the characteristics of the surrounding rock structure and geological conditions. Sensors are arranged at key locations such as the tunnel arch, arch waist, and arch foot, as well as the sides, roof, and bottom of mine roadways. The sensor spacing is determined based on the preliminary assessment results of the surrounding rock stability, generally 5-10 meters. For areas with complex geological conditions, faults, or fracture zones, the sensor arrangement is densified, and the spacing is shortened to 3-5 meters to ensure the comprehensiveness and representativeness of the monitoring data. During sensor installation, positioning and fixing must be strictly carried out according to design requirements to ensure close contact between the sensor and the surrounding rock surface, avoiding data acquisition errors due to improper installation. At the same time, a sensor file is established to record the installation location, model, serial number, and other information of each sensor for subsequent maintenance and data management. Step 102: The initial index data is denoised using a noise reduction autoencoder. Abnormal and missing data are removed by data cleaning, and the data is normalized to obtain the preprocessed data. In this embodiment, initial index data is input into a noise reduction autoencoder. Controllable noise is injected into the initial index data through Gaussian noise perturbation, random masking, and noise intensity control. Noise-reduced data is generated through an encoding-decoding process. A sliding window is constructed, and outliers in the noise-reduced data are identified through unsupervised learning. If the outlier is a single point anomaly, the weighted average of the mean values ​​before and after the outlier and the mean values ​​of the three adjacent sensors at the same time is used for repair. If the outlier is an anomaly of multiple consecutive points, spatial interpolation of the nearest sensors is used for replacement. For short-term missing data, cubic spline interpolation is used for completion. For long-term missing data, a time series prediction model is built using historical data to predict and complete the missing data. The min-max normalization method is used to normalize the cleaned index data to obtain preprocessed data.

[0019] In this embodiment, a three-layer fully connected denoising autoencoder is constructed, which includes an input layer and a hidden layer. The hidden layer includes an encoding layer and a decoding layer. The number of neurons in the input layer is consistent with the dimension of the monitoring index. The encoding layer adopts a two-layer dimensionality reduction structure and uses ReLU as the activation function to achieve data feature compression. The decoding layer has a symmetrical two-layer dimensionality increase structure. The output layer uses a linear activation function to restore the original data dimension.

[0020] In this embodiment, outlier identification uses a combination of statistical methods and machine learning to calculate statistical quantities such as the mean and standard deviation of each indicator data, and sets a reasonable threshold range, such as mean ± 3 times standard deviation. Data exceeding the threshold is considered outlier data. At the same time, machine learning algorithms such as isolated forest and local outlier factor are used to detect outliers in the data to improve the accuracy of outlier identification. For abnormal data, it is processed according to its characteristics. If the abnormal data is a single point of anomaly, it is replaced by interpolation using the mean or median of data from adjacent time points; if it is an anomaly of multiple consecutive points, it is determined whether it is a sensor failure or a data transmission error by combining the on-site sensor status and the actual engineering situation. If it is a sensor failure, it is repaired or replaced in time, and the abnormal data segment of that sensor is removed in subsequent data processing. For short-term missing data, interpolation is suitable for completion because data from adjacent time points typically exhibit strong continuity and linear trends. Linear interpolation assumes a linear relationship between two valid data points before and after the missing point, calculating the value at the missing location using the slope of the line connecting the two points. Cubic spline interpolation further optimizes smoothness by constructing piecewise cubic polynomial curves to ensure continuity of function values, first derivatives, and second derivatives between adjacent intervals, avoiding abrupt changes that may occur with linear interpolation. It is particularly suitable for indicators with physical continuity, such as rock deformation. For example, when the rock deformation rate is stable over a certain period, cubic spline interpolation can more accurately fit the gradual change process at the missing time. The core advantages of these two methods are their reliance on locally adjacent data, low computational complexity, and ability to quickly recover short-term missing high-frequency data while preserving the trend characteristics of the original signal. They are suitable for short-period missing scenarios caused by temporary sensor signal interruptions or data transmission delays. When data is missing for a long period, simple interpolation methods are insufficient to capture long-term trends, periodicity, or nonlinear variations. In such cases, time series forecasting models are needed to predict and complete the missing data. First, based on complete historical data, typically a monitoring sequence of 1-7 days prior to the missing point is selected. The statistical characteristics of the data, such as mean, variance, and autocorrelation, are analyzed to select a suitable model, such as ARIMA or LSTM. Taking the ARIMA model as an example, the data needs to be stabilized first. The model order is determined using the autocorrelation function and partial autocorrelation function. Then, the parameters are fitted using maximum likelihood estimation. Finally, the values ​​for the missing time period are extrapolated based on the fitted model. For rock deformation data with significant nonlinear characteristics, LSTM neural networks are more advantageous. They capture dependencies in long-term sequences through memory units. For example, deformation data from the 24 hours prior to the missing point can be used as input to train the model to learn the time-deformation mapping pattern, thereby predicting missing values ​​for the next few hours. The key to model-based prediction lies in utilizing the overall patterns of historical data, rather than relying solely on local neighboring points. This effectively addresses long-term data gaps caused by sensor failures, network interruptions, and other factors. The completed data not only conforms to short-term trends but also reflects the long-term evolution of surrounding rock stability, such as the effects of periodic loads and the gradual deformation of geological structures. In engineering applications, it is usually necessary to adjust model parameters through cross-validation and evaluate prediction accuracy using indicators such as root mean square error to ensure the reliability of the completed data.

[0021] Step 103: Based on the preprocessed data, construct a sensor spatial topology graph through a graph convolutional network to extract spatial features, and use a multi-head attention mechanism to capture the long-term dependence of deformation rate to extract temporal features. After fusion processing, obtain a fused feature vector. In this embodiment, a sensor spatial topology map is constructed based on the sensor deployment location and network connectivity. Spatial features are extracted using a graph convolutional network based on the sensor spatial topology map to obtain spatial features. A deformation rate sequence is calculated and input into a multi-head attention mechanism. Multiple parallel attention heads simultaneously focus on the dependencies at different time intervals. Each attention head calculates a weighted sum of Q, K, V, and attention weights to capture the long-term dependency features of the deformation rate at different time scales. The spatial feature vector output by the graph convolutional network and the temporal feature vector output by the multi-head attention mechanism are concatenated dimensionally to obtain a fused feature vector.

[0022] In this embodiment, the input layer organizes the preprocessed data into a node feature matrix, calculates the adjacency matrix and degree matrix based on the sensor spatial topology graph according to the sensor location, and adds self-loops; the graph convolutional layer normalizes, propagates, linearly transforms and activates the data of the input layer, learns the spatial dependencies between sensors by aggregating the features of nodes and their adjacent nodes, and obtains the final spatial features through multi-layer graph convolution operations.

[0023] In this embodiment, a sensor spatial topology graph is constructed based on the deployment location and network connection relationship of the sensors. In the graph, nodes represent sensors, edges represent the spatial adjacency relationship between sensors, and the weight of the edge is determined by the reciprocal of the distance between the sensors. The closer the distance, the greater the weight, so as to reflect the spatial correlation between sensors.

[0024] In this embodiment, a self-loop is a special type of edge in graph theory, which refers to a node that is directly connected to itself through an edge, forming a closed loop. In an undirected graph, a self-loop is represented by an edge whose two endpoints are the same node; in a directed graph, it is a directed edge that starts from a node and returns to itself. For example, in a sensor space topology graph, if a sensor not only interacts with adjacent devices but also needs to model its own state, this self-association relationship can be represented by a self-loop.

[0025] Step 104: Input the fused feature vector into the prediction model to obtain the deformation prediction value and stability assessment result of the surrounding rock; In this embodiment, the prediction model adopts a dual-channel modeling architecture: the data-driven channel uses a spatiotemporal Transformer network, and the physics engine channel is simulated based on the finite element method.

[0026] In this embodiment, positional encoding is added to the fused feature vector. The encoded fused feature vector is input into the encoder layer of the spatiotemporal Transformer network. The multi-head self-attention module captures the spatial and temporal dependencies of the features. The decoder layer generates the predicted value of surrounding rock deformation based on the feature vector output by the encoder, thus obtaining the sequence of predicted value of surrounding rock deformation for the data-driven channel. A pre-established three-dimensional finite element model of the surrounding rock is acquired. Based on the current engineering state, excavation boundary conditions and support conditions are set. Based on the finite element method, the stress, strain, and displacement distribution of the surrounding rock under the current working condition are calculated to obtain the physical simulation result of the surrounding rock deformation. The predicted value of deformation corresponding to the sensor position is extracted from the simulation result. The results of the data-driven channel and the physics engine channel are fused using a weighted fusion method to obtain the final predicted value of surrounding rock deformation. The final predicted value of surrounding rock deformation is compared with a preset stability threshold to classify and evaluate the stability state of the surrounding rock, thus obtaining the stability evaluation result.

[0027] In this embodiment, when the predicted value of the surrounding rock deformation is less than the stability threshold, the surrounding rock is determined to be in a stable state; when the predicted value of the surrounding rock deformation is between the stability threshold and the instability threshold, it is determined to be in a basically stable state; when the predicted value of the surrounding rock deformation is greater than or equal to the instability threshold, it is determined to be in an unstable state.

[0028] In this embodiment, the spatiotemporal Transformer network structure includes an input layer, an encoder layer, and a decoder layer. The input layer takes the fused features as input, and after position encoding, it enters the encoder layer. The encoder layer consists of multiple multi-head self-attention modules and a feedforward neural network. It captures the spatial and temporal dependencies of the fused features through the self-attention mechanism to predict the deformation trend of the surrounding rock. The decoder layer generates the predicted value of the surrounding rock deformation based on the feature vector output by the encoder. The model training adopts a supervised learning method, using historical fusion features and corresponding actual values ​​of surrounding rock deformation as training data. The loss function is the mean squared error, and the optimizer is the Adam optimizer. The model parameters are adjusted through the backpropagation algorithm to minimize the error between the predicted value and the actual value. In this embodiment, the physics engine channel uses the finite element method to numerically simulate the rock mass constitutive equation. Based on engineering geological survey data, a three-dimensional finite element model of the surrounding rock is established. Tetrahedral or hexahedral elements are used to mesh the model, and the mesh density is determined according to the complexity of the surrounding rock and the required computational accuracy. Mesh refinement is performed in key areas. According to the lithology and mechanical properties of the surrounding rock, an appropriate rock mass constitutive equation is selected, such as the elastic constitutive equation or the elastoplastic constitutive equation. For rock masses with well-developed joints, a discrete element model or a contact mechanics model considering joint surfaces is used. An initial geostress field is applied, and the initial geostress is determined based on geological exploration data and field measurement results. Then, the engineering excavation process is simulated, taking into account the influence of factors such as excavation sequence and support measures on the stress and deformation of the surrounding rock. Finally, the stress, strain and displacement distribution of the surrounding rock under different working conditions are calculated to obtain the physical simulation results of the surrounding rock deformation.

[0029] Step 105: When the stability of the surrounding rock reaches an unstable state, an automatic risk warning signal is issued to remind relevant personnel to take corresponding support and reinforcement measures.

[0030] In this embodiment, when the stability of the surrounding rock is assessed as unstable, a risk warning signal is automatically triggered. The warning signal includes various methods such as audible alarms, flashing lights, SMS notifications, and platform pop-up prompts to ensure relevant personnel can receive warning information promptly. The warning signal contains detailed information such as the warning time, warning location, stability level, and predicted deformation value, facilitating a quick understanding of the potential danger. When a stable state warning is received, the normal monitoring frequency and inspection system are maintained, with regular checks and maintenance of monitoring equipment and support structures. When a basically stable state warning is received, the monitoring frequency is increased to once every 5 minutes, on-site inspections are strengthened, and close observation of surrounding rock deformation and the working status of the support structure is conducted. Targeted support and reinforcement plans are developed, and construction preparations are made. When an unstable state warning is received, the emergency plan is immediately activated, operations in the relevant area are stopped, personnel are evacuated to a safe area, and professional technicians and construction teams are notified. Emergency measures are implemented according to preset support and reinforcement measures, such as increasing the number of anchor bolts and cables, and shotcreting reinforcement. During the handling process, the deformation and stability of the surrounding rock are continuously monitored until the danger is brought under control.

[0031] Please see Figure 2 A schematic diagram of the structure of an intelligent monitoring system for surrounding rock stability based on artificial intelligence provided in this embodiment of the invention. The system includes: The data acquisition module is used to collect initial index data in real time using micro sensors. The initial index data includes at least the spray layer thickness, anchor bolt stress, and surrounding rock deformation. The data processing module is used to denoise the initial index data using a noise reduction autoencoder, remove abnormal and missing data through data cleaning, and obtain the preprocessed data after normalization. The feature extraction module is used to extract spatial features based on the preprocessed data by constructing a sensor spatial topology map through a graph convolutional network, and to extract temporal features by capturing the long-term dependence of deformation rate using a multi-head attention mechanism. After fusion processing, a fused feature vector is obtained. The prediction module is used to input the fused feature vector into the prediction model to obtain the deformation prediction value and stability assessment result of the surrounding rock; The early warning module is used to automatically issue a risk warning signal when the stability of the surrounding rock reaches an unstable state, reminding relevant personnel to take corresponding support and reinforcement measures.

[0032] In this embodiment, a micro-sensor unit is pre-integrated between the shotcrete and anchor bolt support system to form a structure-sensing integrated support system. The module mainly consists of a high-strength steel mesh, a reinforced fiber concrete layer, and multiple micro-strain and displacement sensing units, and is firmly installed on the surrounding rock surface by pre-embedded anchor bolts. After construction, the internal sensing elements can realize long-term online monitoring of key indicators such as shotcrete thickness, anchor bolt stress, and surrounding rock deformation, and transmit the data to the back-end platform in real time via a wireless network to achieve intelligent diagnosis and risk warning of the support status. It has both flexible and rigid support characteristics, is suitable for temporary or permanent support needs under different geological conditions, significantly improves the prediction and control capabilities of the surrounding rock stability of the cavern group, and has good prospects for engineering promotion and application.

[0033] Figure 3 This is a schematic diagram of the structure of an AI-based intelligent monitoring device for surrounding rock stability provided in an embodiment of the present invention. The AI-based intelligent monitoring device 300 for surrounding rock stability can vary significantly due to different configurations or performance characteristics. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) for storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the AI-based intelligent monitoring device 300 for surrounding rock stability. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the AI-based intelligent monitoring device 300 to implement the method provided in the above embodiment.

[0034] The AI-based intelligent monitoring device for surrounding rock stability 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating devices 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3The structure of the AI-based intelligent monitoring device for surrounding rock stability shown does not constitute a limitation on the computer equipment provided by this invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0035] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the various steps of the artificial intelligence-based intelligent monitoring method for surrounding rock stability provided in the above embodiments.

[0036] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described equipment or apparatus / unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0037] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0038] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples of the present invention and are not intended to limit the invention. Various changes and modifications can be made to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent monitoring method for surrounding rock stability based on artificial intelligence, characterized in that, The method includes the following steps: Initial index data are collected in real time using miniature sensors, including at least the spray layer thickness, anchor bolt stress, and surrounding rock deformation. A noise reduction autoencoder is used to denoise the initial index data. Abnormal and missing data are removed by data cleaning. After normalization, the preprocessed data is obtained. Based on the preprocessed data, a spatial topology graph of the sensor is constructed through a graph convolutional network to extract spatial features. A multi-head attention mechanism is used to capture the long-term dependence of deformation rate to extract temporal features. After fusion processing, a fused feature vector is obtained. By inputting the fused feature vector into the prediction model, the deformation prediction value and stability assessment result of the surrounding rock are obtained. When the stability of the surrounding rock reaches an unstable state, a risk warning signal is automatically issued to remind relevant personnel to take corresponding support and reinforcement measures. Based on the preprocessed data, a sensor spatial topology map is constructed using a graph convolutional network to extract spatial features. A multi-head attention mechanism is used to capture the long-term dependency of deformation rate to extract temporal features. After fusion processing, a fused feature vector is obtained, including: Based on the deployment location and network connectivity of the sensors, a sensor spatial topology map is constructed. Based on the sensor spatial topology map, a convolutional network is used to extract spatial features to obtain spatial features. The deformation rate sequence is calculated and input into a multi-head attention mechanism. Multiple parallel attention heads simultaneously focus on the dependencies at different time intervals. Each attention head calculates a weighted sum of Q, K, V and attention weights to capture the long-term dependency features of the deformation rate at different time scales. The spatial feature vector output by the graph convolutional network and the temporal feature vector output by the multi-head attention mechanism are concatenated by dimension to obtain the fused feature vector; The spatial feature extraction based on the sensor spatial topology map is performed using a graph convolutional network to obtain spatial features, including: The input layer organizes the preprocessed data into a node feature matrix, calculates the adjacency matrix and degree matrix based on the sensor spatial topology graph according to the sensor location, and adds self-loops. Here, a self-loop is a special kind of edge in graph theory, which refers to a node that is directly connected to itself through an edge, forming a closed loop. In an undirected graph, a self-loop is represented by an edge with both endpoints being the same node; in a directed graph, it is a directed edge that starts from a node and returns to itself. Graph convolutional layers normalize, propagate, linearly transform, and activate the data from the input layer. By aggregating the features of nodes and their neighboring nodes, they learn the spatial dependencies between sensors and obtain the final spatial features through multi-layer graph convolution operations.

2. The intelligent monitoring method for surrounding rock stability based on artificial intelligence as described in claim 1, characterized in that, The initial index data is denoised using a noise-reducing autoencoder. Abnormal and missing data are removed through data cleaning, and the data is normalized to obtain preprocessed data, including: The initial index data is input into the noise reduction autoencoder. Controllable noise is injected into the initial index data through Gaussian noise perturbation, random masking, and noise intensity control. The noise-reduced data is generated through the encoding-decoding process. A sliding window is constructed to identify outliers in the denoised data through unsupervised learning. If the outlier is a single point anomaly, the weighted average of the mean values ​​before and after the outlier and the mean values ​​of the three adjacent sensors at the same time is used for repair. If the outlier is an anomaly of multiple consecutive points, spatial interpolation of the nearby sensors is used as a substitute. For short-term missing data, cubic spline interpolation is used to complete the missing data. For long-term missing data, historical data is used to build a time series prediction model to predict and complete the missing data. The cleaned index data is normalized using the min-max normalization method to obtain preprocessed data.

3. The intelligent monitoring method for surrounding rock stability based on artificial intelligence as described in claim 1, characterized in that, The prediction model adopts a dual-channel modeling architecture: the data-driven channel uses a spatiotemporal Transformer network, and the physics engine channel is based on the finite element method for simulation.

4. The intelligent monitoring method for surrounding rock stability based on artificial intelligence as described in claim 3, characterized in that, The step of inputting the fused feature vector into the prediction model to obtain the deformation prediction value and stability assessment result of the surrounding rock includes: Positional encoding is added to the fused feature vector, and the encoded fused feature vector is input into the encoder layer of the spatiotemporal Transformer network. The multi-head self-attention module captures the spatial and temporal dependencies of the features. The decoder layer generates the predicted value of the surrounding rock deformation based on the feature vector output by the encoder, thus obtaining the sequence of predicted value of the surrounding rock deformation of the data-driven channel. Obtain a pre-established three-dimensional finite element model of the surrounding rock, set excavation boundary conditions and support conditions according to the current engineering status, calculate the stress, strain and displacement distribution of the surrounding rock under the current working conditions based on the finite element method, obtain the deformation results of the surrounding rock in physical simulation, and extract the deformation prediction value corresponding to the sensor position from the simulation results. A weighted fusion method is used to fuse the results of the data-driven channel and the physics engine channel to obtain the final predicted value of the surrounding rock deformation. The final predicted value of surrounding rock deformation is compared with the preset stability threshold to classify and evaluate the stability status of the surrounding rock, thus obtaining the stability evaluation result.

5. The intelligent monitoring method for surrounding rock stability based on artificial intelligence as described in claim 4, characterized in that, The step of comparing the final predicted value of the surrounding rock deformation with a preset stability threshold to classify and evaluate the stability state of the surrounding rock and obtain the stability evaluation result includes: When the predicted value of the surrounding rock deformation is less than the stability threshold, the surrounding rock is determined to be in a stable state. When the predicted value of the surrounding rock deformation is between the stability threshold and the instability threshold, it is determined to be in a basically stable state. When the predicted value of surrounding rock deformation is greater than or equal to the instability threshold, it is determined to be in an unstable state.

6. An intelligent monitoring system for surrounding rock stability based on artificial intelligence, used to execute the intelligent monitoring method for surrounding rock stability based on artificial intelligence as described in claim 1, characterized in that, The system includes: The data acquisition module is used to collect initial index data in real time using micro sensors. The initial index data includes at least the spray layer thickness, anchor bolt stress, and surrounding rock deformation. The data processing module is used to denoise the initial index data using a noise reduction autoencoder, remove abnormal and missing data through data cleaning, and obtain the preprocessed data after normalization. The feature extraction module is used to extract spatial features based on the preprocessed data by constructing a sensor spatial topology map through a graph convolutional network, and to extract temporal features by capturing the long-term dependence of deformation rate using a multi-head attention mechanism. After fusion processing, a fused feature vector is obtained. The prediction module is used to input the fused feature vector into the prediction model to obtain the deformation prediction value and stability assessment result of the surrounding rock; The early warning module is used to automatically issue a risk warning signal when the stability of the surrounding rock reaches an unstable state, reminding relevant personnel to take corresponding support and reinforcement measures.

7. An intelligent monitoring device for surrounding rock stability based on artificial intelligence, characterized in that, The AI-based intelligent monitoring device for surrounding rock stability includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the AI-based intelligent monitoring device for surrounding rock stability to execute each step of the AI-based intelligent monitoring method for surrounding rock stability as described in any one of claims 1-5.

8. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement each step of the intelligent monitoring method for surrounding rock stability based on artificial intelligence as described in any one of claims 1-5.

Images