A construction risk assessment method and system for tunneling through a fault fracture zone

Abstract

The application provides a tunnel crossing fault fracture zone construction risk assessment method and system, and belongs to the technical field of safety monitoring of tunnels and underground engineering. The method comprises the following steps: collecting multi-source data in the construction process in real time; performing multi-modal feature fusion processing on the multi-source data to extract a dynamic feature set; constructing a double-channel interactive risk assessment model based on deep reinforcement learning; inputting the dynamic feature set into a physical mechanism channel and a data-driven channel with a bidirectional information interaction mechanism; calculating risk indexes through multi-level risk assessment modules; finally, fusing to obtain a comprehensive risk index; performing risk early warning based on the comprehensive risk index; performing risk traceability analysis based on the output of the risk assessment model; and generating regulation and control decisions based on the risk prediction results. Through multi-source data fusion and double-channel intelligent assessment, the application realizes real-time and accurate identification, accurate traceability and active regulation and control of construction risks, and improves the construction risk control capability of tunnels crossing fault fracture zones.

Description

Technical Field

[0001] This invention relates to the field of tunnel and underground engineering safety monitoring technology, and in particular to a construction risk assessment method and system for tunnels crossing fault fracture zones. Background Technology

[0002] In tunnel engineering, especially in the development of mountain tunnels, traversing fault fracture zones is one of the main sources of risk during construction. Fault fracture zones are typically composed of broken and loose rock masses, with complex geological conditions, variable hydrogeological environments, and significant uncertainties and heterogeneity. Using drill-and-blast methods in such adverse geological sections can easily trigger major engineering disasters such as large deformations of the surrounding rock, collapses, mudslides, water inrushes, and failure of initial support, seriously threatening the safety of construction personnel, project progress and quality, and even causing huge economic losses and social impacts.

[0003] Traditional tunnel construction risk assessment methods mainly rely on the following two approaches: (1) Method based on expert experience and static geological survey: This method relies heavily on preliminary geological survey reports and the personal experience of engineers. However, the accuracy of preliminary surveys is limited, making it difficult to accurately reveal the complex structure and hydrological characteristics within fault fracture zones. At the same time, this method has a significant lag, failing to identify and warn of dynamic risks during drill-and-blast construction in real time. It is a passive management model that responds after the fact, which is difficult to meet the requirements of safe and efficient tunnel construction in modern times.

[0004] (2) Assessment methods based on a single model or simple data fusion: With the development of monitoring technology, some studies have begun to attempt to use real-time data during construction for risk assessment. For example, early warning thresholds are established using drilling and blasting construction parameters (such as blasting parameters and support parameters) or single surrounding rock deformation monitoring data. However, these methods have obvious limitations: Insufficient data utilization: Usually only a single type or a few types of data are focused on, and the geological rock mass parameters (such as rock strength and integrity), parameters of the entire drilling and blasting construction process (such as blasting effect, cycle advance, and support design), surrounding rock conditions (microseismic, ground-penetrating radar), multi-dimensional deformation response (taking into account both cumulative value and rate of change), and hydrological environment, etc., are not deeply integrated. This results in a one-sided perspective on risk assessment, missing information dimensions, and an inability to fully capture the complex risk coupling relationships within the fault fracture zone.

[0005] The existing models are mostly based on a single mechanism: they are either purely data-driven or purely physical mechanism models. Purely data-driven models (such as simple machine learning models) lack consideration for the physical laws of tunnel engineering, such as rock mechanics, blasting dynamics, and support structure mechanics. They have poor generalization ability and weak interpretability when data is scarce or when encountering unseen working conditions. On the other hand, purely physical mechanism models are difficult to handle the high nonlinearity of fault fracture zones and the strong disturbances brought about by drill-and-blast construction, and they cannot effectively learn complex hidden patterns from massive dynamic data.

[0006] Lack of foresight and decision support: Most methods focus on assessing the current state of risk, lacking the ability to accurately predict future risk trends. Furthermore, after identifying a risk, they typically only provide a warning of its existence, without pinpointing the root cause (e.g., whether it's due to excessive blasting explosives, insufficient support strength, or sudden changes in hydrological conditions), thus making it difficult to generate targeted control decisions.

[0007] Therefore, for the high-risk scenario of drilling and blasting tunnels traversing fault fracture zones, there is an urgent need for a new solution that can achieve deep fusion of multi-source information, synergy between mechanism and data-driven approaches, and precise tracing and intelligent decision-making to address the above problems. Summary of the Invention

[0008] The purpose of this invention is to provide a construction risk assessment method and system for tunnels crossing fault fracture zones, which can overcome the shortcomings of existing technologies and achieve real-time, accurate identification, precise source tracing and proactive control of construction risks through multi-source data fusion and dual-channel intelligent assessment.

[0009] To achieve the above objectives, the present invention provides a method for assessing construction risks when tunnels pass through fault fracture zones, comprising the following steps: Step S1: Collect multi-source data in real time during the construction process; Step S2: Perform multimodal feature fusion processing on multi-source data to extract dynamic feature sets; Step S3: Input the dynamic feature set into the risk assessment model to obtain the comprehensive risk index; Step S4: Conduct risk warning based on the comprehensive risk index and generate a comprehensive warning level; Step S5: Based on the comprehensive early warning level and comprehensive risk index, initiate and execute risk tracing analysis to identify key risk factors and key risk transmission paths; Step S6: Generate the optimal control decision.

[0010] Preferably, in step S1, the multi-source data includes: Drill and blast method construction parameters: Geological and rock mass parameters: Groundwater development level Uniaxial compressive strength of rock Rock mass integrity coefficient ; Construction method parameters: Construction method code (Used to identify full-section method, two-step method, three-step method, three-step method with temporary invert arch, etc.); Support parameters: Anchor bolt length Spacing Steel frame model Spacing Shotcrete thickness ; Explosive effect parameters: blast hole utilization rate Circular advance Disturbance blasting coefficient ; Surrounding rock condition sensing data: amplitude of reflected waves from ground-penetrating radar TSP wave velocity anomaly coefficient ; Surrounding rock deformation response data: Absolute value of deformation: Cumulative value of arch settlement Horizontal convergence cumulative value ; Deformation rate of change: rate of settlement of the vault Horizontal convergence rate ; Microseismic activity data: energy release rate of microseismic events Magnitude and frequency value; Environmental data: Unit flow rate pore water pressure .

[0011] Preferably, in step S2, multimodal feature fusion processing is performed on the multi-source data to extract a dynamic feature set, including the following steps: Step S21: Data preprocessing and noise reduction, using wavelet packet transform to decompose and reconstruct the original signal; Step S22: Spatiotemporal registration and alignment to establish a unified spatiotemporal coordinate system; Step S23: Multimodal feature extraction. Based on the multi-source data collected in step S1, extract features including time domain features, frequency domain features, and statistical features. Step S24: Construct hierarchical dynamic feature vectors : ; in, This represents the surrounding rock stability index. Indicates the weight of the rock mass; Indicates the energy gradient of microseismic events; Represents the deformation comprehensive coefficient. This indicates the cumulative allowable value of the arch settlement. Indicates the allowable rate of settlement of the arch. This represents the cumulative allowable value for horizontal convergence. Indicates the horizontal convergence allowable rate; This indicates the blasting disturbance index. Indicates the amplitude of the reflected wave from the ground-penetrating radar. The dominant frequency obtained by Fourier transform Indicates the frequency of magnitude value, Indicates the support strength index, This indicates transpose.

[0012] Preferably, in step S3, the dynamic feature set is input into the risk assessment model to obtain the comprehensive risk index, including the following steps: Step S31, Physical Mechanism Channel: Based on the principles of damage mechanics and rock mechanics, utilizing dynamic feature vectors. The features in the calculation result in three risk indices: First Risk Index : ; in, Indicates the first risk index The scaling factor, The cumulative damage factor is represented by the following formula: ; in, Indicates the damage accumulation coefficient. Representing historical moments The energy release rate of microseismic events Representing historical moments The comprehensive coefficient of surrounding rock deformation, Indicates the decay coefficient of damaged memory. Indicates the current evaluation time; Second Risk Index : ; in, , , , Weighting coefficients representing engineering geological and hydrogeological risks; Third Risk Index : ; in, , , The weighting coefficient represents the risk of construction disturbance. Indicates the stability index of the surrounding rock; Step S32: Data-driven channel, utilizing dynamic feature vectors And its rate of change, using a deep neural network to calculate three risk indices: First Risk Index : ; in, This represents the first deep neural network. This indicates its weight parameters; Second Risk Index : ; in, This represents the second deep neural network. This represents its weight parameters. The first derivative of a dynamic eigenvector is represented. The second derivative of a dynamic eigenvector is represented. Third Risk Index : ; in, This represents the third deep neural network. This indicates its weight parameters; Step S33: The dual-channel interactive fusion module achieves deep fusion of the two channels through a cross-attention mechanism. Define the physical mechanism channel risk vector ; Define data-driven channel risk vector ; The output of cross-attention is then: ; in, , , These represent the query vector, key vector, and value vector of the physical mechanism channel, respectively, and are composed of vectors. Obtained through linear transformation, , , These represent the query vector, key vector, and value vector of the data-driven channel, respectively, obtained from vector D through a linear transformation. The scaling factor representing the dimension of the vector; Step S34: The multi-index fusion module fuses the three risk indices from the dual channels respectively. ; ; ; in, This represents the structural stability risk index after fusion. This indicates the integrated hydrogeological risk index. This indicates the construction disturbance risk index after integration. This indicates the combined weight of the risk indices for each channel; Step S35, Calculation of Comprehensive Risk Index: ; in, This represents the final overall risk index. , , This represents the weighting coefficient of each integration risk index.

[0013] Preferably, the fusion weight coefficient in the multi-index fusion module Adaptive adjustment is achieved through the following mechanisms: Calculate the confidence level of each risk index. : ; in, This represents the confidence scaling factor. Indicates the first Risk index under the first The data quality index of each channel is determined by the signal-to-noise ratio and completeness of the corresponding input data; Weights are adjusted based on confidence level and environmental complexity. : ; in, Indicates the first Risk index under the first Confidence level of each channel This represents the scaling factor for environmental complexity. The environmental complexity index is calculated using the following formula: ; in, , Indicates the weighting coefficient; This represents the reference value for a unit flow rate.

[0014] Preferably, in step S4, risk warning is generated based on the comprehensive risk index, and a comprehensive warning level is generated, including the following steps: Step S41: Set multi-level composite early warning thresholds to form a comprehensive risk index. and various integration risk indices , , Threshold ranges are set for four levels: safe, low risk, medium risk, and high risk. Step S42: Parallel evaluation of individual risk levels, applying the thresholds set in step S41 to the real-time calculated levels. , , , To determine an initial warning level for each index. , , , ; Step S43: Based on the individual risk levels obtained in step S42, perform a fusion decision to calculate the final comprehensive early warning level. : ; in, This represents the mode function, used to determine the most common warning level.

[0015] Preferably, in step S5, based on the comprehensive early warning level and comprehensive risk index, risk tracing analysis is initiated and executed to identify key risk transmission paths, including the following steps: Step S51, when the comprehensive early warning level When the risk level is reached or exceeded, risk tracing analysis is initiated. Step S52: Calculate the dynamic feature set Each feature Comprehensive risk index Contribution Contribution The formula, calculated using the gradient backpropagation method, is as follows: ; in, Indicates the overall risk index Features The partial derivatives; Step S53: Construct a dynamic risk tracing map and identify key risk transmission paths.

[0016] Preferably, in step S53, constructing a dynamic risk source map and identifying key risk transmission paths includes the following steps: Step S531: Construct the node set of the dynamic risk tracing map. and edge set ,in: Node set It includes three types of nodes: construction parameter nodes, geological response nodes, and risk index nodes; Edge set Representing causal relationships between nodes, edge weights Determined in the following ways: If node To the node If there is a direct influence relationship, then the weight... ,in For nodes The contribution of the corresponding feature; If node Influence nodes through multiple intermediate nodes Then the path weight is the product of the weights of each edge on the path; Step S532, based on contribution threshold Prune the graph to retain influential paths with significant weights: Step S533: Identify key risk transmission paths in the pruned graph: Define path importance score : ; in, Indicates the path length. This represents the path length penalty coefficient. Indicates the risk transmission path; Identifying the preceding steps from construction parameter nodes to comprehensive risk index nodes The highest-scoring path serves as a key risk transmission path; Step S534: Output the visualization map of the critical path and the corresponding path importance score.

[0017] Preferably, in step S6, generating the optimal control decision includes the following steps: Step S61: Construct a knowledge base for regulatory strategies, which stores regulatory measures. The characteristics of the risks it can target, the types of risks, and the expected risk mitigation effect. The mapping relationship between them; Step S62: Using the key risk transmission paths and high-contribution characteristics identified by the risk tracing analysis as input, match and filter a set of candidate control measures that can directly affect these paths and characteristics from the strategy knowledge base. ; Step S63: Perform multi-objective optimization decision-making in the set of candidate control measures. Select the optimal control combination Its objective function is: ; in, Indicates the implementation of control combination The subsequent predicted comprehensive risk index Indicates the implementation of control combination Total cost A coefficient representing the trade-off between risk and cost; Step S64: Output decision instructions.

[0018] This invention also provides a construction risk assessment system for tunnels crossing fault fracture zones, comprising: The data acquisition module is used to collect multi-source data in real time during the construction process; The feature extraction module is used to perform multimodal feature fusion processing on multi-source data and extract dynamic feature sets; The risk assessment module includes a dual-channel interactive processing unit, which is used to input dynamic feature sets into the physical mechanism channel and the data-driven channel with a two-way information interaction mechanism, calculate multiple risk indices through multi-level risk assessment, and finally merge them to obtain a comprehensive risk index. The early warning module is used to provide risk warnings based on a comprehensive risk index and generate a comprehensive early warning level. The source tracing analysis module is used to initiate and execute risk source tracing analysis based on the output of the comprehensive early warning level and risk assessment model, and to identify risk factors and key risk transmission paths. The decision-making module is used to generate optimal control decisions based on the comprehensive early warning level and risk source analysis results; The system's modules are connected sequentially to form a closed-loop processing flow from data collection to risk decision-making.

[0019] Therefore, the present invention employs the above-mentioned construction risk assessment method and system for tunnels crossing fault fracture zones, and the beneficial technical effects are as follows: (1) This invention, by constructing a dynamic feature set that deeply integrates the characteristics of drill-and-blast construction, systematically integrates and performs feature engineering on multi-source heterogeneous data such as geological rock mass parameters, parameters of the entire drill-and-blast construction process, surrounding rock conditions, multi-dimensional deformation response, and hydrological environment for the first time. This overcomes the shortcomings of traditional methods, which have a single information dimension and cannot reflect the impact of construction technology in the drill-and-blast scenario. Furthermore, by constructing a dual-channel interactive model with a physical mechanism channel and a data-driven channel, the invention combines the physical laws of rock mechanics, damage mechanics, and blasting dynamics with the ability of the data-driven model to mine complex hidden patterns from massive dynamic data, achieving complementary advantages. The physical mechanism channel ensures the interpretability and physical consistency of the evaluation results, while the data-driven channel enhances the model's ability to fit complex nonlinear relationships. This collaborative evaluation mechanism of physical mechanism and big data in drill-and-blast construction risk assessment improves the accuracy, robustness, and interpretability of risk assessment under complex geological conditions such as fault fracture zones, and solves the problems of poor adaptability of pure mechanism models in the strong disturbance environment of drill-and-blast and weak generalization ability and insufficient decision support of pure data models.

[0020] (2) This invention establishes a complete intelligent closed-loop system for assessment, early warning, source tracing, and decision-making in drill-and-blast construction. The system can not only perceive risk levels in real time through a multi-level composite early warning mechanism, but also accurately locate key construction links (such as blasting and support) or geological factors leading to risks through gradient backpropagation and dynamic risk source tracing maps, revealing the complete transmission path of risk from construction parameters to geological response and then to the risk index. Based on this, multi-objective optimization is performed using a constructed control strategy knowledge base, generating optimal decision-making schemes that directly address the root causes of risks and balance risk suppression effects with control costs (such as adjusting blasting parameters, optimizing support forms, and strengthening drainage). This closed-loop system transforms risk management in drill-and-blast construction from a passive, experience-based post-event response to an active, data-driven pre-event prediction and in-event control, greatly enhancing the safety management capabilities and scientific decision-making in complex geological conditions, and providing core technical support for the safe, efficient, and intelligent construction of tunnels. Attached Figure Description

[0021] Figure 1 This is a flowchart of a construction risk assessment method for tunnels crossing fault fracture zones according to the present invention; Figure 2 This is a diagram of the risk assessment model structure. Figure 3 This is an architecture diagram of a construction risk assessment system for tunnels crossing fault fracture zones according to the present invention. Detailed Implementation

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0024] Example 1 like Figure 1 As shown, a construction risk assessment method for tunnels crossing fault fracture zones includes the following steps: Step S1: Collect multi-source data in real time during the construction process.

[0025] (1) Data collection of drilling and blasting method construction parameters: Geological and rock mass parameter acquisition: The uniaxial compressive strength of the rock was obtained through geological drilling and soil testing. and rock mass integrity coefficient ; Based on the geological survey report and the seepage situation inside the cave, the groundwater development level was comprehensively assessed. (For example, it can be divided into dry, wet, water-sprayed, and water-gushing grades).

[0026] Construction method parameter record: Record the code of the construction method currently used, based on the construction organization design documents. (For example, 01 represents the full-section method, 02 represents the two-step method, 03 represents the three-step method, and 04 represents the three-step method with temporary invert arches).

[0027] Support parameter recording and measurement: Record the anchor bolt length according to the construction drawings and on-site inspection. Spacing ; Record steel frame model (e.g., I18, H175, etc.) and spacing ; The thickness of shotcrete is measured using probe or non-destructive testing methods. .

[0028] Explosion effect parameter collection and calculation: The borehole utilization rate for each blasting cycle was obtained through on-site measurement and recording. (Cyclic advance / hole depth) and cyclic advance ; Calculate the disturbance blasting coefficient : ; ; in, This indicates the amount of explosive consumed per unit volume in real time. This represents a reference consumption amount, determined based on the geological and rock mass parameters of the current construction section. This represents an empirical coefficient related to the construction method and type of explosive. This indicates the rock quality index (which can be calculated from the rock mass integrity coefficient).

[0029] (2) Acquisition of surrounding rock condition sensing data: A ground-penetrating radar system (center frequency 100MHz) was used to detect along the tunnel face and tunnel walls, and the amplitude of reflected waves was collected. ; By deploying seismic sources and sensors on the tunnel sidewall using the TSP303 seismic wave detection system, wave velocity data was collected, and the TSP wave velocity anomaly coefficient was calculated. .

[0030] (3) Acquisition of surrounding rock deformation response data: Deformation absolute value acquisition: A total station was installed at the crown of the arch to collect the cumulative settlement value of the crown. ; Install convergence meters or total stations at measurement points on the tunnel sidewalls to collect the cumulative horizontal convergence values. .

[0031] Deformation rate calculation: Based on time series data of cumulative crown settlement, the crown settlement rate is calculated using differential methods. ; Horizontal convergence rate is calculated by differencing time series data based on the cumulative value of horizontal convergence. .

[0032] Microseismic activity data acquisition: A 16-channel microseismic monitoring system was installed in the rock mass surrounding the tunnel to monitor and calculate the energy release rate of microseismic events in real time. ; Based on microseismic monitoring data, the magnitude frequency is calculated statistically. value.

[0033] (4) Environmental data collection: Install flow meters in tunnel drainage ditches or at water inflow points to collect unit flow rates. .

[0034] A pore water pressure gauge was installed in the borehole to collect pore water pressure data. .

[0035] Step S2: Perform multimodal feature fusion processing on multi-source data to extract dynamic feature sets.

[0036] Step S21, Data Preprocessing and Noise Reduction: The wavelet packet transform is performed using the db4 wavelet basis and decomposed into three levels.

[0037] The signal-to-noise ratio threshold is set to 20dB. Signal components with a signal-to-noise ratio lower than this threshold are filtered.

[0038] Step S22, Spatiotemporal Registration and Alignment: Establish a coordinate system with the tunnel axis as the Z-axis, the vertical upward direction as the Y-axis, and the horizontal rightward direction as the X-axis.

[0039] A timestamp alignment algorithm is used to control the time synchronization accuracy within 0.1 seconds.

[0040] Step S23, Multimodal Feature Extraction: Time-domain characteristics: mean, variance, peak value, kurtosis, etc.

[0041] Frequency domain characteristics: power spectral density, spectral centroid, band energy ratio, etc.

[0042] Statistical characteristics: skewness, kurtosis, autocorrelation coefficient, etc.

[0043] Step S24: Construct hierarchical dynamic feature vectors : ; in, This represents the surrounding rock stability index. Indicates the weight of the rock mass; Indicates the energy gradient of microseismic events; Represents the deformation comprehensive coefficient. This indicates the cumulative allowable value of the arch settlement. Indicates the allowable rate of settlement of the arch. This represents the cumulative allowable value for horizontal convergence. Indicates the horizontal convergence allowable rate; This indicates the blasting disturbance index. Indicates the amplitude of the reflected wave from the ground-penetrating radar. The dominant frequency obtained by Fourier transform Indicates the frequency of magnitude value, Indicates the support strength index, This indicates transpose.

[0044] Step S3: Input the dynamic feature set into the risk assessment model (e.g., Figure 2 This yields a comprehensive risk index.

[0045] Step S31, Physical Mechanism Channel: Based on the principles of damage mechanics and rock mechanics, utilizing dynamic feature vectors. The characteristics in the calculation are used to determine three risk indices.

[0046] First Risk Index (Structural stability risk): ; in, This represents the scaling factor for the first risk index, which is set to 2.5 in this embodiment. The cumulative damage factor is represented by the following formula: ; in, This represents the damage accumulation coefficient, which is set to 0.15 in this embodiment. Representing historical moments The energy release rate of microseismic events Representing historical moments The comprehensive coefficient of surrounding rock deformation, This represents the damage memory decay coefficient, which is set to 0.08 in this embodiment. Indicates the current assessment time.

[0047] Second Risk Index (Hydrogeological risks): ; in, , , , The weighting coefficients representing engineering geological and hydrogeological risks are set to 0.4, 0.35, and 0.25, respectively.

[0048] Third Risk Index (Construction disturbance risk): ; in, , , The weighting coefficients representing the risk of construction disturbance are set to 0.3, 0.3, 0.2, and 0.2, respectively. This represents the stability index of the surrounding rock.

[0049] Step S32: Data-driven channel, utilizing dynamic feature vectors And its rate of change, using a deep neural network to calculate three risk indices: First Risk Index (Instantaneous risk): ; in, This represents its weight parameters. This represents the first deep neural network, which is a feedforward neural network containing a 14-node input layer, a 64-node hidden layer using the ReLU activation function, a dropout layer with a dropout rate of 0.2, a 32-node hidden layer using the ReLU activation function, and a 1-node output layer using the Sigmoid activation function. The network weights are obtained through training, and the initial values ​​are initialized using a He normal distribution. Training uses the Adam optimizer with a learning rate of 0.001, a batch size of 32, and mean squared error as the loss function.

[0050] Second Risk Index (Trend Risk): ; in, This represents its weight parameters. The first derivative of a dynamic eigenvector is represented. The second derivative of a dynamic eigenvector is represented. The second deep neural network is represented by a Long Short-Term Memory (LSTM) network. This network consists of a 28-node input layer, a 50-node LSTM layer, a 25-node fully connected layer using the Tanh activation function, and a 1-node output layer using the Sigmoid activation function. The network weights are obtained through training, initialized using a Glorot uniform distribution. Training employs the RMSprop optimizer with a learning rate of 0.0005, a sequence length of 10 time steps, and uses mean squared error as the loss function.

[0051] Third Risk Index (Unexpected risks): ; in, This represents its weight parameters. This represents the third deep neural network, which employs a feedforward neural network with an attention mechanism. The network consists of a 5-node input layer, a 32-node hidden layer using ReLU activation, an attention mechanism layer, a 16-node hidden layer using ReLU activation, and a 1-node output layer using Sigmoid activation. The network weights are obtained through training, initialized using a He normal distribution. Training employs the Adam optimizer with a learning rate of 0.002, a batch size of 16, and uses binary cross-entropy as the loss function.

[0052] Step S33: The dual-channel interactive fusion module achieves deep fusion of the two channels through a cross-attention mechanism. Define the physical mechanism channel risk vector ; Define data-driven channel risk vector ; The output of cross-attention is then: ; in, , , These represent the query vector, key vector, and value vector of the physical mechanism channel, respectively, and are composed of vectors. Obtained through linear transformation, , , These represent the query vector, key vector, and value vector of the data-driven channel, respectively, obtained from vector D through a linear transformation. This represents the dimension scaling factor of the vector.

[0053] Step S34: The multi-index fusion module fuses the three risk indices from the dual channels respectively. ; ; ; in, This represents the structural stability risk index after fusion. This indicates the integrated hydrogeological risk index. This indicates the construction disturbance risk index after integration. This indicates the combined weight of the risk indices for each channel. It is 0.6. It is 0.4. It is 0.55. It is 0.45. It is 0.5. It is 0.5.

[0054] Fusion weight coefficients in the multi-index fusion module Adaptive adjustment is achieved through the following mechanisms: Calculate the confidence level of each risk index. : ; in, This represents the confidence scaling factor. Indicates the first Risk index under the first The data quality metrics for each channel are determined by the signal-to-noise ratio and completeness of the corresponding input data.

[0055] Weights are adjusted based on confidence level and environmental complexity. : ; in, Indicates the first Risk index under the first Confidence level of each channel This represents the scaling factor for environmental complexity. The environmental complexity index is calculated using the following formula: ; in, , Indicates the weighting coefficient; This represents the reference value for a unit flow rate.

[0056] Step S35, Calculation of Comprehensive Risk Index: ; in, This represents the final overall risk index. , , The weighting coefficients for each integration risk index are 0.4, 0.3, and 0.3, respectively.

[0057] This step achieves deep integration and precise quantification of multi-dimensional risks in tunnel construction by constructing and running a risk assessment model that interacts with both physical mechanism and data-driven approaches. The physical mechanism approach, based on damage mechanics and rock mechanics principles, ensures the physical clarity and interpretability of the risk assessment; the data-driven approach, on the other hand, uses deep neural networks to mine complex nonlinear risk patterns from massive amounts of dynamic data. The dual-channel approach achieves bidirectional deep interaction and adaptive fusion through a cross-attention mechanism, effectively compensating for the shortcomings of pure mechanism models in complex geological conditions and pure data models in terms of weak generalization ability and poor interpretability. The resulting comprehensive risk index comprehensively and dynamically reflects the overall risk status of construction, improving the accuracy, robustness, and reliability of risk assessment and providing core quantitative basis for subsequent risk warning, precise source tracing, and intelligent decision-making.

[0058] Step S4: Conduct risk warning based on the comprehensive risk index and generate a comprehensive warning level.

[0059] Step S41: Set multi-level composite early warning thresholds to form a comprehensive risk index. and various integration risk indices , , Threshold ranges are set for four levels: safe, low risk, medium risk, and high risk.

[0060] The thresholds for the safety level are: a comprehensive risk index below 0.3; the threshold for the attention level is: a comprehensive risk index of 0.3 or higher but not exceeding 0.6; the threshold for the warning level is: a comprehensive risk index of 0.6 or higher but not exceeding 0.8; and the threshold for the danger level is: a comprehensive risk index of 0.8 or higher. All convergence risk indices are classified using the same threshold criteria.

[0061] Step S42: Parallel evaluation of individual risk levels, applying the thresholds set in step S41 to the real-time calculated levels. , , , To determine an initial warning level for each index. , , , .

[0062] Step S43: Based on the individual risk levels obtained in step S42, perform a fusion decision to calculate the final comprehensive early warning level. : ; in, The mode function is used to determine the most common warning level; when multiple modes exist, the warning level with the highest value is taken as the final comprehensive warning level. The comprehensive warning levels, from low to high, correspond to safe, low risk, medium risk, and high risk, respectively.

[0063] The specific meanings of the comprehensive early warning levels and the recommended construction response measures are as follows: Safety( =0): This indicates extremely low risk, stable surrounding rock conditions, and construction can proceed as planned, maintaining the usual monitoring frequency.

[0064] Low risk ( =1): This indicates the existence of potential risk factors, but the surrounding rock condition is basically controllable. Construction can proceed normally, but the monitoring frequency of key parameters (such as deformation rate and microseismic activity) needs to be increased. No special engineering measures are required.

[0065] Medium risk ( =2): This indicates a significant risk and an unfavorable trend in the surrounding rock. Immediate geological exploration and deformation monitoring of the tunnel face are necessary to analyze the root causes of the risk. Based on the risk tracing results, structural reinforcement (such as adding anchor bolts or thickening the shotcrete layer), optimization of support parameters, or adjustment of the blasting plan should be implemented as needed to control the risk's development.

[0066] High risk ( =3): This indicates a very high risk, with an imminent danger of instability or disaster. Construction should be stopped immediately, personnel evacuated from the danger zone, and comprehensive measures such as structural reinforcement, advanced support, and enhanced drainage should be implemented urgently based on the critical path identified by the risk source analysis. Construction may only resume after the risk level has been reduced to "medium risk" or below and assessed.

[0067] This step achieves refined hierarchical management of construction risks by establishing a multi-level composite early warning mechanism. The system sets threshold ranges for four levels—safe, low-risk, medium-risk, and high-risk—for both the comprehensive risk index and each specific risk index, ensuring comprehensive risk identification through parallel evaluation. A fusion decision algorithm is used to integrate the early warning levels of each index, effectively avoiding misjudgments based on a single indicator and improving the reliability of early warnings. This early warning mechanism can perceive changes in risk status in real time, promptly issue tiered early warning signals, and provide differentiated response strategies for different risk levels, ensuring that risk control measures match the risk level and improving the timeliness and effectiveness of risk response.

[0068] Step S5: Based on the comprehensive early warning level and comprehensive risk index, initiate and execute risk tracing analysis to identify key risk transmission paths.

[0069] Step S51, when the comprehensive early warning level When the risk level is reached or exceeded, risk source analysis is initiated.

[0070] Step S52: Calculate the dynamic feature set Each feature Comprehensive risk index Contribution Contribution The formula, calculated using the gradient backpropagation method, is as follows: ; in, Indicates the overall risk index Features The partial derivatives of .

[0071] Step S53: Construct a dynamic risk tracing map and identify key risk transmission paths.

[0072] Step S531: Construct the node set of the dynamic risk tracing map. and edge set ,in: Node set It includes three types of nodes: construction parameter nodes, geological response nodes, and risk index nodes.

[0073] Construction parameter nodes represent controllable construction inputs and activities, and are the origin of risk transmission. They specifically include drilling and blasting design parameters (such as explosive consumption per unit volume of rock and cycle advance), support design parameters (such as anchor length and spacing, and steel frame type), and initial geological conditions (such as rock strength and integrity coefficient). Geological response nodes represent the dynamic changes in rock mass and environment caused by construction activities, and are the intermediate effects and manifestations of risk transmission. They specifically include surrounding rock condition indicators (such as ground-penetrating radar reflected wave amplitude and TSP wave velocity anomaly coefficient), deformation response indicators (such as crown settlement value and rate), microseismic activity indicators (such as microseismic energy release rate), and hydrological environment indicators (such as unit water inflow). Risk index nodes represent the final quantitative output of risk assessment, and are the result of risk transmission. They include various specific risk indices (such as structural stability and hydrogeological risk index) and the comprehensive risk index.

[0074] Edge set Representing causal relationships between nodes, edge weights It is determined in the following ways.

[0075] If node To the node If there is a direct influence relationship, then the weight... ,in For nodes The contribution of the corresponding feature.

[0076] If node Influence nodes through multiple intermediate nodes Then the path weight is the product of the weights of each edge on the path.

[0077] Step S532, based on contribution threshold Prune the graph to retain influential paths with significant weights: Retention conditions: ,in , The significance coefficient is represented, and its value ranges from [0.1, 0.3].

[0078] Significance coefficient According to environmental complexity index Dynamic adjustment: ; in, This represents the baseline significance coefficient, set to 0.2. This represents the environmental sensitivity coefficient, set to 1.5. This represents the hyperbolic tangent function.

[0079] Step S533: Identify key risk transmission paths in the pruned graph: Define path importance score : ; in, Indicates the path length. This represents the path length penalty factor, set to -0.1. It represents the risk transmission path, that is, the node connection sequence that starts from the construction parameter node, passes through the geological response node to the risk index node, and can reflect the risk generation, transmission and amplification process.

[0080] Identifying the preceding steps from construction parameter nodes to comprehensive risk index nodes The highest-scoring path serves as a key risk transmission path.

[0081] Step S534: Output a visual map of the critical path and the corresponding path importance score to provide a basis for risk control decisions.

[0082] This step, combining intelligent triggering mechanisms with in-depth analysis methods, achieves precise tracing from the surface symptoms of risk to its root causes. When the warning level reaches the attention level, the system automatically initiates source tracing analysis, calculating the contribution of each feature to the risk through gradient backpropagation to accurately identify key risk factors. The dynamic risk source tracing map constructed based on these contributions clearly displays the complete causal chain from construction activities to the surrounding rock and geological response, ultimately calculating and generating a risk index, revealing the intrinsic mechanism of risk generation and development. This source tracing method not only locates the current main risk sources but also identifies potential risk transmission chains, providing a clear decision-making direction for targeted regulation and improving the accuracy of risk governance.

[0083] Step S6: Generate the optimal control decision.

[0084] Step S61: Construct a knowledge base for regulatory strategies, which stores regulatory measures. The characteristics of the risks it can target, the types of risks, and the expected risk mitigation effect. The mapping relationship between them.

[0085] Step S62: Using the key risk transmission paths and high-contribution characteristics identified by the risk tracing analysis as input, match and filter a set of candidate control measures that can directly affect these paths and characteristics from the strategy knowledge base. .

[0086] Step S63: Perform multi-objective optimization decision-making in the set of candidate control measures. Select the optimal control combination Its objective function is: ; in, Indicates the implementation of control combination The subsequent predicted comprehensive risk index is calculated using the following formula: ; in, This represents the overall risk index at the current moment. Indicating regulatory measures The execution effect coefficient (with a value range of [0, 1], reflecting the impact of the difference between the actual working conditions and the benchmark working conditions on the effectiveness of the measures).

[0087] Indicates the implementation of control combination The total cost is calculated using the following formula: ; in, , , These represent the control measures. The corresponding time cost, resource consumption cost, and operational complexity cost (all normalized to the [0, 1] interval). , , These represent the weighting coefficients for the corresponding costs.

[0088] This represents the risk-cost trade-off coefficient, with a value range of [0, 1]. The closer to 1, the more the decision-making focuses on risk mitigation; the closer to 0, the more the decision-making focuses on cost control.

[0089] Step S64: Output decision instructions, which include the optimal control combination. The targeted execution targets (core nodes / features in the key risk transmission path) and the expected suppression effect on each risk index.

[0090] Example 2 like Figure 3 As shown, a construction risk assessment system for tunnels crossing fault fracture zones includes: The data acquisition module is used to collect multi-source data in real time during the construction process; The feature extraction module is used to perform multimodal feature fusion processing on multi-source data and extract dynamic feature sets; The risk assessment module includes a dual-channel interactive processing unit, which is used to input dynamic feature sets into the physical mechanism channel and the data-driven channel with a two-way information interaction mechanism, calculate multiple risk indices through multi-level risk assessment, and finally merge them to obtain a comprehensive risk index. The early warning module is used to provide risk warnings based on a comprehensive risk index and generate a comprehensive early warning level. The source tracing analysis module is used to initiate and execute risk source tracing analysis based on the output of the comprehensive early warning level and risk assessment model, and to identify key risk factors and key risk transmission paths. The decision-making module is used to generate optimal control decisions based on the comprehensive early warning level and risk source analysis results; The system's modules are connected sequentially to form a closed-loop processing flow from data collection to risk decision-making.

[0091] It is worth noting that all contents not described in detail in this invention are existing technologies and are well known to those skilled in the art.

[0092] Therefore, this invention employs the aforementioned construction risk assessment method and system for tunnels traversing fault fracture zones, achieving real-time and accurate assessment of construction risks in fault fracture zones. This method establishes a closed-loop management system encompassing dynamic perception, intelligent early warning, root cause analysis, and optimized control, overcoming the shortcomings of traditional methods such as one-sided assessment and delayed early warning, and significantly improving the safety management capabilities and intelligent decision-making level of tunnel construction.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. 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 still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for assessing construction risks when tunnels cross fault fracture zones, characterized in that, Includes the following steps: Step S1: Collect multi-source data in real time during the construction process; Step S2: Perform multimodal feature fusion processing on multi-source data to extract dynamic feature sets; Step S3: Input the dynamic feature set into the risk assessment model to obtain the comprehensive risk index; Step S4: Conduct risk warning based on the comprehensive risk index and generate a comprehensive warning level; Step S5: Based on the comprehensive early warning level and comprehensive risk index, initiate and execute risk tracing analysis to identify key risk factors and key risk transmission paths; Step S6: Generate the optimal control decision.

2. The construction risk assessment method for tunnels crossing fault fracture zones according to claim 1, characterized in that, In step S1, the multi-source data includes: Drill and blast method construction parameters: Geological and rock mass parameters: Groundwater development level uniaxial compressive strength of rock Rock mass integrity coefficient ; Construction method parameters: Construction method code ; Support parameters: Anchor bolt length Spacing Steel frame model Spacing Shotcrete thickness ; Explosive effect parameters: blast hole utilization rate Circular advance Disturbance blasting coefficient ; Surrounding rock condition sensing data: amplitude of reflected waves from ground-penetrating radar TSP wave velocity anomaly coefficient ; Surrounding rock deformation response data: Absolute value of deformation: Cumulative value of arch settlement Horizontal convergence cumulative value ; Deformation rate of change: rate of settlement of the vault Horizontal convergence rate ; Microseismic activity data: energy release rate of microseismic events Magnitude and frequency value; Environmental data: Unit flow rate pore water pressure .

3. The construction risk assessment method for tunnels crossing fault fracture zones according to claim 2, characterized in that, In step S2, multimodal feature fusion processing is performed on the multi-source data to extract a dynamic feature set, including the following steps: Step S21: Data preprocessing and noise reduction, using wavelet packet transform to decompose and reconstruct the original signal; Step S22: Spatiotemporal registration and alignment to establish a unified spatiotemporal coordinate system; Step S23: Multimodal feature extraction. Based on the multi-source data collected in step S1, extract features including time domain features, frequency domain features, and statistical features. Step S24: Construct hierarchical dynamic feature vectors : ； in, This represents the surrounding rock stability index. Indicates the weight of the rock mass; Indicates the energy gradient of microseismic events; Represents the deformation comprehensive coefficient. This indicates the cumulative allowable value of the arch settlement. Indicates the allowable rate of settlement of the arch. This represents the cumulative allowable value for horizontal convergence. Indicates the horizontal convergence allowable rate; This indicates the blasting disturbance index. Indicates the amplitude of the reflected wave from the ground-penetrating radar. The dominant frequency obtained by Fourier transform Indicates the frequency of magnitude value, Indicates the support strength index, This indicates transpose.

4. The construction risk assessment method for tunnels crossing fault fracture zones according to claim 3, characterized in that, In step S3, the dynamic feature set is input into the risk assessment model to obtain the comprehensive risk index, including the following steps: Step S31, Physical Mechanism Channel: Based on the principles of damage mechanics and rock mechanics, utilizing dynamic feature vectors. The features in the calculation result in three risk indices: First Risk Index : ； in, Indicates the first risk index The scaling factor, The cumulative damage factor is represented by the following formula: ； in, Indicates the damage accumulation coefficient. Representing historical moments The energy release rate of microseismic events Representing historical moments The comprehensive coefficient of surrounding rock deformation, Indicates the decay coefficient of damaged memory. Indicates the current evaluation time; Second Risk Index : ； in, , , , Weighting coefficients representing engineering geological and hydrogeological risks; Third Risk Index : ； in, , , The weighting coefficient represents the risk of construction disturbance. Indicates the stability index of the surrounding rock; Step S32: Data-driven channel, utilizing dynamic feature vectors And its rate of change, using a deep neural network to calculate three risk indices: First Risk Index : ； in, This represents the first deep neural network. This indicates its weight parameters; Second Risk Index : ； in, This represents the second deep neural network. This represents its weight parameters. The first derivative of a dynamic eigenvector is represented. The second derivative of a dynamic eigenvector is represented. Third Risk Index : ； in, This represents the third deep neural network. This indicates its weight parameters; Step S33: The dual-channel interactive fusion module achieves deep fusion of the two channels through a cross-attention mechanism. Define the physical mechanism channel risk vector ; Define data-driven channel risk vector ; The output of cross-attention is then: ； in, , , These represent the query vector, key vector, and value vector of the physical mechanism channel, respectively, and are composed of vectors. Obtained through linear transformation, , , These represent the query vector, key vector, and value vector of the data-driven channel, respectively, obtained from vector D through a linear transformation. The scaling factor representing the dimension of the vector; Step S34: The multi-index fusion module fuses the three risk indices from the dual channels respectively. ； ； ； in, This represents the structural stability risk index after fusion. This indicates the integrated hydrogeological risk index. This indicates the construction disturbance risk index after integration. This indicates the combined weight of the risk indices for each channel; Step S35, Calculation of Comprehensive Risk Index: ； in, This represents the final overall risk index. , , This represents the weighting coefficient of each integration risk index.

5. The construction risk assessment method for tunnels crossing fault fracture zones according to claim 4, characterized in that, Fusion weight coefficients in the multi-index fusion module Adaptive adjustment is achieved through the following mechanisms: Calculate the confidence level of each risk index. : ； in, This represents the confidence scaling factor. Indicates the first Risk index under the first The data quality index of each channel is determined by the signal-to-noise ratio and completeness of the corresponding input data; Weights are adjusted based on confidence level and environmental complexity. : ； in, Indicates the first Risk index under the first Confidence level of each channel This represents the scaling factor for environmental complexity. The environmental complexity index is calculated using the following formula: ； in, , Indicates the weighting coefficient; This represents the reference value for the unit flow rate.

6. The construction risk assessment method for tunnels crossing fault fracture zones according to claim 5, characterized in that, In step S4, risk warning is issued based on the comprehensive risk index, and a comprehensive warning level is generated, including the following steps: Step S41: Set multi-level composite early warning thresholds to form a comprehensive risk index. and various integration risk indices , , Threshold ranges are set for four levels: safe, low risk, medium risk, and high risk. Step S42: Parallel assessment of individual risk levels, applying the thresholds set in step S41 to the real-time calculated levels. , , , To determine an initial warning level for each index. , , , ; Step S43: Based on the individual risk levels obtained in step S42, perform a fusion decision to calculate the final comprehensive early warning level. : ； in, This represents the mode function, used to determine the most common warning level.

7. The construction risk assessment method for tunnels crossing fault fracture zones according to claim 6, characterized in that, In step S5, based on the comprehensive early warning level and comprehensive risk index, risk tracing analysis is initiated and executed to identify key risk transmission paths, including the following steps: Step S51, when the comprehensive early warning level When the risk level is reached or exceeded, risk tracing analysis is initiated. Step S52: Calculate the dynamic feature set Each feature Comprehensive risk index Contribution Contribution The formula, calculated using the gradient backpropagation method, is as follows: ； in, Indicates the overall risk index Features The partial derivatives; Step S53: Construct a dynamic risk tracing map and identify key risk transmission paths.

8. The construction risk assessment method for tunnels crossing fault fracture zones according to claim 7, characterized in that, In step S53, constructing a dynamic risk tracing map and identifying key risk transmission paths includes the following steps: Step S531: Construct the node set of the dynamic risk tracing map. and edge set ,in: Node set It includes three types of nodes: construction parameter nodes, geological response nodes, and risk index nodes; Edge set Representing causal relationships between nodes, edge weights Determined in the following ways: If node To the node If there is a direct influence relationship, then the weight... ,in For nodes The contribution of the corresponding feature; If node Influence nodes through multiple intermediate nodes Then the path weight is the product of the weights of each edge on the path; Step S532, based on contribution threshold Prune the graph to retain influential paths with significant weights: Step S533: Identify key risk transmission paths in the pruned graph: Define path importance score : ； in, Indicates the path length. This represents the path length penalty coefficient. Indicates the risk transmission path; Identifying the preceding steps from construction parameter nodes to comprehensive risk index nodes The highest-scoring path serves as a key risk transmission path; Step S534: Output the visualization map of the critical path and the corresponding path importance score.

9. The construction risk assessment method for tunnels crossing fault fracture zones according to claim 8, characterized in that, In step S6, the optimal control decision is generated, including the following steps: Step S61: Construct a knowledge base for regulatory strategies, which stores regulatory measures. The characteristics of the risks it can target, the types of risks, and the expected risk mitigation effect. The mapping relationship between them; Step S62: Using the key risk transmission paths and high-contribution characteristics identified by the risk tracing analysis as input, match and filter a set of candidate control measures that can directly affect these paths and characteristics from the strategy knowledge base. ; Step S63: Perform multi-objective optimization decision-making in the set of candidate control measures. Select the optimal control combination Its objective function is: ； in, Indicates the implementation of control combination The subsequent predicted comprehensive risk index Indicates the implementation of control combination Total cost A coefficient representing the trade-off between risk and cost; Step S64: Output decision instructions.

10. A construction risk assessment system for tunnels crossing fault fracture zones, characterized in that, include: The data acquisition module is used to collect multi-source data in real time during the construction process; The feature extraction module is used to perform multimodal feature fusion processing on multi-source data and extract dynamic feature sets; The risk assessment module includes a dual-channel interactive processing unit, which is used to input dynamic feature sets into the physical mechanism channel and the data-driven channel with a two-way information interaction mechanism, calculate multiple risk indices through multi-level risk assessment, and finally merge them to obtain a comprehensive risk index. The early warning module is used to provide risk warnings based on a comprehensive risk index and generate a comprehensive early warning level. The source tracing analysis module is used to initiate and execute risk source tracing analysis based on the output of the comprehensive early warning level and risk assessment model, and to identify key risk factors and key risk transmission paths. The decision-making module is used to generate optimal control decisions based on the comprehensive early warning level and risk source analysis results; The system's modules are connected sequentially to form a closed-loop processing flow from data collection to risk decision-making.

Images