SCL field safety accident analysis and early warning method and system, medium and program product

Abstract

The invention discloses an SCL field safety accident analysis and early warning method and system, a medium and a program product, and the method comprises the steps: carrying out the distributed cleaning and feature extraction of multi-source heterogeneous monitoring data, and obtaining a multi-dimensional time sequence feature flow; constructing a dynamic association network of a security state; identifying a key influence factor chain, and establishing a semantic analysis model; constructing a safety threshold system; evaluating the multi-dimensional time sequence characteristic flow by using a safety threshold system; when the multi-dimensional time sequence characteristic flow triggers early warning, performing tracing analysis by using a semantic analysis model to obtain a root event triggering the early warning, and simulating the development trend of the root event based on a preset evolution model; determining a predicted risk range and a risk type according to the tracing cause analysis result and the accident development trend; and the corresponding accident early warning scheme is matched in the preset database according to the risk range and the risk type, and the accident early warning scheme is sent to the monitoring terminal, so that the accuracy of judgment and classified early warning of different types of safety accidents is improved.

Description

Technical Field

[0001] The present application belongs to the field of safety accident analysis and early warning, and in particular, relates to a SCL on-site safety accident analysis and early warning method, system, medium and program product. Background Art

[0002] As a technical method widely used in tunnel construction, the SCL (Safety Checklist) method often involves complex geological conditions and construction environments during the construction process. Due to the uncertainty of the construction environment and the influence of human operation factors, safety accidents occur from time to time, such as cracking of shotcrete support structures and excessive deformation of the strata. Traditional safety accident analysis mainly relies on manual experience judgment and regular inspections. This method has a lag in analysis and it is difficult to accurately predict the development trend of accidents.

[0003] In related technologies, a sensor network can be deployed at the construction site to collect data such as support structure stress, ground displacement, construction parameters, etc. in real time, and the collected data can be analyzed using data mining algorithms to establish an accident warning model. When the monitoring data exceeds the preset threshold, the system automatically issues a warning signal, thereby achieving timely warning of safety accidents.

[0004] However, different types of safety accidents will exhibit unique data characteristics during their occurrence and evolution. If the same data analysis model is used, it will be difficult to accurately capture the characteristic fingerprints of various types of accidents, and it will be difficult to make accurate judgments and classified warnings for different types of safety accidents. Summary of the invention

[0005] The present application provides a SCL on-site safety accident analysis and early warning method, system, medium and program product, which are used to improve the accuracy of judgment and classification early warning of different types of safety accidents.

[0006] In the first aspect, the present application provides a SCL on-site safety accident analysis and early warning method, which collects multi-source heterogeneous monitoring data of the tunnel in real time, and performs distributed cleaning and feature extraction on the multi-source heterogeneous monitoring data to obtain a multi-dimensional time series feature stream; A dynamic correlation network of security status is constructed with multi-dimensional time series feature flow. The dynamic correlation network is used to capture the spatiotemporal coupling relationship between different monitoring indicators. According to the coupling strength of the spatiotemporal coupling relationship, the key influencing factor chain is identified, and the semantic parsing model of the safety status is established in combination with the key influencing factor chain. The semantic parsing model is used to reveal the internal mechanism and evolution law of the occurrence of safety accidents; A safety threshold system is built based on the output of the semantic parsing model. The safety threshold system dynamically adjusts the warning threshold through a hierarchical feedback mechanism. Use the security threshold system to evaluate multi-dimensional time series feature flows; When the multi-dimensional time series feature flow triggers an early warning, the semantic parsing model is used to perform causal analysis to obtain the root event that triggered the early warning, and the development trend of the root event is simulated based on the preset evolution model; Determine the expected risk range and risk type based on the results of the causal analysis and the development trend of the accident; According to the risk range and risk type, the corresponding accident warning plan is matched in the preset database, and the accident warning plan is sent to the monitoring terminal.

[0007] By adopting the above technical solutions, multi-source heterogeneous monitoring data is collected in real time and distributed cleaning and feature extraction are performed, so as to obtain a more comprehensive and accurate multi-dimensional time series feature flow, reducing the impact of data noise and redundant information. The dynamic association network constructed by multi-dimensional time series feature flow can accurately capture the spatiotemporal coupling relationship between monitoring indicators. By identifying the key influencing factor chain and establishing a semantic parsing model, the mechanism and evolution law of safety accidents can be analyzed. The safety threshold system constructed based on the semantic parsing model has the ability to adaptively adjust, which improves the accuracy and timeliness of early warning. When the system triggers an early warning, the causal analysis using the semantic parsing model can quickly locate the root event, and the simulation of the accident development trend combined with the preset evolution model can improve the accuracy of predicting the risk range and type. By matching the corresponding accident early warning scheme in the preset database, the accuracy of judgment and classification early warning of different types of safety accidents is improved, and the losses caused by safety accidents are reduced.

[0008] In combination with some embodiments of the first aspect, in some embodiments, a dynamic association network of security status is constructed with a multi-dimensional time series feature flow, specifically including: The multi-dimensional time series feature stream is input into the pre-trained spatiotemporal graph neural network model. The spatiotemporal graph neural network model captures the spatial correlation between monitoring indicators through graph convolution, and captures the temporal evolution law of monitoring indicators through time convolution. The spatial correlation and temporal evolution law are integrated to obtain the spatiotemporal coupling relationship between monitoring indicators, and the spatiotemporal coupling relationship is quantified into a dynamic correlation weight matrix; A dynamic association network is constructed with monitoring indicators as nodes and dynamic association weight matrices as edges.

[0009] By adopting the above technical solution, the multi-dimensional time series feature stream is input into the pre-trained spatiotemporal graph neural network model, and the spatial correlation and temporal evolution law between the monitoring indicators are captured by graph convolution and time convolution respectively, so as to realize the deep mining of the spatiotemporal characteristics of the monitoring data. The spatiotemporal coupling relationship formed by the fusion of spatial correlation and temporal evolution law is quantitatively expressed through the dynamic correlation weight matrix, so that the degree of correlation between the monitoring indicators can be accurately measured. The dynamic correlation network constructed with monitoring indicators as nodes and dynamic correlation weight matrices as edges reduces the problem of insufficient spatiotemporal feature extraction of traditional correlation analysis methods, improves the accuracy of safety status assessment, and can more accurately identify and predict potential safety risks.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, identifying a key influencing factor chain according to the coupling strength of the spatiotemporal coupling relationship specifically includes: Calculate the propagation paths between different monitoring indicator nodes based on the dynamic association network; Determine a propagation path whose path length is less than a preset length threshold and whose cumulative propagation intensity is greater than a preset intensity threshold as an influencing factor chain; Count the triggering frequency of each influencing factor chain; The influencing factor chain whose trigger frequency is higher than the preset frequency threshold and whose corresponding coupling strength is the largest is determined as the key influencing factor chain.

[0011] By adopting the above technical solution, the propagation path between different monitoring indicator nodes in the dynamic association network is calculated, and the influencing factor chain is screened in combination with the path length threshold and the cumulative propagation intensity threshold. The statistical analysis of the triggering frequency of the influencing factor chain is combined with the comparison of the coupling strength. The reliability of the identified key influencing factor chain is improved, the one-sidedness that may be caused by a single indicator evaluation is reduced, and the accuracy of identifying the key influencing factor chain is improved.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, after building a security threshold system based on the output of the semantic parsing model, the method further includes: Map the threshold boundary of the security threshold system to the security state space of each monitoring indicator node in the dynamic association network to obtain the security state interval of each monitoring indicator; When the multi-source heterogeneous monitoring data causes the state of any monitoring indicator node to jump out of the corresponding safe state interval, the state label of the monitoring indicator node is updated, and the updated information of the state label is propagated to the adjacent monitoring indicator nodes; If the number of propagation times of the update information exceeds the preset propagation threshold, the key influencing factor chain corresponding to the update information is marked as an abnormal activation chain.

[0013] By adopting the above technical solution, the security threshold system is mapped to the security state space of each monitoring indicator node in the dynamic association network. When the state of the monitoring indicator node is abnormal, the dynamic changes of the security state can be captured in time through the state label update and information propagation mechanism. The preset propagation threshold is set to determine the propagation range of the abnormal state, and the corresponding key influencing factor chain is marked as the abnormal activation chain, which realizes the precise positioning and tracking of security risks, improves the system's sensitivity to security risks, enhances the real-time and accuracy of security warnings, and makes security monitoring more intelligent and efficient.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, mapping the threshold boundary of the security threshold system to the security state space of each monitoring indicator node in the dynamic association network specifically includes: Construct a multi-layer deep neural network model, and use the threshold parameters corresponding to the threshold boundaries in the safety threshold system as input layer data; Assign weights to the threshold features of different monitoring indicators and extract the correlation patterns between the threshold features; Convert the dynamic association network into a low-dimensional feature vector and fuse it with the threshold feature; The fused threshold features are projected into the high-dimensional state space to form the safety state boundary of each monitoring indicator node.

[0015] By adopting the above technical solution, a multi-layer deep neural network model is constructed to realize the mapping of the security threshold system to the security state space of the monitoring indicator nodes in the dynamic association network, so that the threshold boundary corresponds more accurately to the actual monitoring indicator. The process of weighting the threshold features of different monitoring indicators and extracting the association pattern can accurately reflect the importance and mutual influence of different monitoring indicators in the safety assessment. The dynamic association network is converted into a low-dimensional feature vector and fused with the threshold feature, which reduces the data dimension and improves the calculation efficiency, while retaining the key information in the original network structure. Projecting the fused threshold features into the high-dimensional state space to form the security state boundary can improve the accuracy of identifying the abnormal state of each monitoring indicator, thereby improving the accuracy and reliability of the safety warning.

[0016] In combination with some embodiments of the first aspect, in some embodiments, if the number of propagation times of the update information exceeds a preset propagation threshold, after marking the key influencing factor chain corresponding to the update information as an abnormal activation chain, the method further includes: Construct a risk propagation network and calculate the attention weight of each node in the risk propagation network; Extract a node subset whose attention weight is greater than a preset weight threshold, and determine the temporal change characteristics of each node in the node subset; Determine the abnormal propagation mode based on the temporal variation characteristics; According to the evolutionary characteristics of abnormal propagation patterns, the development path and impact scope of risk propagation are determined.

[0017] By adopting the above technical solutions, a risk propagation network is constructed and the node attention weights are calculated, which enables accurate characterization of the risk propagation process and identification of key areas. Extracting a subset of high attention weight nodes and determining their temporal variation characteristics can improve the accuracy of capturing key nodes and propagation paths of risk propagation. Determining abnormal propagation patterns based on temporal variation characteristics can effectively identify the specific forms and development laws of risk propagation. Determining the development path and scope of risk propagation based on the evolutionary characteristics of abnormal propagation patterns improves the accuracy of predicting the spatial diffusion range and temporal development trend of risks.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, determining the abnormal propagation mode according to the time series variation characteristics specifically includes: Segment the temporal change feature sequence of the node subset according to the preset time window; Calculate the feature change trend and change amplitude within each preset time window; The alternative anomaly propagation mode with the greatest similarity to the feature change trend and change amplitude in the preset feature library matching is determined as the anomaly propagation mode.

[0019] By adopting the above technical solution, the time-series change feature sequence of the node subset is segmented according to the preset time window, and the feature change trend and change amplitude in each time window are calculated, so as to realize the refined analysis of the abnormal propagation process. The feature change trend and change amplitude are matched with the alternative abnormal propagation mode in the preset feature library for similarity, which can improve the accuracy of identifying the specific type and development stage of the current risk propagation and reduce the pattern recognition deviation caused by insufficient feature extraction.

[0020] In the second aspect, an embodiment of the present application provides an SCL on-site safety accident analysis and early warning system, which SCL on-site safety accident analysis and early warning system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and one or more processors call the computer instructions to enable the system to execute the method described in the first aspect and any possible implementation method of the first aspect.

[0021] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, comprising instructions, which, when executed on a system, causes the system to execute the method described in the first aspect and any possible implementation of the first aspect.

[0022] In a fourth aspect, an embodiment of the present application provides a computer program product, characterized in that when the computer program product runs on a system, the system executes the method described in any possible implementation manner in the first aspect.

[0023] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. This application provides a SCL on-site safety accident analysis and early warning method, which collects multi-source heterogeneous monitoring data in real time and performs distributed cleaning and feature extraction, obtains a more comprehensive and accurate multi-dimensional time series feature flow, and reduces the impact of data noise and redundant information. The dynamic association network constructed using the multi-dimensional time series feature flow can accurately capture the spatiotemporal coupling relationship between monitoring indicators. By identifying the key influencing factor chain and establishing a semantic parsing model, the analysis of the mechanism and evolution law of safety accidents is realized. The safety threshold system constructed based on the semantic parsing model has the ability to adaptively adjust, which improves the accuracy and timeliness of the early warning. When the system triggers an early warning, the causal analysis performed using the semantic parsing model can quickly locate the root event, and the simulation of the accident development trend combined with the preset evolution model can improve the accuracy of predicting the risk range and type. By matching the corresponding accident early warning scheme in the preset database, the accuracy of the judgment and classification early warning of different types of safety accidents is improved, and the losses caused by safety accidents are reduced.

[0024] 2. This application provides a SCL on-site safety accident analysis and early warning method, which maps the safety threshold system to the safety state space of each monitoring indicator node in the dynamic association network. When the state of the monitoring indicator node is abnormal, the dynamic changes of the safety state can be captured in time through the state label update and information propagation mechanism. The preset propagation threshold is set to determine the propagation range of the abnormal state, and the corresponding key influencing factor chain is marked as an abnormal activation chain, which realizes the precise positioning and tracking of safety risks, improves the system's sensitivity to safety risks, enhances the real-time and accuracy of safety warnings, and makes safety monitoring more intelligent and efficient.

[0025] 3. The present application provides a method for analyzing and warning of SCL on-site safety accidents, constructs a multi-layer deep neural network model to realize the mapping of the safety threshold system to the safety state space of the monitoring indicator nodes in the dynamic association network, so that the threshold boundary corresponds more accurately to the actual monitoring indicator. The process of weighting the threshold features of different monitoring indicators and extracting the association pattern can accurately reflect the importance and mutual influence of different monitoring indicators in safety assessment. The dynamic association network is converted into a low-dimensional feature vector and fused with the threshold feature, which reduces the data dimension, improves the calculation efficiency, and retains the key information in the original network structure. Projecting the fused threshold features into the high-dimensional state space to form a safe state boundary can improve the accuracy of identifying the abnormal state of each monitoring indicator, thereby improving the accuracy and reliability of the safety warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of an SCL on-site safety accident analysis and early warning method in an embodiment of the present application.

[0027] Figure 2 It is another flow chart of an SCL on-site safety accident analysis and early warning method in an embodiment of the present application.

[0028] Figure 3 It is a schematic diagram of the physical device structure of an SCL on-site safety accident analysis and early warning system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to be used as limitations to the present application. As used in the specification and appended claims of the present application, the singular expressions "one", "a kind of", "said", "above", "the" and "this" are intended to also include plural expressions, unless there is a clear indication to the contrary in the context. It should also be understood that the term "and / or" used in the present application refers to any or all possible combinations comprising one or more listed items.

[0030] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as suggesting or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, unless otherwise specified, "plurality" means two or more.

[0031] The following uses an embodiment and combines Figure 1 , a SCL on-site safety accident analysis and early warning method in an embodiment of the present application is described: See also Figure 1 , which is a flow chart of an SCL on-site safety accident analysis and early warning method in an embodiment of the present application.

[0032] S101, collecting multi-source heterogeneous monitoring data of the tunnel in real time, and performing distributed cleaning and feature extraction on the multi-source heterogeneous monitoring data to obtain a multi-dimensional time series feature flow; In this step, the system first collects multi-source heterogeneous monitoring data from the tunnel construction site in real time. Due to the complex tunnel construction environment, there are many types of monitoring indicators involved, such as surrounding rock displacement, support stress, pore water pressure, construction process parameters, etc., and the data formats and sampling frequencies of different indicators are different, thus forming a multi-source heterogeneous feature.

[0033] In order to efficiently process massive monitoring data, the system uses a distributed computing framework to perform parallel cleaning and feature extraction on the raw data. Data cleaning mainly includes preprocessing operations such as deduplication, denoising, and time alignment to improve data quality; feature extraction uses time-frequency analysis, statistical learning and other methods to mine key features that reflect the safety status of the tunnel from the raw data. The system can use big data processing platforms such as MapReduce and Spark to distribute cleaning and feature extraction tasks to multiple nodes for parallel execution, thereby significantly improving computing efficiency.

[0034] After distributed cleaning and feature extraction, the system will obtain a set of high-quality multi-dimensional time series feature flows, each dimension representing the time series feature value of a type of monitoring indicator. However, when the number of monitoring indicators is too large, the dimension of the feature flow will expand dramatically, which may introduce problems such as data sparsity and redundancy. To this end, the system can also use dimensionality reduction algorithms such as principal component analysis (PCA) and independent component analysis (ICA) to compress the dimension of the feature flow while retaining the core information, making the subsequent security status analysis more efficient and reliable.

[0035] S102, constructing a dynamic association network of security status with multi-dimensional time series feature flow; The system constructs a dynamic association network of security status with multi-dimensional time series feature streams. The dynamic association network is used to capture the spatiotemporal coupling relationship between different monitoring indicators. Specifically: the multi-dimensional time series feature stream is input into the pre-trained spatiotemporal graph neural network model. The spatiotemporal graph neural network model captures the spatial correlation between monitoring indicators through graph convolution, and captures the temporal evolution law of monitoring indicators through time convolution; the spatiotemporal coupling relationship between monitoring indicators is obtained by integrating spatial correlation and temporal evolution law, and the spatiotemporal coupling relationship is quantified into a dynamic association weight matrix; a dynamic association network is constructed with monitoring indicators as nodes and dynamic association weight matrices as edges.

[0036] In this step, the system constructs a dynamic correlation network of safety status with multi-dimensional time series feature flows. This network aims to capture the spatiotemporal coupling relationship between different monitoring indicators and reveal the intrinsic connection and evolution law of the tunnel safety status.

[0037] To achieve the above goals, the system first inputs the multi-dimensional time series feature stream into the pre-trained spatiotemporal graph neural network model. The model uses a combination of graph convolution and time convolution to extract the spatial correlation and time series evolution rules between monitoring indicators. Among them, graph convolution updates node features by aggregating node neighborhood information, which can model local associations between indicators; while time convolution uses a sliding window to scan time series data to capture the dynamic change trend of indicators. Through feature learning in both time and space dimensions, the model can comprehensively characterize the complex coupling relationship between monitoring indicators.

[0038] After extracting the spatiotemporal features, the system further integrates the spatial correlation and temporal evolution law to generate a dynamic correlation weight matrix that reflects the overall coupling degree between monitoring indicators. Each element in the matrix represents the correlation strength between two indicators at a certain moment. The larger the weight value, the more significant the impact between the indicators. Since the tunnel environment and construction status are dynamically changing, the correlation strength between indicators will also change over time, so the correlation matrix has dynamic characteristics.

[0039] Finally, the system uses monitoring indicators as nodes and dynamic association weight matrices as edges to build a complete dynamic association network of safety status. In this network, nodes represent various types of monitoring quantities, and edges represent the spatiotemporal coupling relationship between them. The overall structure of the network can intuitively depict the global situation of tunnel safety. When the tunnel status changes abnormally, the association strength between related indicator nodes will fluctuate significantly. By analyzing the dynamic evolution of the association network, safety hazards can be discovered in a timely manner and accident risks can be estimated.

[0040] S103. Identify a key influencing factor chain according to the coupling strength of the spatiotemporal coupling relationship, and establish a semantic parsing model of the security status in combination with the key influencing factor chain; The system identifies the key influencing factor chain according to the coupling strength of the spatiotemporal coupling relationship, and establishes a semantic parsing model of the safety state in combination with the key influencing factor chain. The semantic parsing model is used to reveal the internal mechanism and evolution law of the occurrence of safety accidents. Among them: the key influencing factor chain is identified according to the coupling strength of the spatiotemporal coupling relationship, specifically: the propagation path between different monitoring indicator nodes is calculated based on the dynamic association network; the propagation path whose path length is less than the preset length threshold and the cumulative propagation intensity is greater than the preset intensity threshold is determined as an influencing factor chain; the triggering frequency of each influencing factor chain is counted; the influencing factor chain with a triggering frequency higher than the preset frequency threshold and the corresponding coupling intensity is the largest is determined as the key influencing factor chain.

[0041] In this step, the system identifies the key influencing factor chain according to the coupling strength of the spatiotemporal coupling relationship, and establishes a semantic parsing model of the safety status based on the key influencing factor chain. This model aims to reveal the internal mechanism and evolution law of safety accidents, and provide intelligent support for accident warning and emergency decision-making.

[0042] First, the system calculates the propagation paths between different monitoring indicator nodes based on the dynamic association network, and describes the cascade diffusion process of the impact of safety accidents in the network. Among them, the propagation path represents the possible impact path from one abnormal indicator node to another node, and each edge on the path corresponds to the local coupling relationship between two indicators.

[0043] Next, the system identifies the influencing factor chain based on the key attributes of the propagation path. Specifically, the path in the propagation path with a path length less than the preset threshold and a cumulative propagation strength greater than the preset threshold is determined as an influencing factor chain. The path length reflects the propagation distance of the abnormal impact. The shorter the length, the more direct the association between the two indicators; while the cumulative propagation strength measures the product of the weights of each edge on the path. The greater the strength, the more significant the cascade effect between the indicators. Through the dual constraints of path length and propagation strength, the system can screen out the propagation path that is most likely to trigger a chain reaction, that is, the influencing factor chain.

[0044] Then, the system further counts the triggering frequency of each influencing factor chain, that is, the number of times the factor chain has abnormal propagation in the historical data. The higher the triggering frequency, the more common the failure mode represented by the factor chain is in tunnel construction. The system identifies the influencing factor chain with a triggering frequency higher than the preset threshold and the largest corresponding cumulative propagation intensity as the key influencing factor chain, as a high-risk cause of safety accidents.

[0045] Finally, the system combines the key influencing factor chain to establish a semantic parsing model of the tunnel safety status. The model represents the logical structure and evolution path of the accident in the form of a causal graph. The nodes in the graph correspond to the key influencing factors, and the directed edges characterize the causal dependency between the factors. By integrating engineering expert knowledge and machine learning algorithms, the system can automatically mine the complex semantic associations between factors from a large number of historical cases, and incrementally optimize the causal graph according to prior rules, and continuously improve the causal logic of accident evolution. The causal graph intuitively presents the necessary and sufficient conditions for the occurrence of the accident, and can deeply reveal the formation mechanism of safety accidents.

[0046] Based on the semantic parsing model, the system can dynamically infer the safety status of the tunnel based on real-time monitoring data. When certain key influencing factors are abnormal, the system traces back the impact chain through the causal graph, predicts the possible path of the accident, and assesses the degree of harm. At the same time, the semantic model can also simulate the dynamic evolution of accidents under different scenarios, estimate the cascading propagation range of faults and the scale of losses, and provide decision-making basis for emergency response.

[0047] S104. Construct a security threshold system based on the output of the semantic parsing model; The system builds a safety threshold system based on the output of the semantic parsing model, and the safety threshold system dynamically adjusts the warning threshold through a hierarchical feedback mechanism.

[0048] In this step, the system builds a safety threshold system based on the output of the semantic parsing model. This threshold system is designed to define the normal fluctuation range and critical state of various monitoring indicators, providing a quantitative basis for tunnel safety assessment and early warning triggering.

[0049] Specifically, the semantic parsing model estimates the impact of abnormal changes in different influencing factors on the safety and stability of the tunnel through causal reasoning and scenario simulation. The model outputs a series of safety status indicators, such as surrounding rock displacement rate, support structure stress level, construction progress deviation, etc. These indicators comprehensively evaluate the safety risk level of the tunnel in its current state.

[0050] Based on the above safety status indicators, the system further constructs a multi-level and multi-granular safety threshold system. In this system, multiple warning thresholds are set for each type of monitoring indicator, corresponding to different risk levels and degrees of danger. When the actual monitoring value exceeds a certain threshold, it indicates that the tunnel safety status is abnormal and corresponding preventive or disposal measures need to be taken.

[0051] The threshold system adopts a hierarchical and progressive structure, setting early warning conditions step by step from local to overall, from minor to serious. For example, the threshold levels can be divided according to the hierarchy of "prompt-warning-alarm-accident", corresponding to different stages of deterioration of the safety status. The higher the level, the greater the risk, and the faster the response and disposal are required.

[0052] In addition, the threshold system is divided and customized for different types of work and different construction stages. For example, different displacement rate thresholds are set in the tunnel excavation stage and the support stage to adapt to the dynamic changes in the stability of the surrounding rock; and different threshold standards are adopted for poor geological sections and normal geological sections according to the differences in geological conditions to cope with the uncertainty of geological risks.

[0053] S105, using a security threshold system to evaluate the multi-dimensional time series feature flow; In this step, the system uses the constructed safety threshold system to conduct real-time evaluation of the multi-dimensional time series feature flow, realizing dynamic monitoring and trend prediction of the tunnel safety status.

[0054] First, the system extracts the features of the collected real-time monitoring data stream and converts it into a multi-dimensional time series feature stream corresponding to the threshold system. The feature value of each dimension represents the quantitative level of a certain type of monitoring indicator at the current moment, such as surrounding rock displacement rate, support structure stress, etc.

[0055] Then, the system compares the indicator values ​​of each dimension of the feature flow with the warning threshold of the corresponding indicator in the threshold system to determine the safety status of each indicator. Specifically, the system uses a rule-based classification algorithm to automatically match the feature value with the threshold condition by setting a series of IF-THEN rules, and gives corresponding security level labels, such as safe, prompt, and warning. When multiple indicators exceed the preset threshold at the same time, the system will also comprehensively analyze the weights and combination logic of each indicator, evaluate the overall security situation, and generate an overall security level.

[0056] On the basis of real-time evaluation, the system will also continuously track the changing trend of characteristic flows and predict the dynamic evolution of safety status. Here, time series analysis and trend extrapolation are mainly used to model the high-order statistics such as the rate of change and acceleration of characteristic values, characterize the dynamic change law of indicators, and extrapolate to the future to estimate the development trend of safety status and identify the signs of increasing danger in advance.

[0057] S106. When the multi-dimensional time series feature flow triggers an early warning, a semantic parsing model is used to perform a causal analysis to obtain the root event that triggers the early warning, and the development trend of the root event is simulated based on a preset evolution model; In this step, when the multi-dimensional time series feature flow triggers an early warning, the system will use the semantic parsing model to conduct a causal analysis to find the root event that triggered the early warning, and based on the preset accident evolution model, simulate the development trend of the root event to predict the subsequent direction of the safety risk.

[0058] Specifically, when the feature flow at a certain moment exceeds the safety threshold and triggers the warning condition, the system first determines the abnormal monitoring indicator corresponding to the warning signal and uses the indicator node as the input of the semantic parsing model. Then, through the causal reasoning mechanism, the system traces back the associated propagation path between nodes along the chain of influencing factors until the source node of the warning signal is found, that is, the root event that triggers the warning is identified.

[0059] For example, when the displacement rate of the tunnel vault is monitored to increase suddenly and exceed the warning threshold, the system starts the tracing diagnosis process. The semantic model analyzes the relationship between displacement acceleration and surrounding rock structure stress and finds that stress mutation is the direct cause of displacement abnormality. It further searches for the semantic association between stress index and construction process parameters and finally identifies the root cause of stress mutation: abnormal vibration frequency of blasting construction, finding the source of the warning.

[0060] After locking the root cause, the system will also combine the feedback data from the construction site to evaluate the severity and impact of the current root cause. For example, by analyzing the current vibration frequency exceeding the standard amplitude, duration and other parameters, the level of damage to the stability of the surrounding rock can be determined; at the same time, the spatial location of the vibration event, geological environmental conditions and other factors must be considered to estimate the spatial scope of its disturbance.

[0061] After mastering the basic situation of the root event, the system will further use the preset accident evolution model to simulate the development trend of the event. The accident evolution model is essentially a causal deduction map that describes the typical development path and evolution law of various tunnel accidents. According to the instability mode of the root event, the system matches the most similar evolution path from the map, takes the subsequent nodes in the path as risk scenarios, and predicts the probability and degree of harm of the subsequent accident chain based on probabilistic reasoning.

[0062] For example, for stress mutation caused by abnormal blasting frequency, the system searches for the evolution path of "abnormal blasting frequency-stress mutation-displacement acceleration-vault collapse" in the accident evolution map. Based on this path, the probability of vault collapse under the current working conditions is predicted, as well as quantitative indicators such as the collapse range and the scale of casualties.

[0063] S107. Determine the expected risk range and risk type based on the results of the causal analysis and the development trend of the accident; S108. Match the corresponding accident warning plan in the preset database according to the risk range and risk type, and send the accident warning plan to the monitoring terminal.

[0064] In this step, the system will further determine the expected risk range and risk type based on the causal analysis results and accident development trends of the previous step, providing a clear goal orientation for the subsequent release of early warning information and formulation of emergency plans.

[0065] First, based on the hazard level assessment of the root event and the prediction of the accident evolution trend, the system estimates the spatial scope of the accident impact, which mainly includes graded areas such as danger zone, threat zone, and affected zone. Among them, the danger zone refers to the core area where the accident occurred, with the highest intensity of hazard factors and the most serious damage; the threat zone is the adjacent area of ​​the danger zone, which is not directly affected by the accident, but has a significant secondary disaster risk; the affected area refers to the peripheral area indirectly affected by the accident, with relatively light damage. Through regional division, the system can more accurately determine the key links and priority targets of risk prevention and control.

[0066] For example, in response to the arch collapse accident caused by the abnormal blasting frequency mentioned above, the system defines the tunnel section where the collapse occurs as a danger zone, focusing on the stability of the excavation surface and adjacent surrounding rock; the tunnel section within a certain range upstream and downstream of the danger zone is designated as a threat zone, mainly to prevent secondary collapse caused by the expansion of the collapse; and the outer tunnel section affected by the collapse vibration, rolling stones, etc. is designated as an affected area, and protective measures such as safety alerts and personnel evacuation are required.

[0067] On the basis of spatial scope, the system should further clarify the risk types that may be caused by the accident and comprehensively analyze the multi-dimensional hazards derived from the accident. The determination of risk types is mainly based on the risk classification standard of tunnel engineering, combined with the factors causing the accident, development trend and on-site environmental conditions for comprehensive assessment. Risk types are usually divided into different dimensions such as casualties, equipment damage, construction delays, economic losses, environmental damage, and social impact, and different risk levels are divided according to the degree of harm.

[0068] In the above embodiment, multi-source heterogeneous monitoring data is collected in real time and distributed cleaning and feature extraction are performed to obtain a more comprehensive and accurate multi-dimensional time series feature flow, reducing the impact of data noise and redundant information. The dynamic association network constructed using the multi-dimensional time series feature flow can accurately capture the spatiotemporal coupling relationship between monitoring indicators. By identifying the key influencing factor chain and establishing a semantic parsing model, the analysis of the mechanism and evolution law of safety accidents is realized. The safety threshold system constructed based on the semantic parsing model has the ability to adaptively adjust, which improves the accuracy and timeliness of early warning. When the system triggers an early warning, the causal analysis performed using the semantic parsing model can quickly locate the root event, and the simulation of the accident development trend combined with the preset evolution model can improve the accuracy of predicting the risk range and type. By matching the corresponding accident early warning scheme in the preset database, the accuracy of the judgment and classification early warning of different types of safety accidents is improved, and the losses caused by safety accidents are reduced.

[0069] In the above embodiment, the system constructs a dynamic association network and establishes a semantic parsing model of the safety status through the collection and processing of multi-source heterogeneous monitoring data, and realizes the analysis and early warning of SCL on-site safety accidents. In order to further improve the early warning accuracy and risk identification ability of the system, another implementation method is introduced below, which focuses on how to deeply integrate the safety threshold system with the dynamic association network, and how to accurately predict the risk propagation path through abnormal propagation pattern analysis. Figure 2 , another SCL on-site safety accident analysis and early warning method in the embodiment of the present application is described: See also Figure 2 , is another flow chart of an SCL on-site safety accident analysis and early warning method in an embodiment of the present application.

[0070] S201, mapping the threshold boundary of the security threshold system to the security state space of each monitoring indicator node in the dynamic association network to obtain the security state interval of each monitoring indicator; The system maps the threshold boundary of the safety threshold system to the safety state space of each monitoring indicator node in the dynamic association network to obtain the safety state range of each monitoring indicator. Specifically: a multi-layer deep neural network model is constructed, and the threshold parameters corresponding to the threshold boundary in the safety threshold system are used as input layer data; weights are assigned to the threshold features of different monitoring indicators, and the association patterns between the threshold features are extracted; the dynamic association network is converted into a low-dimensional feature vector and fused with the threshold feature; the fused threshold feature is projected into the high-dimensional state space to form the safety state boundary of each monitoring indicator node.

[0071] The system maps the threshold boundary of the safety threshold system to the safety state space of each monitoring indicator node in the dynamic association network, thereby obtaining the safety state interval of each monitoring indicator. This step actually merges two different data spaces, that is, combining the threshold system representing the safety operation boundary of the system with the dynamic association network reflecting the real-time operation status of the system. By mapping the threshold boundary and the state space, the safety operation status of the current system can be judged, that is, whether each monitoring indicator is within the safety state interval.

[0072] In order to achieve the mapping of threshold boundaries to state space, the system can use a multi-layer deep neural network model. The threshold boundary parameters of each indicator in the safety threshold system are used as the input layer data of the neural network. By allocating weights and extracting association patterns for threshold features of different indicators, the high-dimensional threshold data is compressed into a low-dimensional feature space. At the same time, the dynamic association network is converted into a low-dimensional vector representation through methods such as graph embedding. After the threshold features and the association network features are fused in the low-dimensional space and then mapped to the high-dimensional state space, the safety state boundary of each monitoring indicator node is obtained.

[0073] S202: when the multi-source heterogeneous monitoring data causes the state of any monitoring indicator node to jump out of the corresponding safe state interval, update the state label of the monitoring indicator node, and propagate the updated information of the state label to adjacent monitoring indicator nodes; When the status of any monitoring indicator node exceeds the safe status range, the system will update the node's status label and propagate the updated information to adjacent nodes. The purpose of this step is to update the node status label in time when an abnormality occurs in a local node, and spread the abnormal information through network propagation, so that adjacent nodes can perceive the abnormality and provide information support for the next step of abnormal location and decision-making.

[0074] In order to achieve status label update and abnormal information propagation, the system first needs to monitor the data flow of each monitoring indicator in real time. Once the current value of an indicator is found to be beyond the safe range, the status label update of the corresponding node is immediately triggered, such as switching from normal state to warning state. While updating the node's own state, the system also needs to traverse all neighbor nodes of the node in the associated network and transmit the abnormal signal through the status update message. In this way, the neighboring nodes can also perceive the abnormal conditions of the local node.

[0075] S203: if the number of propagation times of the update information exceeds a preset propagation threshold, the key influencing factor chain corresponding to the update information is marked as an abnormal activation chain; When the number of propagation of update information exceeds the preset threshold, the system will mark the state update link with abnormal propagation as an abnormal activation link. This step can screen out potential key influencing factors in the network, that is, those node links that can cause large-scale abnormal propagation and bring serious security risks.

[0076] In specific implementation, the system can use graph traversal algorithms such as breadth-first search, taking the state update node as the source point, traversing the node sequence of abnormal state diffusion layer by layer, and recording the propagation depth of each update node. When the propagation depth exceeds the preset threshold, the propagation path composed of all the update nodes currently traversed is marked as an abnormal activation chain. It can be seen that by tracking the network propagation process of the state label update and setting the propagation depth threshold, the key node links that cause chain abnormal reactions can be found.

[0077] S204, constructing a risk propagation network, and calculating the attention weight of each node in the risk propagation network; The system first needs to build a global risk propagation network model based on the historical status data of each monitoring indicator node and the interaction relationship between nodes. This model not only reflects the risk status of each node itself, but also describes the risk transmission path and transmission intensity between nodes. On this basis, the system can use algorithms such as graph attention networks to learn the attention weight of each node by aggregating the node's own characteristics and neighbor information. Intuitively speaking, the larger the attention weight, the more important the node is in the network, and the greater the impact on global risk propagation.

[0078] In the process of building a risk propagation network, we may encounter the problem of dynamic evolution of the network structure, which makes it difficult for static models to adapt. To address this problem, the system can use a dynamic graph neural network to learn the dynamic changes of network topology and node status in real time by introducing the time dimension, so that the risk propagation model can be updated adaptively. In addition, for newly added or exited nodes, the system can flexibly add and delete nodes without changing the model structure by dynamically adjusting the adjacency matrix. Through the above measures, the risk propagation network model can depict the changes in the risk propagation situation caused by network evolution in real time, providing a dynamic perspective for anomaly analysis. At the same time, when calculating the attention weight, the system can also adopt an adaptive weight allocation strategy for different nodes, fully considering factors such as the importance of node business and the difference in the impact of risk events, so that the attention weight can more accurately reflect the importance of the node.

[0079] S205, extracting a node subset whose attention weight is greater than a preset weight threshold, and determining the time series change characteristics of each node in the node subset; The system will extract important nodes whose attention weights are greater than the preset threshold and analyze the characteristic trends of these nodes over time. This step is actually based on the node importance calculated in step S204, further screening key nodes and mining the time series characteristics of these nodes in order to more accurately predict future abnormal propagation behaviors.

[0080] To achieve the above purpose, the system first needs to filter the attention weights obtained in step S204 according to the preset weight threshold, extract the nodes whose weights are significantly higher than the threshold, and form a key node subset. Then, for each key node, the system will collect the time series data of the node state changes over a period of time, and extract the multi-dimensional time series characteristics of the node state changes through the time series analysis method, such as the change rate, fluctuation period, mutation amplitude, etc. These time series characteristics describe the evolution law of key nodes at different time scales.

[0081] S206. Determine the abnormal propagation mode according to the time series variation characteristics, and determine the development path and impact range of risk propagation according to the evolution characteristics of the abnormal propagation mode.

[0082] The system determines the abnormal propagation mode based on the time series change characteristics. Specifically: the time series change characteristic sequence of the node subset is segmented according to the preset time window; the characteristic change trend and change amplitude within each preset time window are calculated; the candidate abnormal propagation mode with the greatest similarity to the characteristic change trend and change amplitude in the preset feature library is determined as the abnormal propagation mode. After that, the system determines the development path and impact range of risk propagation based on the evolution characteristics of the abnormal propagation mode.

[0083] The system first divides the time series feature sequence into several feature sub-segments by sliding the time window based on the key node time series features obtained in step S205. Then, the system calculates the feature change trend and amplitude in each sub-segment, and performs similarity matching with the pre-built abnormal propagation pattern feature library. The pattern with the highest similarity is identified as the abnormal propagation pattern of the current network. The core here is to extract a series of typical abnormal propagation patterns from massive historical abnormal events and establish a corresponding feature library. After identifying the current abnormal pattern, the system further analyzes the historical evolution law of the pattern, and predicts the possible development path of the anomaly through trend extrapolation, similar path clustering and other methods, evaluates its impact range, and finally forms risk warning information.

[0084] In the above embodiment, a multi-layer deep neural network model is constructed to realize the mapping of the security threshold system to the security state space of the monitoring indicator nodes in the dynamic association network, so that the threshold boundary corresponds more accurately to the actual monitoring indicator. The process of weighting the threshold features of different monitoring indicators and extracting the association pattern can accurately reflect the importance and mutual influence of different monitoring indicators in the safety assessment. The dynamic association network is converted into a low-dimensional feature vector and fused with the threshold feature, which reduces the data dimension, improves the calculation efficiency, and retains the key information in the original network structure. Projecting the fused threshold feature into the high-dimensional state space to form a safe state boundary can improve the accuracy of identifying the abnormal state of each monitoring indicator, thereby improving the accuracy and reliability of the safety warning.

[0085] The following describes the system in the embodiment of the present invention from the perspective of hardware processing. Figure 3 , which is a schematic diagram of the physical device structure of an SCL on-site safety accident analysis and early warning system provided in an embodiment of the present application.

[0086] It should be noted that Figure 3 The structure of the system shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0087] like Figure 3As shown, the system includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 302 or the program loaded from the storage part 308 to the random access memory (RAM) 303, such as executing the method in the above embodiment. In RAM 303, various programs and data required for system operation are also stored. CPU 301, ROM 302 and RAM 303 are connected to each other through a bus 304. Input / output (I / O) interface 305 is also connected to bus 304.

[0088] The following components are connected to the I / O interface 305: an input section 306 including a camera, an infrared sensor, etc.; an output section 307 including a liquid crystal display (LCD) and a speaker, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. A removable medium 311, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 310 as needed so that a computer program read therefrom is installed into the storage section 308 as needed.

[0089] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 309, and / or installed from a removable medium 311. When the computer program is executed by the central processing unit (CPU) 301, various functions defined in the present invention are performed.

[0090] It should be noted that the computer-readable medium shown in the embodiment of the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable computer program. Such a propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing.

[0091] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Among them, each box in the flowchart or block diagram can represent a module, a program segment, or a part of the code, and the above-mentioned module, program segment, or a part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0092] As another aspect, the present invention further provides a computer-readable storage medium, which may be included in the system described in the above embodiment; or may exist independently without being assembled into the system. The above storage medium carries one or more computer programs, and when the above one or more computer programs are executed by a processor of a system, the system implements the method provided in the above embodiment.

[0093] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

[0094] As used in the above embodiments, the term "when..." may be interpreted as "if..." or "after..." or "in response to determining..." or "in response to detecting...", depending on the context. Similarly, the phrases "upon determining..." or "if (the stated condition or event) is detected" may be interpreted as "if determining..." or "in response to determining..." or "upon detecting (the stated condition or event)" or "in response to detecting (the stated condition or event)", depending on the context.

[0095] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk), etc.

[0096] Those skilled in the art can understand that to implement all or part of the processes in the above-mentioned embodiments, the processes can be completed by computer programs to instruct related hardware, and the programs can be stored in computer-readable storage media. When the programs are executed, they can include the processes of the above-mentioned method embodiments. The aforementioned storage media include: ROM or random access memory RAM, magnetic disk or optical disk and other media that can store program codes.

Claims

1. A SCL on-site safety accident analysis and early warning method, characterized in that: include: Collect multi-source heterogeneous monitoring data of the tunnel in real time, and perform distributed cleaning and feature extraction on the multi-source heterogeneous monitoring data to obtain a multi-dimensional time series feature flow; Constructing a dynamic association network of security status with the multi-dimensional time series feature flow, wherein the dynamic association network is used to capture the spatiotemporal coupling relationship between different monitoring indicators; Identify a key influencing factor chain according to the coupling strength of the spatiotemporal coupling relationship, and establish a semantic parsing model of the safety state in combination with the key influencing factor chain, wherein the semantic parsing model is used to reveal the internal mechanism and evolution law of the occurrence of safety accidents; Building a safety threshold system based on the output of the semantic parsing model, wherein the safety threshold system dynamically adjusts the warning threshold through a hierarchical feedback mechanism; Using the security threshold system to evaluate the multi-dimensional time series feature flow; When the multi-dimensional time series feature flow triggers an early warning, the semantic parsing model is used to perform causal analysis to obtain the root event that triggers the early warning, and the development trend of the root event is simulated based on the preset evolution model; Determine the expected risk range and risk type based on the results of the causal analysis and the development trend of the accident; A corresponding accident warning plan is matched in a preset database according to the risk range and the risk type, and the accident warning plan is sent to a monitoring terminal.

2. The method according to claim 1, characterized in that The dynamic association network of the security status is constructed by using the multi-dimensional time series feature flow, specifically including: Inputting the multidimensional time series feature stream into a pre-trained spatiotemporal graph neural network model, the spatiotemporal graph neural network model captures the spatial correlation between monitoring indicators through graph convolution, and captures the temporal evolution law of the monitoring indicators through time convolution; The spatial correlation and the temporal evolution law are integrated to obtain the spatiotemporal coupling relationship between the monitoring indicators, and the spatiotemporal coupling relationship is quantified into a dynamic correlation weight matrix; A dynamic association network is constructed with the monitoring indicators as nodes and the dynamic association weight matrix as edges.

3. The method according to claim 1, characterized in that The identifying of the key influencing factor chain according to the coupling strength of the spatiotemporal coupling relationship specifically includes: Calculate the propagation paths between the different monitoring indicator nodes based on the dynamic association network; Determine a propagation path whose path length is less than a preset length threshold and whose cumulative propagation strength is greater than a preset strength threshold as an influencing factor chain; Counting the triggering frequency of each of the influencing factor chains; The influencing factor chain whose trigger frequency is higher than the preset frequency threshold and whose corresponding coupling strength is the largest is determined as the key influencing factor chain.

4. The method according to claim 1, characterized in that: After constructing the security threshold system based on the output of the semantic parsing model, the method further includes: Mapping the threshold boundary of the security threshold system to the security state space of each monitoring indicator node in the dynamic association network to obtain the security state interval of each monitoring indicator; When the multi-source heterogeneous monitoring data causes the state of any monitoring indicator node to jump out of the corresponding safe state interval, updating the state label of the monitoring indicator node, and propagating the update information of the state label to adjacent monitoring indicator nodes; If the number of propagation times of the update information exceeds a preset propagation threshold, the key influencing factor chain corresponding to the update information is marked as an abnormal activation chain.

5. The method according to claim 4, characterized in that Mapping the threshold boundary of the security threshold system to the security state space of each monitoring indicator node in the dynamic association network specifically includes: Constructing a multi-layer deep neural network model, using the threshold parameters corresponding to the threshold boundaries in the safety threshold system as input layer data; Assigning weights to the threshold features of different monitoring indicators and extracting correlation patterns between the threshold features; Converting the dynamic association network into a low-dimensional feature vector and fusing it with the threshold feature; The fused threshold features are projected into a high-dimensional state space to form the safety state boundary of each monitoring indicator node.

6. The method according to claim 4 or 5, characterized in that: If the number of propagation times of the update information exceeds a preset propagation threshold, after marking the key influencing factor chain corresponding to the update information as an abnormal activation chain, the method further includes: Constructing a risk propagation network and calculating the attention weight of each node in the risk propagation network; Extracting a node subset whose attention weight is greater than a preset weight threshold, and determining a time series variation feature of each node in the node subset; Determining an abnormal propagation mode according to the time series variation characteristics; According to the evolution characteristics of the abnormal propagation pattern, the development path and impact scope of risk propagation are determined.

7. The method according to claim 6, characterized in that Determining the abnormal propagation mode according to the time series variation characteristics specifically includes: Segmenting the temporal variation feature sequence of the node subset according to a preset time window; Calculating the characteristic change trend and change amplitude within each of the preset time windows; The candidate abnormal propagation mode with the greatest similarity to the characteristic change trend and the change amplitude in the preset characteristic library matching is determined as the abnormal propagation mode.

8. A SCL on-site safety accident analysis and early warning method system, characterized in that: The system comprises: One or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to cause the system to execute the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that: When the instructions are executed on a system, the system is caused to execute the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that When the computer program product is run on a system, the system is caused to execute the method according to any one of claims 1 to 7.

Images