Underground engineering geological safety dynamic risk assessment method based on multi-source data fusion

Abstract

The invention discloses an underground engineering geological safety dynamic risk assessment method based on multi-source data fusion, which relates to the technical field of risk assessment, and comprises the following steps: collecting multi-source heterogeneous data related to underground engineering, extracting implicit information, modeling underground engineering geological safety risk factors into a risk network, and establishing a risk network model; calculating the comprehensive importance of the nodes based on a Stacking integration algorithm, and identifying key risk factors; acquiring characteristic parameters of key risk factors by using spatio-temporal characteristics of implicit information, introducing a random walk mechanism to acquire a dynamic accident prediction chain, and performing learning representation by using a graph attention network to acquire probability distribution of an accident evolution path; and assessing the vulnerability of the connection edge in the risk network, establishing a dynamic risk assessment model based on the node importance and the edge vulnerability, and obtaining a dynamic risk value corresponding to the accident according to the accident occurrence probability and the risk mitigation factor. According to the invention, intelligent identification, dynamic evaluation and accurate early warning of risk factors are realized, and the accuracy and real-time performance of risk identification and evaluation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of risk assessment, and more specifically, to a dynamic risk assessment method for underground engineering geological safety based on multi-source data fusion. Background Art

[0002] Underground engineering faces complex geological conditions and dynamic safety risks during construction and operation. Traditional geological safety risk assessment methods mainly rely on single data sources or static analysis, such as geological exploration reports, on-site monitoring data, or empirical models, etc., which are difficult to comprehensively reflect the dynamic risk changes of underground engineering during construction and operation. With the development of information technology, multi-source data fusion technology provides a new solution for underground engineering geological safety risk assessment, which can integrate multi-dimensional information such as geology, construction, monitoring, and environment to achieve dynamic identification and assessment of risks. Traditional underground engineering geological safety risk assessment mainly relies on the following methods: geological exploration and empirical models, single-source monitoring data analysis, and static assessment models. The limitations of these methods are isolated data and lagging updates, which cannot achieve dynamic perception and early warning of risks, resulting in an increased probability of engineering accidents (such as collapses, water inrushes, rock bursts, etc.).

[0003] Multi-source data fusion technology can significantly improve the accuracy and real-time performance of risk assessment by integrating data from different sources and scales (such as geological exploration data, InSAR remote sensing monitoring, sensor real-time data, construction logs, environmental factors, etc.) and combining methods such as machine learning, deep learning, and Bayesian networks. The geological conditions of underground engineering are characterized by concealment, mutation, and uncertainty, and traditional static assessment methods are difficult to meet the requirements of engineering safety management. Therefore, how to implement a dynamic risk assessment method for underground engineering geological safety based on multi-source data and provide more scientific and reliable technical support for the safe construction and operation of underground engineering is an urgent problem to be solved. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes a dynamic risk assessment method for underground engineering geological safety based on multi-source data fusion. By integrating multi-dimensional data such as geological exploration, construction monitoring, and environmental factors, and combining complex network theory and machine learning algorithms, it realizes the dynamic assessment and accurate early warning of risk factors.

[0005] The first aspect of the present invention provides a dynamic risk assessment method for underground engineering geological safety based on multi-source data fusion, including the following steps: Collect multi-source heterogeneous data related to underground engineering, preprocess the multi-source heterogeneous data, extract implicit information from the preprocessed multi-source heterogeneous data, and analyze the spatio-temporal characteristics of the implicit information; Obtain the geological safety risk factors of underground engineering, model the geological safety risk factors of underground engineering as a risk network, calculate the comprehensive importance of nodes in the risk network based on the Stacking integration algorithm, and identify key risk factors; Use the spatio-temporal characteristics of implicit information to obtain the characteristic parameters of key risk factors, introduce a random walk mechanism to obtain a dynamic accident prediction chain, and use a graph attention network to learn and represent the dynamic accident prediction chain to obtain the probability distribution of the accident evolution path; Evaluate the vulnerability of the connection edges in the risk network, establish a dynamic risk assessment model based on the node comprehensive importance and edge vulnerability, obtain the dynamic risk value corresponding to the accident according to the accident occurrence probability and risk mitigation factor, and visualize the underground engineering accidents and dynamic risk values.

[0006] In this solution, multi-source heterogeneous data related to underground engineering is collected, the multi-source heterogeneous data is preprocessed, and implicit information is extracted from the preprocessed multi-source heterogeneous data. Specifically: Collect geological exploration data, construction process data, real-time monitoring data, environmental data, and historical accident data involved in the whole life cycle of underground engineering as multi-source heterogeneous data, fill in missing values, process outliers, and perform normalization on the structured data in the multi-source heterogeneous data, and perform text mining on the unstructured data in the multi-source heterogeneous data; Perform spatio-temporal registration on the preprocessed multi-source heterogeneous data, cluster the spatio-temporally registered multi-source heterogeneous data according to data characteristics, obtain a low-dimensional embedding representation through eigen-decomposition of the graph Laplacian matrix, construct a multi-modal similarity matrix, use the multi-modal similarity matrix to judge and generate clustering clusters, and obtain the final clustering result; Perform graph processing on spatio-temporally correlated data to construct a spatio-temporal graph. In the spatio-temporal graph, the nodes are monitoring points, and the edges include spatial adjacency relationships and temporal correlations. Extract spatial features through graph convolutional layers, couple LSTM layers to capture time dynamics, and output corresponding data features; Use a multi-task learning network for physically coupled data to obtain corresponding data features through different task branches, and use an attention mechanism combined with a Transformer network for environment-responsive data for feature extraction to obtain corresponding data features; Integrate the data features of different clusters to generate heterogeneous features, perform feature standardization processing on the heterogeneous features, calculate the mutual information between the data features of different clusters, construct high-order interaction features based on the mutual information, and generate implicit information according to the heterogeneous features and high-order interaction features.

[0007] In this solution, analyze the spatio-temporal characteristics of implicit information. Specifically: Use the improved CEEMDAN to perform time series decomposition on the implicit information. Construct a preprocessing sequence by adding white noise with a specific ratio through multiple iterations, perform EMD decomposition on the preprocessing sequence to obtain intrinsic mode components, and screen out the intrinsic mode components greater than the preset correlation coefficient threshold. Divide the screened intrinsic mode components into long-term trend terms, periodic fluctuation terms, and sudden anomaly terms. Use the sliding window improved rescaled range method for Hurst exponent analysis, associate the Hurst exponents of different components with construction log events, and generate the time characteristics of the implicit information. Calculate the range and sill value by rock layer according to the geological data of the location of the underground project, perform modeling physical constraints through Kriging interpolation, generate the continuous spatial distribution corresponding to the implicit information based on the physical constraints combined with the CNN model, and calculate the spatial gradient of the continuous spatial distribution to generate the spatial characteristics of the implicit information. Unify the time characteristics and spatial characteristics of the implicit information into a tensor. Based on the attention mechanism, calculate the weights for the time characteristics and spatial characteristics after tensor unification, and use the weights for feature fusion to obtain the spatio-temporal characteristics of the implicit information.

[0008] In this solution, obtain the geological safety risk factors of the underground project, model the geological safety risk factors of the underground project as a risk network, calculate the comprehensive importance of the nodes in the risk network based on the Stacking ensemble algorithm, and identify the key risk factors. Specifically: Use the big data retrieval method to obtain the multi-dimensional risk factors of the geological safety accidents of the underground project, construct an interaction matrix between the multi-dimensional risk factors, convert each risk factor and accident case into network nodes, and use directed edges to represent the causal direction to construct a network topology to generate a risk network. Calculate the degree centrality, betweenness centrality, clustering coefficient, and closeness centrality indicators of the nodes based on the risk network. Introduce the Stacking ensemble strategy for node importance modeling, and use random forest, gradient boosting tree, and graph attention network as the base learners. Among them, the random forest processes the non-linear interaction of degree centrality and betweenness centrality, the gradient boosting tree captures the monotonic relationship between the clustering coefficient and closeness centrality, and the graph attention network learns the implicit features of the network topology structure. Generate a three-dimensional feature vector for each node according to the features output by the base learners, add construction stage labels and geological module codes to the three-dimensional feature vector, use the three-dimensional feature vector to train the meta-learner, and add an attention weighting mechanism and engineering constraint terms to the meta-learner to output the comprehensive importance. Preset a comprehensive importance threshold, mark the nodes greater than the preset comprehensive importance threshold, and identify the key risk factors according to the marked nodes.

[0009] In this solution, the spatio-temporal features of implicit information are used to obtain the characteristic parameters of key risk factors, and a random walk mechanism is introduced to obtain a dynamic accident prediction chain. Specifically: Based on the spatio-temporal features of the implicit information corresponding to multi-source heterogeneous data related to underground engineering, a characteristic parameter system of key risk factors is constructed, initial values are assigned to the key risk factor nodes, a template for MetaPath random walk is preset, the spatio-temporal feature similarity of the nodes is obtained based on the initial values of the key risk factor nodes, and a basic transition probability matrix is obtained by combining the network edge weights; Historical accident cases at different construction stages are obtained, the occurrence frequencies of different accident categories are analyzed according to the historical accident cases, and the accident preference coefficient of the current construction node is obtained through the occurrence frequencies of different accident categories as a correction term to correct the basic transition probability matrix to generate a dynamic transition probability matrix; Select key risk factor nodes with initial values greater than a preset threshold, initiate walks simultaneously from the selected key risk factor nodes, use the dynamic transition probability matrix for walk path sampling, and dynamically adjust the walk granularity according to the comprehensive importance of the key risk factor nodes, and merge similar node sequences to construct a prediction chain; The prediction chain is compared with the historical accident chain in a dynamic time warping mode, the historical accident chain that meets the requirements of the dynamic time warping distance is used to enhance the prediction chain, a chain weight is constructed according to the comprehensive importance of the key risk factor nodes, the edge weight and the time effect attenuation factor, and the preset number of dynamic accident prediction chains is selected using the chain weight for priority ranking.

[0010] In this solution, a graph attention network is used to learn and represent the dynamic accident prediction chain to obtain the probability distribution of the accident evolution path. Specifically: Obtain a preset number of dynamic accident prediction chains, use a graph attention network to learn and represent the dynamic accident prediction chain, set a multi-head attention mechanism guided by engineering semantics in the graph attention layer, and match the attention heads with different engineering semantics; Use the multi-head attention mechanism to calculate the dynamic attention, and adopt a deformable convolution kernel to adapt to the spatial relationship changes in different construction stages. An LSTM module is inserted between the graph attention layers to remember the state evolution history of the nodes; Gradually aggregate the neighborhood information through three layers of graph attention. Finally, obtain the context-aware embedding representation and edge embedding representation of each node, and apply special position encoding to the head and tail nodes of the dynamic accident prediction chain, and calculate the path scoring function using the context-aware embedding representation and edge embedding representation of each node; Calculate the relative probability through softmax adapted to the construction stage, generate the probability distribution of different accident evolution paths, and display the top 5 accident evolution paths in real time.

[0011] In this solution, the vulnerability of the connection edges in the risk network is evaluated. Based on the comprehensive importance of nodes and the edge vulnerability, a dynamic risk assessment model is established. The dynamic risk value corresponding to an accident is obtained according to the accident occurrence probability and the risk mitigation factor. Specifically: In the risk network, the vulnerability of the connection edges is obtained according to the dynamic betweenness centrality, dynamic load rate, and failure impact degree. A dynamic risk assessment model is established. In the dynamic risk assessment model, the state of the risk network is quantified. Based on the comprehensive importance of nodes and the edge vulnerability, the node state index and the edge activity are calculated respectively; According to the node state index and the edge activity, the path activation probability is calculated for each dynamic accident prediction chain. The global probability is integrated according to the path activation probability. The structural measure factor and the management measure factor are introduced as risk mitigation factors. The dynamic risk value is calculated according to the global probability and the risk mitigation factors; According to the dynamic risk value, the risk level is divided, and early warning information is generated based on the divided risk level.

[0012] The second aspect of the present invention provides a dynamic risk assessment system for the geological safety of underground engineering based on multi-source data fusion. The system includes a multi-source data collection and preprocessing module, an implicit information mining module, a risk network modeling and analysis module, an accident chain prediction module, a dynamic risk assessment module, and a visualization early warning module; The multi-source data collection and preprocessing module is responsible for collecting multi-source heterogeneous data related to underground engineering, preprocessing the multi-source heterogeneous data, and realizing the standardized access and fusion processing of multi-source heterogeneous data; The implicit information mining module is responsible for extracting implicit information from the preprocessed multi-source heterogeneous data and analyzing the spatio-temporal characteristics of the implicit information; The risk network modeling and analysis module is responsible for obtaining the geological safety risk factors of underground engineering, modeling the geological safety risk factors of underground engineering as a risk network, calculating the comprehensive importance of nodes in the risk network based on the Stacking integration algorithm, identifying key risk factors, and evaluating the vulnerability of the connection edges in the risk network; The accident chain prediction module is responsible for generating potential accident evolution paths based on the random walk mechanism, obtaining dynamic accident prediction chains, and using a graph attention network to learn and represent the dynamic accident prediction chains to obtain the probability distribution of the accident evolution paths; The dynamic risk assessment module is responsible for establishing a dynamic risk assessment model based on the node importance and the edge vulnerability, and obtaining the dynamic risk value corresponding to an accident according to the accident occurrence probability and the risk mitigation factors; The visualization early warning module visualizes the underground engineering accidents and the dynamic risk values, and displays the high-risk areas and the risk time series changes through a multi-dimensional early warning dashboard.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: The dynamic risk assessment method based on multi-source data fusion of the present invention integrates multi-dimensional data such as geological exploration, construction monitoring, and environmental factors, combines complex network theory and machine learning algorithms, and realizes the whole-chain closed-loop management of data perception-network modeling-intelligent prediction-dynamic assessment-decision support for underground engineering risks, significantly improving the accuracy of risk identification and the timeliness of early warning. A risk factor interaction network is constructed based on complex network theory, and the importance of nodes is dynamically calculated through the Stacking integration algorithm to reflect the risk evolution path in real time. A random walk mechanism is introduced to generate an accident prediction chain, the risk propagation mode is learned by combining the graph attention network, and real-time monitoring data is assimilated through the Bayesian network to realize the minute-level update of risk probability, enhancing the accuracy and interpretability of risk early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments or exemplifications of the present invention, the following will briefly introduce the drawings required for use in the embodiments or exemplifications. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts.

[0015] Figure 1 Shows the flowchart of the dynamic risk assessment method for geological safety of underground engineering based on multi-source data fusion; Figure 2 Shows the flowchart of calculating the comprehensive importance of risk network nodes to identify key risk factors; Figure 3 Shows the flowchart of introducing a random walk mechanism to obtain a dynamic accident prediction chain; Figure 4 Shows the block diagram of the dynamic risk assessment system for geological safety of underground engineering based on multi-source data fusion. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0016] In order to better understand the above-mentioned objects, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments can be combined with each other.

[0017] Many specific details are set forth in the following description to facilitate a thorough understanding of the present invention. However, the present invention can be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0018] Figure 1The flowchart of the dynamic risk assessment method for underground engineering geological safety based on multi-source data fusion is shown.

[0019] As Figure 1 shown, this embodiment provides a dynamic risk assessment method for underground engineering geological safety based on multi-source data fusion, including: S102, collecting multi-source heterogeneous data related to underground engineering, preprocessing the multi-source heterogeneous data, extracting implicit information from the preprocessed multi-source heterogeneous data, and analyzing the spatio-temporal characteristics of the implicit information; S104, obtaining the underground engineering geological safety risk factors, modeling the underground engineering geological safety risk factors as a risk network, calculating the comprehensive importance of nodes in the risk network based on the Stacking integration algorithm, and identifying the key risk factors; S106, using the spatio-temporal characteristics of the implicit information to obtain the characteristic parameters of the key risk factors, introducing a random walk mechanism to obtain a dynamic accident prediction chain, and using a graph attention network to learn and represent the dynamic accident prediction chain to obtain the probability distribution of the accident evolution path; S108, evaluating the vulnerability of the connection edges in the risk network, establishing a dynamic risk assessment model based on the node comprehensive importance and edge vulnerability, obtaining the corresponding dynamic risk value of the accident according to the accident occurrence probability and the risk mitigation factor, and visualizing the underground engineering accident and the dynamic risk value.

[0020] It should be noted that the geological exploration data, construction process data, real-time monitoring data, environmental data, and historical accident data involved in the whole life cycle of underground engineering are collected as multi-source heterogeneous data; the address exploration data includes borehole data, geological profiles, rock mass mechanical parameters, hydrogeological data, etc.; the construction process data includes excavation progress, support parameters, construction logs, quality inspection records, etc.; the real-time monitoring data includes time series data such as displacement monitoring, stress monitoring, seepage pressure monitoring, and microseismic monitoring, and the environmental data includes external factor data such as rainfall, temperature, and vibration monitoring; the historical accident data includes the accident case library and treatment plans of similar projects. Missing value filling, outlier processing, and normalization processing are performed on the structured data (monitoring data, construction parameters) in the multi-source heterogeneous data, and text mining is performed on the unstructured data (geological reports, construction logs) in the multi-source heterogeneous data; the preprocessed multi-source heterogeneous data is spatio-temporally registered to establish a mapping relationship between the data. For example, coordinate system conversion is performed on the geological exploration data (borehole coordinates), monitoring point layout positions, and construction area models to unify them to the engineering coordinate system, and timestamp standardization is performed on the time series data (monitoring sensors, construction records).

[0021] Cluster the multi-source heterogeneous data after spatio-temporal registration according to data characteristics, and design clustering features according to different data types. For structured data, use time series features (mean, variance, trend slope) and spatial features (correlation of adjacent points) as clustering features; for unstructured data, use Doc2Vec document vectors and geological entity recognition results as clustering features; for image data, extract histograms of oriented gradients combined with texture features as clustering features. Obtain a low-dimensional embedding representation through eigen-decomposition of the graph Laplacian matrix, construct a multi-modal similarity matrix, use Euclidean distance for structured data, cosine similarity for text, and SSIM structural similarity for images, and use the multi-modal similarity matrix to judge and generate clustering clusters to obtain the final clustering result; perform graph processing on spatio-temporal associated data, construct a spatio-temporal graph, where the nodes in the spatio-temporal graph are monitoring points, and the edges contain spatial adjacency relationships and temporal correlations, extract spatial features through graph convolutional layers, such as deformation propagation patterns, and couple LSTM layers to capture temporal dynamics, such as displacement acceleration trends, and output corresponding data features; for physically coupled data, use a multi-task learning network to obtain corresponding data features through different task branches. For example, use 1D CNN to extract local fluctuation features of stress data, perform wavelet packet decomposition energy spectrum analysis of vibration signals in the first task branch, and perform regression modeling of construction parameters and stress responses in the second task branch to obtain corresponding dynamic load influence coefficients and construction disturbance sensitivities. For environment-responsive data, use an attention mechanism combined with a Transformer network for feature extraction to obtain corresponding data features; for example, perform discrete wavelet transform on seepage pressure data to decompose multi-scale features, use attention weights to calculate the differential impact of rainfall events on each monitoring point, and fuse hydrogeological parameters through cross-modal attention to output hydraulic coupling strengths, etc. Integrate the data features of different clusters to generate heterogeneous features, perform feature standardization processing on the heterogeneous features, and calculate the mutual information between the data features of different clusters, such as the correlation between vibration sensitivity and rock mass deformation synergy. Based on the mutual information, construct high-order interaction features, such as the rainfall-seepage pressure-support stress three-modal coupling index, and generate implicit information according to the heterogeneous features and high-order interaction features. Efficiently transform the multi-source heterogeneous data from the original information to the risk implicit features, providing an accurate data basis for dynamic risk assessment.

[0022] It should be noted that the improved CEEMDAN is used to decompose the implicit information in time series. A preprocessing sequence is constructed by adding white noise with a specific ratio through multiple iterations to effectively separate the mode mixing components. According to the characteristics of underground engineering data, the noise amplitude is taken as 5-10% of the standard deviation of the monitoring data. The EMD decomposition is performed on the preprocessing sequence to obtain the intrinsic mode components, and the intrinsic mode components greater than the preset correlation coefficient threshold are screened; the screened intrinsic mode components are divided into a long-term trend term, a periodic fluctuation term, and a sudden anomaly term. The long-term trend term is a low-frequency component with a cumulative variance contribution greater than 60%, reflecting the slow evolution of the engineering structure. The periodic fluctuation term is an intermediate-frequency component containing significant periods, corresponding to the influence of the construction cycle. The sudden anomaly term is a high-frequency residual component, characterizing abnormal events such as equipment failures or geological mutations. The Hurst exponent analysis is performed using the sliding window improved rescaled range method to judge the feature persistence. The Hurst exponents of different components are associated with the construction log events to generate the time features of the implicit information; the range and sill value are calculated for each rock stratum according to the geological data of the location of the underground project, and the physical constraints are modeled through Kriging interpolation to ensure that the prediction conforms to the geological laws. Based on the physical constraints, the continuous spatial distribution corresponding to the implicit information is generated by combining with the CNN model. The spatial gradient is calculated for the continuous spatial distribution, and the high-risk transition zone is identified through the spatial gradient field to generate the spatial features of the implicit information; the time features and spatial features of the implicit information are tensor unified, and the weights are calculated for the time features and spatial features after tensor unification based on the attention mechanism, and the weights are used for feature fusion to obtain the spatio-temporal features of the implicit information.

[0023] Figure 2 The flowchart showing the calculation of the comprehensive importance of the risk network nodes to identify the key risk factors is shown.

[0024] According to the embodiments of the present invention, the geological safety risk factors of the underground project are obtained, the geological safety risk factors of the underground project are modeled as a risk network, and the comprehensive importance of the nodes in the risk network is calculated based on the Stacking ensemble algorithm to identify the key risk factors, specifically: S202, using the big data retrieval method to obtain the multi-dimensional risk factors of the geological safety accidents of the underground project, and constructing an interaction matrix between the multi-dimensional risk factors. Each risk factor and accident case are converted into network nodes, and the causal direction is represented by a directed edge to construct the network topology to generate the risk network; S204. Calculate the degree centrality, betweenness centrality, clustering coefficient, and closeness centrality metrics of the nodes in the risk network, introduce the Stacking integration strategy for node importance modeling, and use random forest, gradient boosting tree, and graph attention network as base learners. Among them, the random forest processes the non-linear interaction between degree centrality and betweenness centrality, the gradient boosting tree captures the monotonic relationship between clustering coefficient and closeness centrality, and the graph attention network learns the implicit features of the network topology; S206. Generate a three-dimensional feature vector for each node according to the features output by the base learners, add construction stage labels and geological module codes to the three-dimensional feature vector, and use the three-dimensional feature vector to train the meta-learner. Add an attention weighting mechanism and engineering constraint terms to the meta-learner to output the comprehensive importance; S208. Preset a comprehensive importance threshold, mark the nodes greater than the preset comprehensive importance threshold, and identify key risk factors based on the marked nodes.

[0025] It should be noted that a big data retrieval method is used to obtain multi-dimensional risk factors of underground engineering geological safety accidents, including geological body factors, engineering response factors, and environmental inducement factors; the geological body factors include rock mass quality indicators, geological structure characteristics, hydrogeological parameters, etc., the engineering response factors include surrounding rock deformation, support status, construction disturbance, etc., and the environmental inducement factors include meteorology and hydrology, surrounding loads, etc. Establish deterministic associations based on geomechanics principles as the physical interaction relationships between factors, mine potential associations through historical data as the statistical correlation relationships between factors, and integrate domain knowledge to supplement implicit associations as the empirical correlation relationships between factors. Each risk factor is converted into a network node, and the node attributes include basic attributes such as factor type and monitoring location, and dynamic attributes such as current value and change trend. Directed edges are constructed between nodes through different causal relationships, and different edge weights are determined.

[0026] In the calculation of degree centrality index, the sum of the weights of all directed edges starting from the current node is statistically counted to quantify the external influence ability of the node. For example, the high out-degree of the excavation speed node reflects its significant influence on multiple support monitoring points; and the sum of the weights of all directed edges pointing to the node is statistically counted to characterize the degree to which the node is vulnerable to other factors. For example, the weak interlayer node usually has the characteristic of high in-degree, and the degree value is mapped to the interval [0, 1]. In the calculation of betweenness centrality, only the actual effective propagation paths with a length ≤ 3 are considered. Based on the maximum flow minimum cut theory, the bottleneck role of the node in risk propagation is analyzed. For example, the groundwater pressure node often shows a high betweenness in the seepage path. In the calculation of clustering coefficient, only the triangular closure of adjacent nodes within the same geological unit is calculated. For example, the monitoring point cluster within the fault fracture zone usually shows a high clustering property, and two modes of risk input clustering and risk output clustering are distinguished. In the calculation of closeness centrality, 1 / edge weight is used as the distance metric, and the local closeness is evaluated in units of construction partitions to track the evolution trend of the closeness of key nodes during the construction process.

[0027] In the Stacking ensemble modeling stage, random forest, gradient boosting tree and graph attention network are used as base learners to obtain the three-dimensional feature vectors generated by each node. Engineering features such as construction stage labels and geological unit codes are added for spatio-temporal context enhancement, and the contribution weights of each base model are automatically adjusted according to the current network density. The base learners are trained independently, and 5-fold time series cross-validation is adopted. The meta-features output by the base learners are used to train the meta-learner. An attention weighting mechanism is designed in the meta-learner to ensure that key construction factors maintain high weights, and the comprehensive importance output by the meta-learner is used. Through Stacking ensemble, the evaluation of the importance of risk factors has achieved a leap from a single perspective to multi-dimensional fusion. The established comprehensive importance index has both statistical significance and engineering interpretability, providing a quantitative decision-making basis for the precise prevention and control of underground engineering risks.

[0028] Figure 3 The flowchart showing the acquisition of the dynamic accident prediction chain by introducing the random walk mechanism is shown.

[0029] According to the embodiments of the present invention, the characteristic parameters of key risk factors are obtained using the spatio-temporal characteristics of implicit information, and the random walk mechanism is introduced to obtain the dynamic accident prediction chain. Specifically: S302, construct a characteristic parameter system of key risk factors based on the spatio-temporal characteristics of implicit information corresponding to multi-source heterogeneous data related to underground engineering, perform initial assignment for the key risk factor nodes, preset the template of MetaPath random walk, obtain the node spatio-temporal feature similarity based on the initial values of the key risk factor nodes, and combine the network edge weights to obtain the basic transition probability matrix; S304. Obtain historical accident cases in different construction stages, analyze the occurrence frequencies of different accident categories based on the historical accident cases, obtain the accident preference coefficient of the current construction node as a correction term through the occurrence frequencies of different accident categories, and correct the basic transition probability matrix to generate a dynamic transition probability matrix; S306. Select key risk factor nodes with initial values greater than a preset threshold, initiate random walks simultaneously from the selected key risk factor nodes, sample the random walk paths using the dynamic transition probability matrix, dynamically adjust the random walk granularity according to the comprehensive importance of the key risk factor nodes, and merge similar node sequences to construct a prediction chain; S308. Compare the prediction chain with the historical accident chain in a dynamic time warping pattern, use the historical accident chain that meets the requirements of the dynamic time warping distance to enhance the prediction chain, construct a chain weight according to the comprehensive importance of the key risk factor nodes, edge weights, and time decay factors, and select a preset number of dynamic accident prediction chains using the chain weight.

[0030] It should be noted that by presetting the template of MetaPath random walk through engineering semantics, such as excavation parameters → vibration → loosening zone → settlement, rock formation → seepage pressure → displacement → support stress, rainfall → groundwater level → weak interlayer → slip. Generate the basic probability by combining the network edge weights with the node spatio-temporal feature similarity, analyze the occurrence frequencies of different accident categories based on the historical accident cases, obtain the accident preference coefficient of the current construction node as a correction term through the occurrence frequencies of different accident categories, such as strengthening the geological structure related paths during the support period. Initiate random walks simultaneously from the selected key risk factor nodes, sample the random walk paths using the dynamic transition probability matrix, along the main risk conduction path, the selection probability of nodes with high comprehensive importance is increased, and refined search is carried out inside the local high-risk cluster. Dynamically adjust the random walk granularity according to the comprehensive importance of the key risk factor nodes, merge similar node sequences to construct a prediction chain, verify the time series relationship through Granger causality test for causal verification, and finally obtain the prediction chain. Construct a chain weight according to the comprehensive importance of the key risk factor nodes, edge weights, and time decay factors, , and use engineering semantics to annotate the chain type, such as annotating seepage erosion type, vibration accumulation type, etc.

[0031] It should be noted that to obtain a preset number of dynamic accident prediction chains, a graph attention network is used to learn and represent the dynamic accident prediction chains. A multi-head attention mechanism guided by engineering semantics is set in the graph attention layer to match different engineering semantics with attention heads. Among them, the geology head focuses on features such as rock formation continuity and tectonic strike, the mechanics head focuses on stress-strain conduction relationships, and the construction head tracks the impact of construction processes and technologies. The multi-head attention mechanism is used to calculate dynamic attention, and a deformable convolutional kernel is adopted to adapt to changes in spatial relationships at different construction stages. The convolutional radius is dynamically adjusted according to the excavation progress. An LSTM module is inserted between graph attention layers to memorize the state evolution history of nodes. The model is trained using single accident types, multi-type mixtures, and extreme working conditions with added noise. Neighborhood information is gradually aggregated through three layers of graph attention, and finally, a context-aware embedding representation and edge embedding representation of each node are obtained. Special position encodings are applied to the head and tail nodes of the dynamic accident prediction chain, and a path scoring function is calculated using the context-aware embedding representation and edge embedding representation of each node; , is a learnable parameter matrix, is a type bias term. The relative probability is calculated through softmax adapted to the construction stage, , where is the temperature coefficient. During forward propagation, 20% of the attention heads are randomly discarded, and the variance of the probability distribution is obtained through 50 samplings. When using Monte Carlo dropout for forward propagation, 20% of the attention heads are randomly discarded, and the variance of the probability distribution is obtained through 50 samplings to generate the probability distribution of different accident evolution paths, and the top 5 accident evolution paths with probabilities are displayed in real time. A data closed-loop of prediction-disposal-verification is established. The weight of the path samples correctly predicted is increased by 30%, and false alarm paths trigger model fine-tuning.

[0032] It should be noted that in the risk network, the vulnerability of the connection edges is obtained based on the dynamic betweenness, dynamic load ratio, and failure impact degree; for the betweenness index, an improved flow-sensitive calculation method is adopted to count the proportion of all effective risk conduction paths passing through this edge; the dynamic load ratio calculates the ratio of the current risk volume transmitted by the edge (based on the node state difference) to its theoretical bearing capacity. For example, the load ratio of the edge between the support structure and the surrounding rock = measured axial force / design allowable axial force × time-varying reduction coefficient. The failure impact degree is obtained by randomly disconnecting this edge through Monte Carlo simulation and observing the decline in the global efficiency of the network, especially considering the cascading effect guided by geological structures. For example, the failure of an edge near a fault zone will have a wider impact. Identify the peak time periods of edge vulnerability during the construction process, divide the vulnerability benchmark areas according to geological exploration results, and use similar correction coefficients for edges within the same rock formation unit. Force the vulnerability of edges that do not conform to mechanical principles (such as those with no medium action over long distances) to be zero.

[0033] A dynamic risk assessment model is established. In the dynamic risk assessment model, the risk network state is quantified, and the node state index and the edge activity are calculated based on the comprehensive importance of the nodes and the vulnerability of the edges respectively. The node state index quantifies the node risk level and is expressed as , is the material nonlinear coefficient. The edge activity reflects the risk conduction intensity of the connecting edge and is expressed as , is the construction quality control level. According to the node state index and the edge activity, the path activation probability is calculated for each dynamic accident prediction chain , which characterizes the possibility of the occurrence of the dynamic accident prediction chain and is expressed as , is the edge activity of the i-th edge, represents the node state index of the j-th node, is the path length penalty factor. According to the path activation probability, the global probability is integrated. The global risk probability considers the integrated risk of all critical paths and is expressed as , is the probability of the k-th path. The structural measure factor and the management measure factor are introduced as risk mitigation factors. The structural measure factor is expressed as: , and the management measure factor is expressed as: . Preferably, the emergency plan score is collected from historical expert scores. According to the global probability and the risk mitigation factors, the dynamic risk value is calculated, , is the engineering importance coefficient. According to the dynamic risk value, the risk level is divided. Based on the historical accident data, a dynamic risk value-loss curve is established, and the threshold is set according to the engineering risk preference. During the critical construction stage (such as crossing a fault), the overall threshold is lowered, and warning information is generated based on the divided risk levels.

[0034] Figure 4 Fig. shows the block diagram of the underground engineering geological safety dynamic risk assessment system based on multi-source data fusion.

[0035] The second embodiment of the present invention provides an underground engineering geological safety dynamic risk assessment system 4 based on multi-source data fusion, including: a multi-source data collection and preprocessing module 401, a hidden information mining module 402, a risk network modeling and analysis module 403, an accident chain prediction module 404, a dynamic risk assessment module 405, and a visualization warning module 406; The multi-source data collection and preprocessing module 401 is responsible for collecting multi-source heterogeneous data related to underground engineering, preprocessing the multi-source heterogeneous data, and realizing the standardized access and fusion processing of the multi-source heterogeneous data; The implicit information mining module 402 is responsible for extracting implicit information from the preprocessed multi-source heterogeneous data and analyzing the spatio-temporal characteristics of the implicit information; The risk network modeling and analysis module 403 is responsible for obtaining the geological safety risk factors of underground engineering, modeling the geological safety risk factors of underground engineering as a risk network, calculating the comprehensive importance of nodes in the risk network based on the Stacking integration algorithm, identifying key risk factors, and evaluating the vulnerability of the connection edges in the risk network; The accident chain prediction module 404 is responsible for generating potential accident evolution paths based on the random walk mechanism, obtaining a dynamic accident prediction chain, using a graph attention network to learn and represent the dynamic accident prediction chain, and obtaining the probability distribution of the accident evolution path; The dynamic risk assessment module 405 is responsible for establishing a dynamic risk assessment model based on the node importance and edge vulnerability, and obtaining the dynamic risk value corresponding to the accident according to the accident occurrence probability and the risk mitigation factor; The visualization and early warning module 406 visualizes the underground engineering accidents and the dynamic risk values, and displays the high-risk areas and the temporal changes of the risks through a multi-dimensional early warning dashboard.

[0036] In a third aspect of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium includes a program for the method of dynamically assessing the geological safety of underground engineering based on multi-source data fusion. When the program for the method of dynamically assessing the geological safety of underground engineering based on multi-source data fusion is executed by a processor, the steps of the method of dynamically assessing the geological safety of underground engineering based on multi-source data fusion are implemented.

[0037] In several embodiments provided in the present application, it should be understood that the disclosed method and system can be implemented in other ways. The system embodiments described above are only illustrative. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed with each other can be through some interfaces, and the indirect coupling or communication connection of devices or modules can be electrical, mechanical, or other forms. In addition, in each embodiment of the present invention, the various functional modules can all be integrated in one processing module, or each module can be separately used as one module, or two or more modules can be integrated in one module; the above integrated modules can be implemented in the form of hardware, or in the form of a combination of hardware and software functional modules.

[0038] Those of ordinary skill in the art can understand that all or part of the steps to implement the above method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including the above method embodiments. The aforementioned storage medium includes various media that can store program codes, such as removable storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs.

[0039] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered within the protection scope of the present invention.

Claims

1. A dynamic risk assessment method for geological safety of underground engineering based on multi-source data fusion, characterized in that It includes the following steps: Collect multi-source heterogeneous data related to underground engineering, preprocess the multi-source heterogeneous data, extract implicit information from the preprocessed multi-source heterogeneous data, and analyze the spatio-temporal characteristics of the implicit information; Obtain the geological safety risk factors of underground engineering, model the geological safety risk factors of underground engineering as a risk network, calculate the comprehensive importance of nodes in the risk network based on the Stacking integration algorithm, and identify key risk factors; Use the spatio-temporal characteristics of the implicit information to obtain the characteristic parameters of the key risk factors, introduce a random walk mechanism to obtain a dynamic accident prediction chain, use a graph attention network to learn and represent the dynamic accident prediction chain, and obtain the probability distribution of the accident evolution path; Evaluate the vulnerability of the connection edges in the risk network, establish a dynamic risk assessment model based on the node comprehensive importance and edge vulnerability, obtain the dynamic risk value corresponding to the accident according to the accident occurrence probability and the risk mitigation factor, and visualize the underground engineering accident and the dynamic risk value.

2. The method for dynamically assessing the geological safety risks of underground engineering based on multi-source data fusion according to claim 1, wherein Collect multi-source heterogeneous data related to underground engineering, preprocess the multi-source heterogeneous data, and extract implicit information from the preprocessed multi-source heterogeneous data. Specifically: Collect geological exploration data, construction process data, real-time monitoring data, environmental data, and historical accident data involved in the whole life cycle of underground engineering as multi-source heterogeneous data. Fill in missing values, process outliers, and perform normalization on the structured data in the multi-source heterogeneous data, and perform text mining on the unstructured data in the multi-source heterogeneous data; Perform spatio-temporal registration on the preprocessed multi-source heterogeneous data, cluster the spatio-temporally registered multi-source heterogeneous data according to data characteristics, obtain a low-dimensional embedding representation through eigen-decomposition of the graph Laplacian matrix, construct a multi-modal similarity matrix, use the multi-modal similarity matrix to judge and generate clustering clusters, and obtain the final clustering result; Perform graph processing on spatio-temporal correlation data, construct a spatio-temporal graph. In the spatio-temporal graph, the nodes are monitoring points, and the edges include spatial adjacency relationships and temporal correlations. Extract spatial features through graph convolutional layers, couple LSTM layers to capture time dynamics, and output corresponding data features; Use a multi-task learning network for physically coupled data to obtain corresponding data features through different task branches, and use an attention mechanism combined with a Transformer network for environment-responsive data to extract features and obtain corresponding data features; Integrate the data features of different clusters to generate heterogeneous features, perform feature standardization processing on the heterogeneous features, calculate the mutual information between the data features of different clusters, construct high-order interaction features based on the mutual information, and generate implicit information according to the heterogeneous features and high-order interaction features.

3. The method for dynamically assessing the geological safety risk of underground engineering based on multi-source data fusion according to claim 2, wherein Analyze the spatio-temporal characteristics of the implicit information. Specifically: Use the improved CEEMDAN to perform time series decomposition on the implicit information, construct a preprocessing sequence by adding white noise with a specific proportion through multiple iterations, perform EMD decomposition on the preprocessing sequence to obtain intrinsic mode components, and screen out the intrinsic mode components greater than the preset correlation coefficient threshold; The screening and acquisition of intrinsic mode components are divided into long-term trend terms, periodic fluctuation terms, and sudden anomaly terms. The Hurst exponent analysis is carried out by using the sliding window improved rescaled range method. The Hurst exponents of different components are associated with construction log events to generate time characteristics of implicit information; Calculate the range and sill value by rock stratum according to the geological data of the location of the underground project, perform physical constraint modeling through Kriging interpolation, and generate the continuous spatial distribution corresponding to the implicit information based on the physical constraint combined with the CNN model. Calculate the spatial gradient of the continuous spatial distribution to generate the spatial characteristics of the implicit information; Unify the time characteristics and spatial characteristics of the implicit information into a tensor. Based on the attention mechanism, calculate the weights for the time characteristics and spatial characteristics after tensor unification, and use the weights for feature fusion to obtain the spatio-temporal characteristics of the implicit information.

4. The method for dynamically evaluating the geological safety risks of underground engineering based on multi-source data fusion according to claim 1, wherein Obtain the geological safety risk factors of the underground project, model the geological safety risk factors of the underground project as a risk network, and calculate the comprehensive importance of nodes in the risk network based on the Stacking integration algorithm to identify key risk factors. Specifically: Use the big data retrieval method to obtain multi-dimensional risk factors of geological safety accidents of underground projects, construct an interaction matrix between multi-dimensional risk factors, convert each risk factor and accident case into network nodes, and use directed edges to represent the causal direction to construct a network topology to generate a risk network; Based on the risk network, calculate the degree centrality, betweenness centrality, clustering coefficient, and closeness centrality indicators of nodes, introduce the Stacking integration strategy for node importance modeling, and use random forest, gradient boosting tree, and graph attention network as base learners. Among them, the random forest processes the non-linear interaction between degree centrality and betweenness centrality, the gradient boosting tree captures the monotonic relationship between clustering coefficient and closeness centrality, and the graph attention network learns the implicit features of the network topology structure; Generate a three-dimensional feature vector for each node according to the features output by the base learner, add construction stage labels and geological module codes to the three-dimensional feature vector, use the three-dimensional feature vector to train the meta-learner, and add an attention weighting mechanism and engineering constraint terms to the meta-learner to output the comprehensive importance; Preset a comprehensive importance threshold, mark the nodes greater than the preset comprehensive importance threshold, and identify key risk factors according to the marked nodes.

5. The method for dynamically assessing the geological safety risks of underground engineering based on multi-source data fusion according to claim 1, characterized in that Use the spatio-temporal characteristics of the implicit information to obtain the characteristic parameters of the key risk factors, and introduce the random walk mechanism to obtain the dynamic accident prediction chain. Specifically: Construct a characteristic parameter system of key risk factors based on the spatio-temporal characteristics of implicit information corresponding to multi-source heterogeneous data related to underground projects, perform initial assignment for key risk factor nodes, preset the template of MetaPath random walk, obtain the spatio-temporal feature similarity of nodes based on the initial values of key risk factor nodes, and combine the network edge weights to obtain the basic transition probability matrix; Obtain historical accident cases at different construction stages, analyze the occurrence frequencies of different accident categories based on the historical accident cases, and obtain the accident preference coefficient of the current construction node as a correction term through the occurrence frequencies of different accident categories to correct the basic transition probability matrix to generate a dynamic transition probability matrix; Select key risk factor nodes with initial values greater than a preset threshold, initiate walks simultaneously from the selected key risk factor nodes, sample walk paths using the dynamic transition probability matrix, and dynamically adjust the walk granularity according to the comprehensive importance of key risk factor nodes, and merge similar node sequences to construct a prediction chain; Compare the prediction chain with the historical accident chain in a dynamic time warping pattern, use the historical accident chain that meets the requirements of the dynamic time warping distance to enhance the prediction chain, construct a chain weight according to the comprehensive importance of key risk factor nodes, edge weights, and time decay factors, and use the chain weight to perform priority sorting to select a preset number of dynamic accident prediction chains.

6. The method for dynamically evaluating the geological safety risk of underground engineering based on multi-source data fusion according to claim 1, characterized in that, Use a graph attention network to learn the representation of the dynamic accident prediction chain and obtain the probability distribution of the accident evolution path, specifically: Obtain a preset number of dynamic accident prediction chains, use a graph attention network to learn the representation of the dynamic accident prediction chain, set a multi-head attention mechanism guided by engineering semantics in the graph attention layer, and match the attention heads with different engineering semantics; Calculate the dynamic attention using the multi-head attention mechanism, and adopt a deformable convolutional kernel to adapt to the spatial relationship changes in different construction stages. Insert an LSTM module between the graph attention layers to memorize the state evolution history of the nodes; Gradually aggregate the neighborhood information through three layers of graph attention, finally obtain the context-aware embedding representation of each node and the edge embedding representation, and apply special position encoding to the start and end nodes of the dynamic accident prediction chain, and calculate the path scoring function using the context-aware embedding representation of each node and the edge embedding representation; Calculate the relative probability through softmax adapted to the construction stage, generate the probability distribution of different accident evolution paths, and display the top 5 accident evolution paths in real time.

7. The method for dynamically assessing the geological safety risks of underground engineering based on multi-source data fusion according to claim 1, wherein Evaluate the vulnerability of the connecting edges in the risk network, establish a dynamic risk assessment model based on the comprehensive importance of nodes and the edge vulnerability, and obtain the dynamic risk value corresponding to the accident according to the accident occurrence probability and the risk mitigation factor, specifically: Obtain the edge vulnerability in the risk network according to the dynamic edge betweenness, dynamic load rate, and failure impact degree, establish a dynamic risk assessment model, quantify the state of the risk network in the dynamic risk assessment model, and calculate the node state index and edge activity respectively based on the comprehensive importance of nodes and the edge vulnerability; Calculate the path activation probability for each dynamic accident prediction chain according to the node state index and edge activity, integrate the global probability according to the path activation probability, introduce the structural measure factor and the management measure factor as risk mitigation factors, and calculate the dynamic risk value according to the global probability and the risk mitigation factor; Classify the risk levels according to the dynamic risk value, and generate early warning information based on the classified risk levels.

8. An underground engineering geological safety dynamic risk assessment system based on multi-source data fusion, characterized in that, Implement the dynamic risk assessment method for underground engineering geological safety based on multi-source data fusion as described in any one of claims 1-7. The system includes a multi-source data acquisition and preprocessing module, a hidden information mining module, a risk network modeling and analysis module, an accident chain prediction module, a dynamic risk assessment module, and a visualization warning module; The multi-source data acquisition and preprocessing module is responsible for collecting multi-source heterogeneous data related to underground engineering, preprocessing the multi-source heterogeneous data, and realizing the standardized access and fusion processing of multi-source heterogeneous data; The hidden information mining module is responsible for extracting hidden information from the preprocessed multi-source heterogeneous data and analyzing the spatio-temporal characteristics of the hidden information; The risk network modeling and analysis module is responsible for obtaining the geological safety risk factors of underground engineering, modeling the geological safety risk factors of underground engineering as a risk network, calculating the comprehensive importance of nodes in the risk network based on the Stacking integration algorithm, identifying key risk factors, and evaluating the vulnerability of the connection edges in the risk network; The accident chain prediction module is responsible for generating potential accident evolution paths based on the random walk mechanism, obtaining the dynamic accident prediction chain, using the graph attention network to learn and represent the dynamic accident prediction chain, and obtaining the probability distribution of the accident evolution path; The dynamic risk assessment module is responsible for establishing a dynamic risk assessment model based on the node importance and edge vulnerability, and obtaining the dynamic risk value corresponding to the accident according to the accident occurrence probability and the risk mitigation factor; The visualization warning module visualizes the underground engineering accidents and the dynamic risk values, and displays the high-risk areas and the risk time-series changes through a multi-dimensional warning dashboard.

Images