An analysis method and system for the influence of hard rock blasting on the stability of a tunneling roadway

Abstract

This invention discloses a method and system for analyzing the impact of hard rock blasting on the stability of tunneling roadways, relating to the field of intelligent rock mass monitoring and analysis technology. The method includes: acquiring blasting parameters and surrounding rock attribute data from multiple consecutive blasting cycles; registering the data to construct a multi-source spatiotemporal dataset of the blasted surrounding rock; performing multi-dimensional reliability evaluation based on the multi-source spatiotemporal dataset, establishing a permissible blasting response domain, and identifying abnormal monitoring data in conjunction with blasting response constraints; constructing a dynamic spatiotemporal graph and adaptively updating the association weights between nodes based on the monitoring results of the current blasting cycle; aggregating neighborhood information of node features based on the dynamic spatiotemporal graph to obtain the blasting disturbance state and current surrounding rock damage state corresponding to each surrounding rock region; and comprehensively calculating the stability index of each monitoring region to analyze the impact on tunneling roadway stability. The method of this invention quantifies the fatigue deterioration and stress recovery effect of hard rock through a cumulative damage recursion mechanism.

Description

Technical Field

[0001] This invention relates to the field of intelligent analysis technology for rock mass monitoring, and in particular to an analysis method and system for the impact of hard rock blasting on the stability of tunnels. Background Technology

[0002] Periodic blasting dynamic loads can generate strong dynamic disturbances to the surrounding rock of tunnels, easily inducing the initiation and propagation of microcracks within the rock mass, and even evolving into engineering disasters such as rock bursts, spalling, or large-scale collapses. To accurately quantify the impact of blasting disturbances on the stability of the surrounding rock, multi-source heterogeneous monitoring data, including microseismic, acoustic emission, stress, and displacement data, are widely introduced into engineering sites. By collecting monitoring characteristics throughout the entire time period and using data-driven models, the nonlinear mechanical response of rocks under complex geological conditions can be mapped.

[0003] Existing methods for analyzing the stability of tunnel blasting are prone to sensor malfunctions or artifact noise in deep, humid, dusty, and electromagnetically disturbed underground environments. Current technologies typically input multi-source data directly into the model after simple threshold filtering, lacking response boundary constraints based on physical and mechanical mechanisms. They fail to establish a multi-dimensional reliability evaluation system for monitoring data, including data continuity and physical consistency, leading to confusion between sensor noise anomalies and actual physical anomalies in the surrounding rock, thus weakening the robustness of the evaluation results. When processing distributed monitoring data, existing analysis models often employ isolated single-point feature extraction or static spatial distance weighting, neglecting the spatiotemporal heterogeneity and topological dynamic evolution of blasting stress waves propagating in jointed and fractured rock masses. When local fracturing or stress redistribution occurs in the rock mass, the correlation response intensity between monitoring nodes inevitably changes abruptly, making it difficult to adaptively update this complex spatial topological map. Furthermore, they neglect the fatigue degradation, stress relaxation, and rheological recovery effects of hard rock under repeated cyclic dynamic impacts. Static analysis paradigms that lack historical memory and cumulative damage regression mechanisms cannot truly reflect the complex full life cycle characteristics of hard rock tunnels as they gradually evolve from microscopic damage to macroscopic instability. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an analytical method for the impact of hard rock blasting on the stability of tunnels, which solves the problems in the prior art where the reliability of multi-source monitoring data is difficult to quantify effectively, sensor anomalies are difficult to distinguish from the actual physical anomalies of the surrounding rock, and the dynamic spatiotemporal correlation between monitoring nodes is difficult to accurately characterize.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an analytical method for the impact of hard rock blasting on the stability of tunneling roadways, which includes acquiring blasting parameters and roadway surrounding rock property data in multiple consecutive tunneling blasting cycles, registering the data, and constructing a multi-source spatiotemporal dataset of blasting surrounding rock. Based on the multi-source spatiotemporal dataset, a multi-dimensional credibility evaluation is performed, a blasting response allowable domain is established, and abnormal monitoring data is identified in combination with blasting response constraints. Using the tunnel monitoring area as nodes and multi-source monitoring features as node features, a dynamic spatiotemporal map is constructed, and the association weights between nodes are adaptively updated based on the monitoring results of the current blasting cycle. Based on the dynamic spatiotemporal map, neighborhood information is aggregated for node features to obtain the blasting disturbance state corresponding to each surrounding rock area; cumulative damage is recursively calculated to obtain the current surrounding rock damage state; and the stability index of each monitoring area is comprehensively calculated to conduct an analysis of the impact on the stability of the tunnel.

[0007] As a preferred embodiment of the analytical method for the impact of hard rock blasting on the stability of tunnels described in this invention, the data registration includes, for continuous... The tunneling and blasting cycle is numbered, and the current cycle is the [number]th cycle. One loop; collect the blasting parameter set for each loop. To obtain information about the surface and deep layers of the surrounding rock in the tunnel. Multi-source heterogeneous monitoring data from the first sensor node; acquire the first... The detonation command trigger time of each cycle maps absolute system time to relative detonation time. Construct a unified time grid matrix and set the time step. For sampling intervals greater than The low-frequency stress and displacement data were upsampled using cubic spline interpolation and aligned to a unified time grid matrix.

[0008] As a preferred embodiment of the analytical method for the impact of hard rock blasting on the stability of tunneling roads described in this invention, the construction of a multi-source spatiotemporal dataset of the blasted surrounding rock includes, for sampling intervals less than... High-frequency vibration data, within a time step Peak features are obtained by performing max pooling within the sequence; for discrete microseismic and acoustic emission event sequences, the summation is distributed to the time of event occurrence. Within the time interval; the length of the generated time dimension is Synchronous multi-source timing matrix; During the tunneling process, the working face dynamically moves forward; a global coordinate system for the tunnel is established. Let the first The equivalent global coordinates of the center of the loop are: , No. The global coordinates of each sensor node are ; Transform the global coordinate system translation to the current equivalent center of the explosion. A relative local coordinate system with the origin; for each node Calculate the spatial topological feature vector, including absolute blast center distance, spatial azimuth, and structural surface intercept; The synchronized multi-source time series matrix after time synchronization is fused with the spatially registered topological feature vector; for continuous... Each iteration constructs a spatiotemporal dataset represented by a multidimensional tensor. .

[0009] As a preferred embodiment of the analytical method for the impact of hard rock blasting on the stability of tunneling roads as described in this invention, the step of conducting a multi-dimensional reliability evaluation includes, for the first... One blasting cycle, for any monitoring node The system calculates the state health score from four dimensions: data continuity score, physical consistency score, spatial correlation score, and historical drift score, and then performs a linear weighting to obtain the dynamic reliability of the node. ; Before extraction Using historical high-confidence monitoring data from the previous cycle, a nonlinear regression fit was performed to calculate the... Second blasting node Theoretical predicted mean response ; Calculate the probability density function of historical prediction residuals and set the confidence level; Construct nodes Explosive response allowable range ; Conduct initial screening to identify nodes In the Substituting the actual monitored peak value into the loop, if the actual monitored peak value does not fall within the allowable range of the blast response... If so, the anomaly detection mechanism will be triggered; When node When an exception is triggered, retrieve the node. The first-order candidate neighborhood node set in the spatial topology Participate in joint discrimination; calculate the blasting stress wave and elastic wave generated by potential surrounding rock fracture from the nodes based on the longitudinal and transverse wave velocities of the rock mass. Propagate to neighboring nodes Theoretical time delay Perform state coordination determination; If node An abnormal pulse is generated, in Within the time window, the set of candidate neighborhood nodes If all node monitoring values ​​are within the permissible range for blasting response and no coordinated response occurs, then the node is determined to be... The current data indicates a sensor malfunction. If node The monitored data exceeded the allowed range, and there was a corresponding time delay. Within the window, the set of candidate neighboring nodes If it also exhibits homologous abnormal fluctuations that exceed the allowable range, it is determined to be a real anomaly in the surrounding rock; Data identified as sensing anomalies is masked based on a set of candidate neighboring nodes. The data is spatially reconstructed and replaced using graph neural networks; for data determined to be real anomalies in the surrounding rock, high attention feature weights are assigned to nodes and affected neighborhood edges when constructing the graph network.

[0010] As a preferred embodiment of the analysis method for the impact of hard rock blasting on the stability of tunneling roadways described in this invention, wherein: the adaptive updating of the correlation weights between nodes based on the monitoring results of the current blasting cycle includes, for the first... One tunneling and blasting cycle, sensors are deployed in the surrounding rock of the tunnel. Each monitoring area is defined as a set of nodes in the graph; multi-source spatiotemporal datasets are extracted, and the nodes are... Vibration characteristics, stress and displacement state quantities, and calculated dynamic reliability The layers are stitched together, with dimension alignment performed using a fully connected layer, and initialized to the [number]th [dimension]. The node feature vector of each cycle This constitutes the node feature matrix of the entire graph; Without considering the real-time state of the nodes, the nodes are calculated based on the laws of rock mechanics. With nodes The initial physical edge weights between them are obtained by multiplying the physical decay terms together; To prevent errors from propagating in graph networks, nodes are used. and nodes In the Dynamic reliability of loops and The physical prior edge weights are corrected using a soft mask to obtain nodes with corrected credibility. and nodes The edge weights are used to construct the corrected physical adjacency matrix; An adaptive graph attention mechanism (GAT) is introduced, which utilizes the actual monitoring data features of the current loop to perform secondary optimization of weights; for nodes... and nodes eigenvectors and Perform a linear mapping and calculate the attention coefficient; The physical adjacency matrix is ​​used as a priori mask for the attention distribution, and normalized using the Softmax function to obtain the adaptively updated dynamic association weight matrix. .

[0011] As a preferred embodiment of the analytical method for the impact of hard rock blasting on the stability of tunneling roads described in this invention, the step of obtaining the current surrounding rock damage state includes, based on a dynamic correlation weight matrix... , for the The node features in each burst loop are spatially convolutionally aggregated; for any monitoring node... Using neighborhood node set The features are weighted and transmitted, and after multi-layer graph convolution operations, the hidden feature vectors of the fused spatial context of the region are obtained. ; The aggregated hidden feature vector The input is fed into a fully connected feedforward neural network for nonlinear mapping, dimensionality reduction, and decoding into the current blasting perturbation state vector. From the state of explosive disturbance Extract the incremental damage to the surrounding rock directly caused by the current single blast. ; Get the The absolute physical time interval between the next blast and the previous blasting cycle Construct a nonlinear memory decay function with physical time constraints. ; Set nodes In the The cyclic comprehensive surrounding rock damage state vector is Apply the time recovery decay function directly to the historical state to calculate the experience. Residual historical damage after the interval; The residual historical damage and the surrounding rock damage increment are fused to obtain the updated state. Current surrounding rock damage state after the end of each cycle .

[0012] As a preferred embodiment of the analytical method for the impact of hard rock blasting on the stability of tunneling roadways described in this invention, the comprehensive calculation of stability indicators for each monitoring area includes, for any monitoring area node within the roadway... Extract the first The feature vectors obtained from iterative calculations are mapped to four-dimensional physical quantification indicators, including damage indexes. Disturbance index Deformation rate index Abnormal frequency index ; The comprehensive stability index is calculated using a variable weight comprehensive evaluation model. Preset initial constant weights for four dimensional indicators; calculate the monitoring area. Comprehensive stability index Clustering was used to determine three critical thresholds. Based on the calculation The monitoring areas of the tunnel are divided into four risk zones: like This indicates a stable region; single-blast disturbance degree Within the range of material elastic recovery, the surrounding rock of the tunnel is in a safe state; like and This area is the dominant blasting sensitive zone; the cumulative damage in the area is not large, but it is extremely sensitive to the vibration of a single blast, and the charge structure and micro-delay time need to be adjusted in the next cycle. like and and The dominant area is the cumulative damage zone; when the rock mass experiences fatigue damage and creep, stronger support is required. like If the area is in a potentially unstable zone, the structural surface is about to be breached, posing a high risk. Tunneling must be stopped and emergency reinforcement carried out.

[0013] Secondly, this invention provides an analysis system for the impact of hard rock blasting on the stability of tunneling roadways, comprising: a multi-source data construction module for time synchronization and spatial registration of data to construct a multi-source spatiotemporal dataset of blasted surrounding rock; a data reliability analysis module connected to the multi-source data construction module for determining the dynamic reliability of each monitoring node; a blasting disturbance analysis module for constructing a dynamic spatiotemporal correlation model based on the spatial relationship between monitoring nodes, blasting wave propagation characteristics, geological connection relationships, and dynamic reliability; and a stability analysis module for determining the stability state of different areas of the tunneling roadway by comprehensively considering the cumulative damage state, blasting disturbance state, surrounding rock deformation characteristics, and disturbance propagation characteristics.

[0014] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the method for analyzing the impact of hard rock blasting on the stability of tunneling as described in the first aspect of the present invention.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for analyzing the impact of hard rock blasting on the stability of tunneling roadways as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By using multi-dimensional reliability evaluation and a blasting response allowable domain based on physical laws, it can accurately separate sensor noise caused by harsh downhole environments from actual surrounding rock fracture signals, fundamentally ensuring the reliability of monitoring and evaluation data. Constructing a dynamic spatiotemporal diagram and adaptively updating node association weights based on monitoring results breaks the limitations of traditional static spatial weighting, truly mapping the dynamic propagation mechanism and spatial correlation evolution of blasting stress waves in complex rock structures. Introducing a cumulative damage recursion mechanism with historical memory upgrades the fragmented static assessment of a single blast to a dynamic progressive assessment under continuous tunneling cycles, scientifically quantifying the fatigue degradation and stress recovery effects of hard rock. By comprehensively calculating multi-dimensional stability indicators and performing refined four-level risk zoning, it provides an intuitive and quantitative scientific basis for the adaptive optimization of on-site blasting parameters and the dynamic adjustment of support schemes. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the analytical method for the impact of hard rock blasting on the stability of tunnels.

[0019] Figure 2 A schematic diagram of a computer device used to analyze the impact of hard rock blasting on the stability of tunnels. Detailed Implementation

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0021] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0022] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0023] Reference Figures 1-2This is one embodiment of the present invention, which provides a method for analyzing the impact of hard rock blasting on the stability of tunnels, comprising the following steps: S1: Obtain blasting parameters and roadway surrounding rock attribute data from multiple consecutive tunneling and blasting cycles, register the data, and construct a multi-source spatiotemporal dataset of blasting surrounding rock.

[0024] The data registration includes, for continuous... The tunneling and blasting cycle is numbered, and the current cycle is the [number]th cycle. One loop; collect the blasting parameter set for each loop. To obtain information about the surface and deep layers of the surrounding rock in the tunnel. Multi-source heterogeneous monitoring data from the first sensor node; acquire the first... The detonation command trigger time of each cycle maps absolute system time to relative detonation time. Construct a unified time grid matrix and set the time step. For sampling intervals greater than The low-frequency stress and displacement data were upsampled using cubic spline interpolation and aligned to a unified time grid matrix.

[0025] Furthermore, for continuous The tunneling and blasting cycle is numbered, and the current cycle is the [number]th cycle. One cycle, Collect the set of blasting parameters for the current loop. The surrounding rock property data of the tunnel is obtained by deploying data on the surface and deep layers of the surrounding rock of the tunnel. Multi-source heterogeneous monitoring data collected by sensor nodes.

[0026] The blasting parameters include the three-dimensional spatial coordinates of each borehole, the charge amount of each borehole, the delay time of the differential detonator, and the trigger time of the detonation command.

[0027] The surrounding rock attribute data of the tunnel includes surrounding rock geological structure data and surrounding rock dynamic monitoring response data.

[0028] The surrounding rock geological structure data includes longitudinal wave velocity, transverse wave velocity, elastic modulus, distribution location of rock layer joints, fissures and faults, thickness of weak interlayers, and energy transmission coefficient of structural surfaces.

[0029] The dynamic monitoring response data of the surrounding rock includes time-varying peak velocity (PPV), dominant frequency and band energy, acoustic emission / microseismic events, and surrounding rock stress and displacement data collected by sensor nodes.

[0030] Due to the differences in data structures among different types of sensors, feature pre-extraction is performed first. This includes high-frequency waveform data, such as vibration, acoustic emission / microseismic event data, etc. Vibration sensors typically have sampling frequencies in the kHz range; short-time Fourier transform is used to extract time-varying peak velocity (PPV), dominant frequency, and bandgap energy. For acoustic emission / microseismic events, the occurrence time, three-dimensional location coordinates, ring count, and released energy for each event are extracted. Low-frequency state data includes stress and displacement data. Sampling frequencies are typically in the Hz or minute range; their discrete time-series values ​​are directly retained.

[0031] Due to the cross-scale differences in the sampling rates of various sensors, a time synchronization algorithm based on a unified clock network and detonation origin is adopted to unify the absolute system time of the hardware of each monitoring device.

[0032] Get the The detonation command trigger time for each cycle Mapping absolute system time to relative blast time Construct a unified time grid matrix and set the base time step. Perform alignment operations on data with different sampling rates.

[0033] For sampling interval greater than For low-frequency stress and displacement data, cubic spline interpolation was used for upsampling interpolation, aligning them to a uniform time grid. For sampling intervals less than [missing information], [missing information]... High-frequency vibration data, within a time step Peak features are obtained by using the maximum value pool.

[0034] For discrete acoustic emission / microseismic event sequences, they are accumulated and distributed to the time of event occurrence. Within a time interval. This generates a time dimension of length [length missing]. Synchronous multi-source timing matrix .

[0035] During tunneling, the blast source face dynamically moves forward. A global coordinate system for the tunnel is established. Let the first... The equivalent global coordinates of the center of the loop are: , No. The global coordinates of each sensor node are To eliminate spatial coordinate drift caused by tunneling, the global coordinate system translation is converted to be based on the current equivalent blast center. A relative local coordinate system with the origin. For each sensor node. Calculate its spatial topological feature vector .

[0036] Absolute Heart Distance The formula is expressed as: Spatial azimuth, including yaw and pitch angles, describes the spatial orientation of a node relative to the blast center and the tunnel axis.

[0037] Structural intercept representation, calculation of sensor and heartbreaking The number and thickness of the lines that pass through known weak interlayers or faults characterize the discontinuity of the medium.

[0038] Furthermore, for each sensor node Constructing the current equivalent explosion core Point to monitoring node The propagation path of the blast wave is calculated, and the intersection characteristics of the propagation path with the structural planes of faults, joints and weak interlayers are calculated. The structural plane intersection characteristics include the number of intersecting structural planes, the distance from the intersection point of each structural plane to the equivalent blast center, the penetration length of the propagation path in each weak interlayer, and the energy transmission coefficient of each structural plane, which are used to characterize the blocking and attenuation effect of geological structures on the propagation of blast stress waves.

[0039] The construction of the multi-source spatiotemporal dataset for blasted surrounding rock includes, for sampling intervals less than... High-frequency vibration data, within a time step Peak features are obtained by performing max pooling within the sequence; for discrete microseismic and acoustic emission event sequences, the summation is distributed to the time of event occurrence. Within the time interval; the length of the generated time dimension is Synchronous multi-source timing matrix.

[0040] During the tunneling process, the working face dynamically moves forward; a global coordinate system for the tunnel is established. Let the first The equivalent global coordinates of the center of the loop are: , No. The global coordinates of each sensor node are .

[0041] Transform the global coordinate system translation to the current equivalent center of the explosion. A relative local coordinate system with the origin; for each node Calculate the spatial topological feature vector, including absolute burst center distance, spatial azimuth, and structural intercept.

[0042] The time-synchronized multi-source time series matrix and the spatially registered topological feature vector are then compared. Perform feature splicing and fusion.

[0043] For continuous Each iteration constructs a spatiotemporal dataset represented by a multidimensional tensor. Among them, the input feature tensor ,Include Each node Within a time step 3D monitoring features, 3D relative spatial topological features and 3D burst parameter features (extended to the time dimension via broadcasting mechanism); output label tensor This dataset represents the actual damage or deformation state of the surrounding rock, as assessed based on subsequent monitoring data. The tensor structure of this dataset is directly used as input to the subsequent Dynamic Spatiotemporal Graph Neural Network (GNN).

[0044] S2: Based on the multi-source spatiotemporal dataset, perform multi-dimensional credibility evaluation, establish the blasting response allowable domain, and identify abnormal monitoring data in combination with blasting response constraints.

[0045] The multidimensional credibility evaluation includes, for the first One blasting cycle, for any monitoring node The system calculates the state health score from four dimensions: data continuity score, physical consistency score, spatial correlation score, and historical drift score, and then performs a linear weighting to obtain the dynamic reliability of the node. .

[0046] Before extraction Using historical high-confidence monitoring data from the previous cycle, a nonlinear regression fit was performed to calculate the... Second blasting node Theoretical predicted mean response ; Calculate the probability density function of historical prediction residuals and set the confidence level; Construct nodes Explosive response allowable range .

[0047] Perform initial screening to identify nodes In the Actual monitoring peak value of the cycle Substitute, if If this occurs, the anomaly detection mechanism will be triggered.

[0048] When node When an exception is triggered, retrieve the node. The first-order candidate neighborhood node set in the spatial topology Participate in joint discrimination; calculate the blasting stress wave and elastic wave generated by potential surrounding rock fracture from the nodes based on the longitudinal and transverse wave velocities of the rock mass. Propagate to neighboring nodes Theoretical time delay Then, perform state coordination determination.

[0049] If node An abnormal pulse is generated, in Within the time window, the set of candidate neighborhood nodes If all node monitoring values ​​are within the permissible range for blasting response and no coordinated response occurs, then the node is determined to be... The current data indicates a sensor malfunction.

[0050] If node The monitored data exceeded the allowed range, and there was a corresponding time delay. Within the window, the set of candidate neighboring nodes If it also exhibits homologous abnormal fluctuations that exceed the allowable range, it is determined to be a true anomaly in the surrounding rock.

[0051] Data identified as sensing anomalies is masked based on a set of candidate neighboring nodes. The data is spatially reconstructed and replaced using graph neural networks; for data determined to be real anomalies in the surrounding rock, high attention feature weights are assigned to nodes and affected neighborhood edges when constructing the graph network.

[0052] Furthermore, regarding the first One blasting cycle, for any monitoring node The node's health score is calculated from four independent dimensions, and then linearly weighted using either entropy weighting or expert experience weighting to obtain the node's dynamic credibility. : Data continuity score : Count the nodes within the current loop time window The effective sampling point missing rate and communication interruption frequency are considered. The higher the missing rate, the lower the score.

[0053] Physical consistency score This involves examining the coupling relationship between different physical quantities at the same node. For example, according to elasticity, stress increment should be positively correlated with strain / displacement increment; the time-series Pearson correlation coefficient between the two is calculated, and if a complete divergence occurs, the current score is reduced.

[0054] Spatial correlation score Extracting nodes and its neighboring nodes (distance radius) Calculate the cosine similarity of the feature vectors for the low-frequency state baseline features (within the region). If the node The baseline showed a sharp, isolated jump, resulting in a decrease in score.

[0055] Historical drift score Using a sliding window to extract nodes in the past The baseline mean is used in each iteration to calculate the drift rate of the current baseline mean relative to the historical mean. Finally, the dynamic reliability of the node is calculated by weighted summation of the four components. .

[0056] To determine whether the data is abnormal, a physical benchmark based on rock dynamics needs to be introduced to establish an allowable range for the blast response at a specific confidence level.

[0057] Based on the monitored blasting vibration velocity For example, before extraction Historical high-confidence monitoring data from each cycle were used for nonlinear regression fitting using the improved Sadovsky formula, expressed as follows: in, For the amount of explosives, The distance between the centers of the explosion; and To match the wave velocity of the rock mass Dynamic attenuation parameters related to geological conditions.

[0058] Based on the aforementioned nonlinear regression fitting, the first... Second blasting node Theoretical predicted mean response In this embodiment, it is a node. The theoretical blasting vibration prediction velocity at the location.

[0059] Calculate the probability density function of historical prediction residuals and set confidence levels, such as the boundary coefficients corresponding to a 95% confidence level. Nodes are thus constructed. Allowable range of blast response: Perform initial screening to identify nodes In the Actual peak monitoring speed of the cycle Substitute and compare, if If this occurs, the anomaly detection mechanism will be triggered.

[0060] When node When an anomaly is triggered, the cause could be sensor malfunction, such as loosening or electromagnetic interference, or it could be due to actual micro-fractures or instability in the surrounding rock. In this case, the node should be retrieved. The first-order candidate neighborhood node set in the spatial topology Participate in joint judgment.

[0061] Propagation delay calculation: based on the longitudinal wave velocity of the rock mass and transverse wave velocity Calculate the blasting stress wave and the elastic wave generated by potential surrounding rock fracture from the node. Propagate to neighboring nodes , Theoretical time delay .

[0062] If the node is identified as having a sensor malfunction, then... An abnormal pulse is generated, but time delay is taken into account. Time window Within, the set of candidate neighboring nodes If the monitoring values ​​of the vast majority of nodes remain stable within their respective blasting response allowable ranges and no coordinated response occurs, then the nodes are considered to be... The current data is either sensor artifacts or communication interference. Represents a node The start time of the abnormal pulse. This indicates the propagation delay tolerance.

[0063] If node The monitoring data significantly exceeded the allowed range, and there was a corresponding time delay. Within the window, the set of candidate neighboring nodes If the rock exhibits abnormal fluctuations from the same source that exceed the permissible range, or abrupt changes in stress or displacement steps, and conforms to the spatial propagation law of stress waves or energy release, then it is determined to be a real anomaly in the surrounding rock, and real micro-fractures or structural plane slippage have occurred in the current area.

[0064] Data identified as sensor anomalies is masked, and a candidate neighborhood node set is used. The data is spatially reconstructed and replaced using a graph neural network, while the dynamic reliability of the lower adjustment point in the next cycle is also considered.

[0065] For data identified as real anomalies in the surrounding rock, it is retained as a real label. In step S3, when constructing the graph network, higher attention feature weights are assigned to the node and its affected neighborhood edges so that the model can focus on the high-risk area.

[0066] S3: Using the tunnel monitoring area as nodes and multi-source monitoring features as node features, a dynamic spatiotemporal map is constructed, and the association weights between nodes are adaptively updated based on the monitoring results of the current blasting cycle.

[0067] The adaptive update of the association weights between nodes based on the monitoring results of the current blasting cycle includes, for the first... One tunneling and blasting cycle, sensors are deployed in the surrounding rock of the tunnel. Each monitoring area is defined as a set of nodes in the graph; multi-source spatiotemporal datasets are extracted, and the nodes are... Vibration characteristics, stress and displacement state quantities, and calculated dynamic reliability The layers are stitched together, with dimension alignment performed using a fully connected layer, and initialized to the [number]th [dimension]. The node feature vector of each cycle The node feature matrix that constitutes the entire graph .

[0068] Without considering the real-time state of the nodes, the nodes are calculated based on the laws of rock mechanics. With nodes Initial physical edge weights between The weights are obtained by multiplying the physical decay terms together.

[0069] To prevent errors from propagating in graph networks, nodes are used. and nodes In the Dynamic reliability of loops and The physical prior edge weights are corrected using a soft mask to obtain the confidence-corrected edge weights. Construct the corrected physical adjacency matrix , Represents the physical adjacency matrix No. line, number The elements of a column, i.e., nodes With nodes The specific edge weights after credibility correction.

[0070] An adaptive graph attention mechanism (GAT) is introduced, which utilizes the actual monitoring data features of the current loop to perform secondary optimization of weights; for nodes... and nodes eigenvectors and Perform a linear mapping and calculate the attention coefficient.

[0071] Physical adjacency matrix The prior mask used for the attention distribution is normalized using the Softmax function to obtain the adaptively updated dynamic association weights. .

[0072] Furthermore, taking the tunnel monitoring area as a node and multi-source monitoring features as node features, a dynamic spatiotemporal map is constructed based on node spatial distance, blast wave propagation time, rock mass structure relationship, and node dynamic reliability. The association weights between nodes are adaptively updated according to the monitoring results of the current blasting cycle. The specific steps include the following.

[0073] Regarding the first One tunneling and blasting cycle, sensors are deployed in the surrounding rock of the tunnel. Each monitoring region is defined as a set of nodes in the graph. The multi-source spatiotemporal dataset, after anomaly cleaning, is extracted, and the nodes are... Vibration characteristics (such as PPV, dominant frequency), stress / displacement state variables, and calculated dynamic reliability. The layers are stitched together and aligned to dimension using a fully connected layer, initialized to the first dimension. The node feature vector of each cycle This constitutes the node feature matrix of the entire graph: Without considering the real-time state of the nodes, the nodes are calculated based on the laws of rock mechanics. With nodes Initial physical edge weights between This weight is obtained by multiplying three physical decay terms together.

[0074] Spatial distance attenuation term, based on the Euclidean distance between the two nodes. Construct Gaussian radial basis functions: in, This is the distance scaling control parameter.

[0075] Explosion wave propagation time correlation term : Get the current equivalent core burst to the node and nodes Theoretical stress wave arrival time and .like The smaller the value, the more closely the two nodes are positioned on the blast wave front, and their vibration response exhibits a high degree of resonance at the same frequency. A correlation term is defined, and the formula is expressed as: in, This is the time decay coefficient.

[0076] For rock mass structure blocking, ray tracing is performed using a 3D geological model to detect nodes. With nodes Number of known weak interlayers or fault planes through which the line passes Introducing a medium penetration attenuation model: in, For the first The energy transmission coefficients of each structural surface are synthesized to obtain the elements in the physical prior adjacency matrix: Considering that individual monitoring areas may be severely disturbed in the current cycle, resulting in data distortion and low dynamic reliability, the weight of their influence on neighboring nodes should be weakened to prevent errors from propagating in the graph network.

[0077] Utilizing nodes and nodes In the Dynamic reliability of loops and The physical prior edge weights are corrected using a softmask to obtain the confidence-corrected edge weights: Based on this, a revised physical adjacency matrix is ​​constructed. The matrix defines the physical constraints of information transmission in a graph neural network.

[0078] The distance cutoff threshold is determined based on the sensor spacing, the effective propagation range of blasting disturbance, and the rock mass wave velocity. The Euclidean distance between the two nodes No more than Nodes that are not completely blocked by structural surfaces are included in the node list. Neighborhood node set based on physical distance truncation .

[0079] Although the physical adjacency matrix provides the theoretical connection strength, the actual mechanical response relationship changes dynamically due to the nonlinear rock-breaking process of each blast. An adaptive graph attention mechanism (GAT) is introduced to perform secondary optimization of the weights using the actual monitoring data features of the current cycle.

[0080] For nodes and nodes eigenvectors and Perform a linear mapping and calculate the attention coefficient. This characterizes the degree of interdependence between the actual response features of the two nodes under the current blasting cycle: in, The weight matrix is ​​a learnable feature transformation matrix. For the feedforward network parameters of the attention mechanism, This indicates feature splicing.

[0081] Physical adjacency matrix As a priori mask for attention distribution, it ensures that attention allocation obeys physical laws (i.e., no attention is allocated between nodes that are physically completely blocked by structural surfaces). Finally, it is normalized through the Softmax function to obtain the adaptively updated dynamic association weights. : in, For nodes truncated based on physical distance The set of neighboring nodes. This weight matrix. This will serve as the core operator for node feature space aggregation in step S4.

[0082] S4: Based on the dynamic spatiotemporal map, the neighborhood information of the node features is aggregated to obtain the blasting disturbance state corresponding to each surrounding rock area; cumulative damage is recursively calculated to obtain the current surrounding rock damage state; the stability index of each monitoring area is comprehensively calculated to conduct stability impact analysis of the tunnel.

[0083] Obtaining the current surrounding rock damage state includes, based on the dynamic correlation weight matrix. , for the The node features in each burst loop are spatially convolutionally aggregated; for any monitoring node... Using neighborhood node set The features are weighted and transmitted, and after multi-layer graph convolution operations, the hidden feature vectors of the fused spatial context of the region are obtained. .

[0084] The aggregated hidden feature vector The input is fed into a fully connected feedforward neural network for nonlinear mapping, dimensionality reduction, and decoding into the current blasting perturbation state vector. From the state of explosive disturbance Extract the incremental damage to the surrounding rock directly caused by the current single blast. .

[0085] Get the The absolute physical time interval between the next blast and the previous blasting cycle Construct a nonlinear memory decay function with physical time constraints. .

[0086] Set nodes In the The cyclic comprehensive surrounding rock damage state vector is Apply the time recovery decay function directly to the historical state to calculate the experience. Equivalent residual historical damage after interval .

[0087] The decayed historical state is fused with the current damage increment to update the state, resulting in the [number]th [state]. Current surrounding rock damage state after the end of each cycle .

[0088] Furthermore, based on the dynamic spatiotemporal graph, neighborhood information is aggregated for node features to obtain the blasting disturbance state corresponding to each surrounding rock area; combined with the surrounding rock damage state of the previous blasting cycle, the current blasting disturbance state, and the time interval between adjacent blasting cycles, cumulative damage recursion with historical memory and recovery effect is performed to obtain the current surrounding rock damage state.

[0089] Based on the adaptively updated dynamic correlation weight matrix obtained in step S3 , for the The node features in each burst loop are spatially convolutionally aggregated. For any monitoring node... Utilizing its neighborhood node set The features are weighted and passed through multiple layers of graph convolution (GCN) to obtain the hidden feature vector of the fused spatial context of the region. : in, The learnable weight matrix of the graph convolutional layer. For bias terms, It is a non-linear activation function.

[0090] The aggregated hidden feature vector The input is fed into a fully connected feedforward neural network for nonlinear mapping, which reduces the dimensionality and decodes it into a current blasting perturbation state vector with mechanical representation significance. .

[0091] Furthermore, using a multilayer perceptron (MLP) pre-trained based on rock damage constitutive relations, the blasting disturbance state is analyzed. Extract the incremental damage to the surrounding rock directly caused by the current single blast. .

[0092] Get the The next blast cycles with the previous blast (the... The absolute physical time interval between (times) Considering the rheological properties of surrounding rock in geotechnical engineering (such as stress relaxation and compressive closure of microcracks), the apparent dynamic damage caused by the previous blast will recover or redistribute to some extent as time passes. A nonlinear memory decay function with physical time constraints is constructed. : in, The rheological recovery coefficient of the rock mass is obtained through well logging data or rock creep tests; The residual irreversible damage baseline coefficient represents the permanent plastic failure boundary that cannot be recovered regardless of the time interval.

[0093] Instead of the purely data-driven forget gate mechanism in traditional gated recurrent units (GRUs), a physical mechanism-driven cumulative damage recursion unit is constructed.

[0094] Set nodes In the The cyclic comprehensive surrounding rock damage state vector is Applying the time recovery decay function directly to the historical state, the experience is calculated. Equivalent residual historical damage after interval : Subsequently, the decayed historical state is fused with the current damage increment to update the state to obtain the [number]th [state]. Current surrounding rock damage state after the end of each cycle : in, and The parameters of the state transition matrix are... This represents vector concatenation. This represents the Hadamard product (element-to-element multiplication). This recursive process retains the characteristics of long-term sequences while strictly adhering to the time accumulation and decay laws in rock mechanics. Indicates the current state of blasting disturbance. Extract the incremental damage to the surrounding rock directly caused by the current single blast.

[0095] For any monitoring area node within the tunnel Extract the first The feature vectors obtained from the iterative calculations are mapped to four dimensions of physical quantification indicators: Damage index The calculated current surrounding rock damage state vector The scalar norm is calculated and then normalized using Min-Max to characterize the macroscopic development of microcracks within the rock mass.

[0096] Disturbance index Extract the blasting disturbance state vector The vibration kinetic energy term in the equation, combined with the elastic wave impedance of the rock mass in the region, is used to calculate the additional dynamic ratio caused by a single blast.

[0097] Deformation rate index Extracting nodes from raw low-frequency monitoring data The convergence displacement increment, divided by the time interval This yields the rate of change of displacement for the current period.

[0098] Anomaly frequency index Historical records identified as genuine anomalies in the surrounding rock were traced back, and calculation nodes were established. and its first-order neighbors in the past The frequency of actual exception events occurring within a loop.

[0099] To prevent misjudgments caused by the failure of a single indicator, a variable-weight comprehensive evaluation model is used to calculate the comprehensive stability index. A non-linear penalty weighting mechanism based on the degree of indicator deterioration is introduced: Let the initial constant weights of the four dimension indicators be... If a single indicator (such as deformation rate) If the value increases dramatically and approaches its physical limit threshold, the danger weight of this indicator should be amplified exponentially. Calculate the dynamic penalty weight. : in, For the first Observed values ​​of the indicators This is the risk amplification factor. After normalizing the weights, the monitoring area is calculated. Comprehensive stability index (range of values) (A higher value indicates a higher risk of instability). Three critical thresholds are determined using K-Means clustering or a support vector machine (SVM) trained based on historical instability cases. (in According to the calculation Based on the characteristics of individual indicators, the monitoring areas of the roadway are divided into four risk zones: Stable region ( The overall stability index is low, and the disturbance degree of a single blast is also low. Within the range of material elastic recovery, the surrounding rock of the tunnel is in a safe state.

[0100] Explosion-sensitive areas ( and Dominant): The overall index is at a low to medium level, but the disturbance index... Highlighted. This indicates that the cumulative damage in this area is not significant, but it is extremely sensitive to the vibrations of a single blast (such as being in a stress concentration zone or a zone where blast waves overlap), requiring adjustments to the charge structure or timing in the next cycle.

[0101] Cumulative damage area ( and or (Dominant): The overall indicators are at a medium-to-high level. Individual disturbances are not significant, but the degree of damage or deformation rate is significant. This indicates that the rock mass has experienced fatigue damage or creep, and stronger support is required.

[0102] Potential instability zone ( ): Extremely high comprehensive indicators or dense surrounding rock anomalies. This indicates that the structural plane is about to be broken through or there is an extremely high risk of rockburst / collapse, requiring immediate cessation of excavation, evacuation of personnel, and emergency reinforcement.

[0103] This embodiment also provides an analysis system for the impact of hard rock blasting on the stability of tunnels, including: The multi-source data construction module performs time synchronization and spatial registration of data to construct a multi-source spatiotemporal dataset for blasting surrounding rock.

[0104] The data reliability analysis module, connected to the multi-source data construction module, is used to determine the dynamic reliability of each monitoring node.

[0105] The blasting disturbance analysis module is used to construct a physical adjacency matrix based on the initial physical edge weights and dynamic credibility between monitoring nodes, define the physical constraints of information transmission in the graph neural network, and calculate the dynamic association weights to obtain a dynamic spatiotemporal association model.

[0106] The stability analysis module comprehensively considers cumulative damage, blasting disturbance, surrounding rock deformation characteristics, and disturbance propagation characteristics to determine the stability status of different areas of the tunnel.

[0107] The dynamic spatiotemporal correlation model is a graph model composed of monitoring area nodes, node features, and dynamic correlation weights between nodes. Spatial correlation is represented by the distance between nodes and the attenuation effect of structural surfaces, while temporal correlation is represented by the arrival time difference of the blast wave. Each blasting cycle updates the correlation weights based on the current monitoring features and dynamic reliability, describing the perturbation correlation between different monitoring areas as the blasting process progresses.

[0108] This embodiment also provides a computer device, such as... Figure 2 As shown, the method for analyzing the impact of hard rock blasting on the stability of tunneling includes: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the method for analyzing the impact of hard rock blasting on the stability of tunneling as proposed in the above embodiments.

[0109] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0110] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the analysis method for the impact of hard rock blasting on the stability of tunnels as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0111] In summary, this invention achieves the following: 1) Obtaining blasting parameters and surrounding rock attribute data from multiple consecutive tunneling blasting cycles; 2) Registering the data to construct a multi-source spatiotemporal dataset of the blasting surrounding rock; 3) Based on the multi-source spatiotemporal dataset, performing multi-dimensional reliability evaluation, establishing a permissible blasting response domain, and identifying abnormal monitoring data in conjunction with blasting response constraints; 4) Using the tunnel monitoring area as a node and multi-source monitoring features as node features, constructing a dynamic spatiotemporal graph, and adaptively updating the association weights between nodes based on the monitoring results of the current blasting cycle; 5) Aggregating neighborhood information of node features based on the dynamic spatiotemporal graph to obtain the blasting disturbance state corresponding to each surrounding rock area; 6) Performing cumulative damage recursion to obtain the current surrounding rock damage state; 7) Comprehensively calculating the stability index of each monitoring area to conduct a stability impact analysis of the tunneling roadway.

[0112] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for analyzing the impact of hard rock blasting on the stability of tunnels, characterized in that: This includes acquiring blasting parameters and roadway surrounding rock attribute data from multiple consecutive tunneling and blasting cycles, registering the data, and constructing a multi-source spatiotemporal dataset of blasting surrounding rock. Based on multi-source spatiotemporal datasets, multi-dimensional credibility evaluation is performed, a blasting response allowable domain is established, and abnormal monitoring data is identified by combining blasting response constraints. Using the tunnel monitoring area as nodes and multi-source monitoring features as node features, a dynamic spatiotemporal map is constructed, and the association weights between nodes are adaptively updated based on the monitoring results of the current blasting cycle. Based on the dynamic spatiotemporal graph, neighborhood information is aggregated for node features to obtain the blasting disturbance state corresponding to each surrounding rock area. Accumulated damage is extrapolated to obtain the current state of surrounding rock damage; stability indices of each monitoring area are calculated comprehensively to analyze the impact on the stability of the tunnel.

2. The method for analyzing the impact of hard rock blasting on the stability of tunnels as described in claim 1, characterized in that: The data registration includes, for continuous... The tunneling and blasting cycle is numbered, and the current cycle is the [number]th cycle. One loop; collect the blasting parameter set for each loop. To obtain information about the surface and deep layers of the surrounding rock in the tunnel. Multi-source heterogeneous monitoring data from the first sensor node; acquire the first... The detonation command trigger time of each cycle maps absolute system time to relative detonation time. Construct a unified time grid matrix and set the time step. For sampling intervals greater than The low-frequency stress and displacement data were upsampled using cubic spline interpolation and aligned to a unified time grid matrix.

3. The method for analyzing the impact of hard rock blasting on the stability of tunnels as described in claim 2, characterized in that: The construction of the multi-source spatiotemporal dataset for blasted surrounding rock includes, for sampling intervals less than... High-frequency vibration data, within a time step Peak features are obtained by performing max pooling within the sequence; for discrete microseismic and acoustic emission event sequences, the features are accumulated and distributed to the time of event occurrence. Within the time interval; the length of the generated time dimension is Synchronous multi-source timing matrix; During the tunneling process, the working face dynamically moves forward; a global coordinate system for the tunnel is established. Let the first The equivalent global coordinates of the center of the loop are: , No. The global coordinates of each sensor node are ; Transform the global coordinate system translation to the current equivalent center of the explosion. A relative local coordinate system with the origin; for each node Calculate the spatial topological feature vector, including absolute blast center distance, spatial azimuth, and structural surface intercept; The synchronized multi-source time series matrix after time synchronization is fused with the spatially registered topological feature vector; for continuous... Each iteration constructs a spatiotemporal dataset represented by a multidimensional tensor. .

4. The method for analyzing the impact of hard rock blasting on the stability of tunnels as described in claim 3, characterized in that: The multidimensional credibility evaluation includes, for the first One blasting cycle, for any monitoring node The system calculates the state health score from four dimensions: data continuity score, physical consistency score, spatial correlation score, and historical drift score, and then performs a linear weighting to obtain the dynamic reliability of the node. ; Before extraction Using historical high-confidence monitoring data from the previous cycle, a nonlinear regression fit was performed to calculate the... Second blasting node Theoretical predicted mean response ; Calculate the probability density function of historical prediction residuals and set the confidence level; Construct nodes Explosive response allowable range ; Perform initial screening to identify nodes In the Substituting the actual monitored peak value into the loop, if the actual monitored peak value does not fall within the allowable range of the blast response... If so, the anomaly detection mechanism will be triggered; When node When an exception is triggered, retrieve the node. The first-order candidate neighborhood node set in the spatial topology Participate in joint discrimination; calculate the blasting stress wave and elastic wave generated by potential surrounding rock fracture from the nodes based on the longitudinal and transverse wave velocities of the rock mass. Propagate to neighboring nodes Theoretical time delay Perform state coordination determination; If node An abnormal pulse is generated. Within the time window, the set of candidate neighborhood nodes If all node monitoring values ​​are within the permissible range for blasting response and no coordinated response occurs, then the node is determined to be... The current data indicates a sensor malfunction. If node The monitored data exceeded the allowed range, and there was a corresponding time delay. Within the window, the set of candidate neighboring nodes If it also exhibits homologous abnormal fluctuations that exceed the allowable range, it is determined to be a real anomaly in the surrounding rock; Data identified as sensing anomalies is masked based on a set of candidate neighboring nodes. The data is spatially reconstructed and replaced using graph neural networks; for data determined to be real anomalies in the surrounding rock, high attention feature weights are assigned to nodes and affected neighborhood edges when constructing the graph network.

5. The method for analyzing the impact of hard rock blasting on the stability of tunnels as described in claim 4, characterized in that: The adaptive update of the association weights between nodes based on the monitoring results of the current blasting cycle includes, for the first... One tunneling and blasting cycle, sensors are deployed in the surrounding rock of the tunnel. Each monitoring area is defined as a set of nodes in the graph; Extracting multi-source spatiotemporal datasets and dividing nodes Vibration characteristics, stress and displacement state quantities, and calculated dynamic reliability The layers are stitched together, with dimension alignment performed using a fully connected layer, and initialized to the [number]th [dimension]. The node feature vector of each cycle This constitutes the node feature matrix of the entire graph; Without considering the real-time state of the nodes, the nodes are calculated based on the laws of rock mechanics. With nodes The initial physical edge weights between them are obtained by multiplying the physical decay terms together; To prevent errors from propagating in a graph network, nodes are used. and nodes In the Dynamic reliability of loops and The physical prior edge weights are corrected using a soft mask to obtain nodes with corrected credibility. and nodes The edge weights are used to construct the corrected physical adjacency matrix; An adaptive graph attention mechanism (GAT) is introduced, which utilizes the actual monitoring data features of the current loop to perform secondary optimization of weights; for nodes... and nodes eigenvectors and Perform a linear mapping and calculate the attention coefficient; Using the physical adjacency matrix as a priori mask for the attention distribution, and normalizing it using the Softmax function, we obtain the adaptively updated dynamic association weight matrix. .

6. The method for analyzing the impact of hard rock blasting on the stability of tunnels as described in claim 5, characterized in that: Obtaining the current surrounding rock damage state includes, based on the dynamic correlation weight matrix. , for the The node features in each burst loop are subjected to spatial dimension convolutional aggregation; for any monitoring node Using neighborhood node set The features are weighted and transmitted, and after multi-layer graph convolution operations, the hidden feature vectors of the fused spatial context of the region are obtained. ; The aggregated hidden feature vector The input is fed into a fully connected feedforward neural network for nonlinear mapping, dimensionality reduction, and decoding into the current blasting perturbation state vector. From the state of explosive disturbance Extract the incremental damage to the surrounding rock directly caused by the current single blast. ; Get the The absolute physical time interval between the next blast and the previous blasting cycle Construct a nonlinear memory decay function with physical time constraints. ; Set nodes In the The cyclic comprehensive surrounding rock damage state vector is Apply the time recovery decay function directly to the historical state to calculate the experience. Residual historical damage after the interval; The residual historical damage and the surrounding rock damage increment are fused to obtain the updated state. Current surrounding rock damage state after the end of each cycle .

7. The method for analyzing the impact of hard rock blasting on the stability of tunnels as described in claim 6, characterized in that: The comprehensive calculation of stability indicators for each monitoring area includes, for any monitoring area node within the roadway. Extract the first The feature vectors obtained from iterative calculations are mapped to four-dimensional physical quantification indicators, including damage indexes. Disturbance index Deformation rate index Abnormal frequency index ; The comprehensive stability index is calculated using a variable weight comprehensive evaluation model. The initial constant weights of the four dimensions are preset; Calculate the monitoring area Comprehensive stability index Clustering was used to determine three critical thresholds. Based on the calculation The monitoring areas of the tunnel were divided into four risk zones: like This indicates a stable region; single-blast disturbance degree Within the range of material elastic recovery, the surrounding rock of the tunnel is in a safe state; like and This area is the dominant blasting sensitive zone; the cumulative damage in the area is not large, but it is extremely sensitive to the vibration of a single blast, and the charge structure and micro-delay time need to be adjusted in the next cycle. like and and The dominant area is the area of ​​cumulative damage; The rock mass has suffered fatigue damage and creep, requiring enhanced support. like If the area is in a potentially unstable zone, the structural surface is about to be breached, posing a high risk. Tunneling must be stopped and emergency reinforcement carried out.

8. An analysis system for the impact of hard rock blasting on the stability of tunneling roadways, based on the analysis method for the impact of hard rock blasting on the stability of tunneling roadways as described in any one of claims 1 to 7, characterized in that: This includes a multi-source data construction module, which performs time synchronization and spatial registration of data to construct a multi-source spatiotemporal dataset for blasting surrounding rock; The data reliability analysis module, connected to the multi-source data construction module, is used to determine the dynamic reliability of each monitoring node; the blasting disturbance analysis module is used to construct a dynamic spatiotemporal correlation model based on the spatial relationship between monitoring nodes, blasting wave propagation characteristics, geological connection relationship and dynamic reliability; the stability analysis module, by comprehensively considering the cumulative damage state, blasting disturbance state, surrounding rock deformation characteristics and disturbance propagation characteristics, determines the stability state of different areas of the tunnel.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the method for analyzing the impact of hard rock blasting on the stability of tunneling roadways as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the method for analyzing the impact of hard rock blasting on the stability of tunnels as described in any one of claims 1 to 7.

Images