Railway engineering hidden danger identification method based on multi-dimensional feature deconstruction

Abstract

The invention relates to the technical field of railway engineering safety management, in particular to a railway engineering hidden danger recognition method based on multi-dimensional feature deconstruction, which comprises the following steps: performing multi-dimensional analysis on safety risks and determining safety risk types through feature extraction, association analysis and model classification based on historical safety data and multi-modal sensing data; dividing monitoring partitions and collecting multi-modal sensing data; judging an abnormal condition and acquiring key hidden danger parameters and risk thresholds of the core risk area; early warning triggering and risk levels are judged based on sensor feedback data; and adjusting the acquisition frequency or generating early warning information according to the risk level. The method can improve the hidden danger recognition accuracy, reduces the false alarm and the missing alarm, optimizes the data processing efficiency, achieves the precise early warning and dynamic monitoring, and improves the safety management level of railway engineering construction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent safety monitoring and early warning of railway engineering, in particular to a railway engineering hidden danger identification method based on multi-dimensional feature decomposition. BACKGROUND

[0002] The safety management of railway engineering construction sites is gradually moving towards digitization and intelligentization, but the existing safety hidden danger identification method still has certain limitations. On the one hand, most systems rely on manual inspection and experience judgment, the identification dimension is relatively single and subjective, and it is insufficient when covering complex safety risks caused by the interaction of multiple factors such as "people, machines, materials, methods, environment, and measurement"; on the other hand, the safety hidden dangers of different professions (such as subgrade, bridge, tunnel, track, station building, and four-electricity) show diversification and time-varying characteristics, and the traditional static rules or linear model has certain challenges in timely perception and accurate identification of potential risks.

[0003] In addition, the existing hidden danger management system generally lacks an effective closed-loop mechanism. The data connection between hidden danger identification, rectification dispatching and review feedback is not close enough, which can easily lead to faults in the hidden danger disposal process, incomplete rectification, and difficulty in forming a dynamically optimized risk control system. At the same time, the construction of the knowledge base relies on manual input and static maintenance, and it is difficult to learn and evolve according to the actual scene and disposal results, which limits the experience accumulation and knowledge inheritance, affecting the continuous improvement ability of the safety management system.

[0004] In view of the above problems, an intelligent safety management system that can integrate multi-source perception data, has multi-dimensional hidden danger analysis capability, and can realize closed-loop control and knowledge self-evolution is urgently needed to improve the intrinsic safety level of railway engineering construction. SUMMARY

[0005] Therefore, the present application provides a railway engineering hidden danger identification method based on multi-dimensional feature decomposition, which aims to solve the problems of single dimension, reliance on manual experience, lack of intelligent identification ability of multi-professional and multi-factor interaction, and imperfect hidden danger management closed loop and difficulty in self-learning and continuous optimization of the knowledge base in the current technology.

[0006] The present application provides a railway engineering hidden danger identification method based on multi-dimensional feature decomposition, which comprises: determining the safety risk type of the construction area based on the historical safety data of the railway engineering construction site and the corresponding multi-modal perception data; dividing the construction area into several monitoring sub-zones based on the safety risk type, and periodically collecting multi-modal perception data of any monitoring sub-zone through multi-source perception equipment; determine whether an abnormal situation exists in the corresponding monitoring partition based on the multi-modal perception data collected in the target time period, wherein if an abnormal situation exists, multi-modal perception data and sensor feedback data of each monitoring partition in the target time period are obtained; determine a core risk area based on the multi-modal perception data of each monitoring partition, and determine a key hidden danger parameter for the core risk area and a risk threshold corresponding to the key hidden danger parameter; determine whether to trigger a warning based on the risk threshold corresponding to the key hidden danger parameter and the sensor feedback data, and determine the risk level of the core risk area when the warning is triggered, the risk level including a low risk level and a high risk level; determine the warning strategy of the construction area based on the risk level of the core risk area, and generate and send warning information.

[0007] Further, when determining the warning strategy of the construction area based on the risk level of the core risk area, it includes: adjust the collection frequency based on the sensor feedback data of the core risk area; generate and send warning information based on the area identifier of the construction area and the risk threshold of the key hidden danger parameter.

[0008] Further, when determining the safety risk type of the construction area based on the historical safety data and the corresponding multi-modal perception data of the railway engineering construction site, it includes: determine a plurality of key safety parameters of the construction area based on the historical safety data and the corresponding multi-modal perception data; determine a comprehensive correlation feature value based on the correlation between the key safety parameters; determine the safety risk type of the construction area based on the comprehensive correlation feature value and the safety risk classification model.

[0009] Further, when dividing the construction area into a plurality of monitoring partitions based on the safety risk type of the construction area, it includes: determine the corresponding risk feature value based on the safety risk type of the construction area; determine the number of monitoring partitions corresponding to the construction area based on the risk feature value; divide the construction area into a plurality of monitoring partitions based on the number of monitoring partitions.

[0010] Further, when determining whether an abnormal situation exists in the corresponding monitoring partition based on the multi-modal perception data collected in the target time period, it includes: predict the environmental perception data in the future preset time period based on the environmental perception data collected in the target time period to obtain predicted environmental perception data; determining critical environment perception data in a future preset time period based on multi-modal perception data collected in a target time period; determining whether an abnormal situation exists in the corresponding monitoring partition based on a comparison result of the predicted environment perception data and the critical environment perception data.

[0011] Further, when determining the core risk area based on the multi-modal perception data of each monitoring partition, the method comprises: determining a key hidden danger parameter as a safety parameter that does not meet a standard safety parameter range in the multi-modal perception data in each monitoring partition, and determining a risk threshold value corresponding to the key hidden danger parameter based on the standard safety parameter range; determining the standard safety parameter range of the construction area based on historical safety data of the construction area and corresponding multi-modal perception data; determining the core risk area based on a comparison result of the standard safety parameter range and the multi-modal perception data of each monitoring partition; determining the core risk area based on a number of safety parameters that do not meet the standard safety parameter range in the multi-modal perception data in each monitoring partition.

[0012] Further, when determining whether to trigger a warning based on the risk threshold value corresponding to the key hidden danger parameter and the sensor feedback data, and determining a risk level of the core risk area when the warning is triggered, the method comprises: determining a key deviation value based on a comparison result of the risk threshold value corresponding to the key hidden danger parameter and the key hidden danger parameter data; determining a perception deviation value based on the sensor feedback data and a preset feedback standard; determining whether to trigger a warning based on the key deviation value and the perception deviation value.

[0013] Further, when determining a warning strategy of the construction area based on the risk level of the core risk area, the method comprises: if the risk level of the core risk area is a low risk level, adjusting a collection frequency of the collected data based on sensor feedback data of the core risk area; if the risk level of the core risk area is a high risk level, generating and sending a warning information based on a region identifier of the construction area and a risk threshold value of the key hidden danger parameter.

[0014] Further, the multi-modal perception data comprises environment parameters, equipment operating states, and personnel activity information.

[0015] Compared with the prior art, the beneficial effects of the present application are that: by combining historical safety data and construction site multi-modal sensing data, the safety risk types of the construction area can be comprehensively analyzed, so that the hidden danger identification not only depends on artificial experience, but also covers various safety risk factors, thereby improving the comprehensiveness and scientificity of risk identification. Secondly, by dividing the construction area into monitoring sub-zones and periodically collecting multi-modal sensing data of each sub-zone, the method realizes real-time dynamic monitoring of the construction site, can timely discover abnormal conditions and locate the core risk area, and improves the timeliness and accuracy of hidden danger identification. Thirdly, by extracting the key hidden danger parameters of the core risk area and its risk threshold, and combining sensor feedback data for judgment and warning, multi-dimensional risk analysis and hierarchical management can be realized, so that the warning result is more accurate, and the construction management personnel can formulate corresponding disposal strategies according to different risk levels. Finally, based on the risk level of the core risk area, the warning strategy of the construction area is generated and the warning information is sent, realizing the closed-loop management from hidden danger monitoring to warning decision, enhancing the intelligentization, initiative and operability of the construction site safety management, which helps to reduce the probability of safety accidents and improve the overall construction safety level. BRIEF DESCRIPTION OF DRAWINGS

[0016] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are for purposes of illustration only and are not intended to limit the application thereto. Moreover, in the drawings, like reference numerals designate like parts throughout the several views. In the drawings: Figure 1 A flow chart of a railway engineering hidden danger identification method based on multi-dimensional feature decomposition provided by an embodiment of the present application is shown in the figure. Figure 2 A flow chart of a railway engineering hidden danger identification method based on multi-dimensional feature decomposition provided by an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0017] Exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to convey the scope of the present disclosure to those skilled in the art. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0018] As shown in Figures 1-2 some embodiments of the present application, the present embodiment provides a railway engineering hidden danger identification method based on multi-dimensional feature decomposition, comprising: Step S100: Determine the type of safety risk for the construction area based on historical safety data of the railway construction site and corresponding multimodal perception data.

[0019] Specifically, when determining the type of safety risk for a construction area based on historical safety data and corresponding multimodal perception data from railway construction sites, the process includes: determining several key safety parameters for the construction area based on historical safety data and corresponding multimodal perception data; determining comprehensive correlation feature values ​​based on the correlation relationships between the key safety parameters; and determining the type of safety risk for the construction area based on the comprehensive correlation feature values ​​and a safety risk classification model.

[0020] Understandably, by analyzing historical safety data and corresponding multimodal sensing data from railway construction sites, several key safety parameters of the construction area are extracted. These key parameters reflect the safety status of the construction site, including personnel work behavior, equipment operating status, material stacking, environmental conditions, and monitoring data, providing basic data for subsequent risk assessment. Secondly, the method calculates comprehensive correlation feature values ​​by analyzing the correlations between various key safety parameters. This step reveals the inherent connections and interactive effects between multiple factors, reflecting the complexity of the overall safety status of the construction area, thus capturing potential risk characteristics more accurately than single-parameter analysis. Finally, based on the obtained comprehensive correlation feature values, a pre-trained safety risk classification model is used to determine the risk type of the construction area. By classifying the comprehensive features through the model, the construction area can be divided into different safety risk types, achieving scientific identification and classification of potential hazards, and providing quantitative basis for subsequent monitoring, early warning, and management decisions.

[0021] Step S200: Divide the construction area into several monitoring zones based on the type of safety risk, and periodically collect multimodal sensing data of any monitoring zone through multi-source sensing devices.

[0022] Specifically, when dividing a construction area into several monitoring zones based on the type of safety risks in the construction area, the process includes: determining the corresponding risk characteristic value based on the type of safety risks in the construction area; determining the number of monitoring zones corresponding to the construction area based on the risk characteristic value; and dividing the construction area into several monitoring zones based on the number of monitoring zones.

[0023] Specifically, multimodal sensing data includes environmental parameters, equipment operating status, and personnel activity information.

[0024] It can be understood that, by taking the safety risk type of the construction area as the basis, the risk characteristic value corresponding to the construction area is determined by analyzing the historical data and multi-modal perception data of the construction site. The risk characteristic value can quantitatively reflect the potential danger degree and complexity of the construction area, including possible accident types, hidden danger density, and safety management difficulty, etc., providing a scientific and quantitative basis for subsequent monitoring partition division. By introducing the risk characteristic value, the method can convert the safety risk of the construction area from an abstract concept to an operational numerical indicator, making the risk assessment process more objective and repeatable. Secondly, the number of monitoring partitions is determined based on the risk characteristic value of the construction area. High-risk areas have more potential safety hazards or complex risk factors, and can be divided into more monitoring partitions to ensure fine and comprehensive monitoring of the construction site. Relatively low-risk areas are divided into fewer monitoring partitions to optimize the use efficiency of monitoring resources. Through this dynamic partitioning strategy, not only is the differentiated management of construction areas with different risk levels achieved, but also the monitoring equipment and data processing resources are reasonably allocated under the premise of ensuring monitoring accuracy, thereby improving the overall monitoring efficiency. Thirdly, the construction area is actually divided according to the determined number of monitoring partitions, forming several monitoring partitions that can be independently managed and monitored. Each monitoring partition is equipped with multi-source perception equipment to periodically collect multi-modal perception data, including environmental parameters (such as temperature, humidity, wind speed, air quality, etc.), equipment operating status (such as mechanical operation, abnormal alarm signals, power consumption, etc.), and personnel activity information (such as work behavior, location trajectory, work posture, etc.). These multi-modal data not only comprehensively reflect the real-time safety status of each partition, but also reveal the development trend of potential hazards, providing multi-dimensional reference for abnormal situation determination. Finally, by comprehensively analyzing the environmental data, equipment status and personnel activity information, the method can establish a dynamic risk portrait of the monitoring partition, realizing continuous and safe perception of the construction site. This dynamic monitoring mechanism not only can timely discover potential hazards and provide early warning, but also can provide a reliable data basis for subsequent core risk area identification, hidden danger parameter extraction, and risk level determination.

[0025] It can be seen that, through the partition management, multi-modal perception data collection and prediction comparison mechanism, real-time, dynamic and fine safety monitoring of the construction area is realized, providing basic technical support for subsequent core risk area identification and early warning.

[0026] Step S300, based on the multi-modal perception data collected in the target time period, determine whether there is an abnormal situation in the corresponding monitoring partition, if there is, obtain the multi-modal perception data and sensor feedback data of each monitoring partition in the target time period.

[0027] Specifically, when determining whether an abnormal situation exists in the corresponding monitoring partition based on the multi-modal perception data collected in the target time period, the following steps are included: predicting the environmental perception data in a future preset time period based on the environmental perception data collected in the target time period to obtain predicted environmental perception data; determining critical environmental perception data in the future preset time period based on the multi-modal perception data collected in the target time period; and determining whether an abnormal situation exists in the corresponding monitoring partition based on a comparison result of the predicted environmental perception data and the critical environmental perception data.

[0028] It can be understood that by analyzing the multi-modal perception data collected in the target time period, real-time evaluation of the safety state of the monitoring partition is realized. The multi-modal perception data includes environmental parameters, equipment operating states, and personnel activity information, etc., which can comprehensively reflect the overall safety situation of the monitoring partition. By centralized processing and feature extraction of these data, the system can obtain dynamic safety indicators of each monitoring partition, providing basic data support for abnormal situation determination. Secondly, for the safety situation in the future preset time period, a prediction mechanism is introduced. Based on the environmental perception data collected in the current time period, the system constructs a prediction model to calculate the future environmental perception data, obtaining predicted environmental perception data. This prediction process can capture potential abnormal trends in advance, enabling the system to have forward-looking risk identification capability, rather than relying solely on static judgment of the current state. At the same time, by analyzing the multi-modal perception data in the target time period, critical environmental perception data in the future preset time period, i.e., safety threshold or risk tolerance range, is determined. The critical data can be set according to construction specifications, historical accident data, and professional experience, reflecting the safety limit of each monitoring partition under different conditions. Finally, by comparing the predicted environmental perception data with the critical environmental perception data, the system can determine whether an abnormal situation exists in each monitoring partition. When the predicted data exceeds the critical value, the system identifies potential abnormal partitions and can trigger an alarm or further data collection and processing. Through this combination of prediction analysis and threshold determination, the method not only enables timely discovery of current abnormalities, but also dynamically warns of potential risks in the future, thereby improving the initiative and accuracy of construction site safety management.

[0029] Specifically, the predicted environmental perception data can use time series prediction models, such as: LSTM (Long Short-Term Memory Network): suitable for capturing the time dependence of environmental data (such as temperature, humidity, vibration, etc.). Prophet (Facebook open source prediction model): suitable for predicting environmental data with seasonal and holiday effects. ARIMA / SARIMA: suitable for environmental variables with strong linear trend and periodicity.

[0030] Specifically, the critical environmental perception data determination can employ statistical methods or physical modeling methods, for example: Extreme Value Theory (EVT): used to determine critical values under extreme environmental conditions. Monte Carlo simulation: simulate a variety of combinations of environmental variables to find boundary conditions that may cause structural abnormalities. Physics-based finite element analysis (FEA): calculate the response limit of the structure according to material properties and load conditions.

[0031] It can be seen that through multi-modal data acquisition, predictive analysis and threshold comparison, real-time identification and prospective early warning of abnormal conditions in monitoring partitions are realized, providing a reliable technical foundation for subsequent core risk area identification and safety management decisions.

[0032] Step S400, determine the core risk area based on the multi-modal perception data of each monitoring partition, and determine the key hidden danger parameters corresponding to the key hidden danger parameters and the risk threshold value of the core risk area.

[0033] Specifically, when determining the core risk area based on the multi-modal perception data of each monitoring partition, the safety parameters in each monitoring partition that do not meet the standard safety parameter range are determined as the key hidden danger parameters, and the risk threshold value corresponding to the key hidden danger parameters is determined based on the standard safety parameter range; determine the standard safety parameter range of the construction area based on the historical safety data and the corresponding multi-modal perception data of the construction area; determine the core risk area based on the comparison result of the standard safety parameter range and the multi-modal perception data of each monitoring partition; determine the core risk area based on the number of safety parameters in each monitoring partition that do not meet the standard safety parameter range.

[0034] Specifically, the core risk area identification can employ: Weighted scoring model: Weighted scoring of parameter deviation for each monitoring partition; Decision tree / random forest classifier: train the model to identify high-risk areas; GIS spatial analysis + heat map visualization: combine geographic information for risk area visualization.

[0035] Specifically, the standard safety parameter range determination can employ: Clustering analysis (such as K-means, DBSCAN): identify parameter distribution under normal conditions; Principal component analysis (PCA): extract the main variation direction of multi-modal data to assist in determining the normal range; Statistical methods based on historical data: such as mean ± 3σ as the safety range.

[0036] It can be understood that by analyzing the multi-modal perception data collected in each monitoring partition, potential risk points existing in the construction site are identified. Specifically, the safety parameters in the monitoring partition that do not meet the standard safety parameter range are identified as key hidden danger parameters. The safety parameters can include environmental conditions (such as temperature, humidity, wind speed), equipment operating status (such as load, vibration, abnormal alarm), and personnel behavior (such as operation posture, position deviation, etc.), which can reflect the actual safety status of the construction site. Secondly, based on the historical safety data and multi-modal perception data of the construction area, the standard safety parameter range of the construction area is constructed. The standard parameter range can quantify the reasonable fluctuation interval of each safety index of the construction site, providing a reference benchmark for risk judgment. By comparing the real-time collected data with the standard parameter range, the safety parameters deviating from the normal range can be accurately identified, thereby realizing the preliminary screening of potential hidden dangers. Thirdly, according to the comparison result and the number of deviated standard parameters, the core risk area is determined. The more key hidden danger parameters deviating from the standard range, the higher the safety risk existing in the monitoring partition, so it can be designated as a core risk area. This process not only considers the abnormal situation of a single safety parameter, but also comprehensively evaluates the regional risk by counting the number of abnormal parameters, making the identification of the core risk area more comprehensive and scientific. Finally, for the core risk area, the method further determines the risk threshold value corresponding to the key hidden danger parameter. The risk threshold value is set based on the standard safety parameter range and can be used to judge the severity of the hidden danger and the risk level. By matching the key hidden danger parameters and the risk threshold value, the system can quantify the safety risk and provide a reliable basis for subsequent early warning strategy formulation and safety management decision-making.

[0037] It can be seen that by combining historical data with real-time multi-modal perception data, using standard safety parameter range and abnormal parameter quantity comprehensive evaluation, accurately identifying core risk area and determining key hidden danger parameters and their risk threshold value, providing technical support for the scientific, fine and intelligent of construction site safety management.

[0038] Step S500, based on the risk threshold value corresponding to the key hidden danger parameter and the sensor feedback data, determine whether to trigger an early warning, and determine the risk level of the core risk area when the early warning is triggered, the risk level includes low risk level and high risk level.

[0039] Specifically, based on the risk threshold value corresponding to the key hidden danger parameter and the sensor feedback data, determine whether to trigger an early warning, and determine the risk level of the core risk area when the early warning is triggered, including: determining a key deviation value based on the risk threshold value corresponding to the key hidden danger parameter and the comparison result of the key hidden danger parameter data; determining a perception deviation value based on the sensor feedback data and a preset feedback standard; determining whether to trigger an early warning based on the key deviation value and the perception deviation value.

[0040] Specifically, the key deviation value calculation: absolute deviation: | current value - risk threshold |, standardized deviation: (current value - mean value) / standard deviation, dynamic deviation: update the deviation value according to the time window sliding Specifically, the perception deviation value calculation, combined with the confidence, error range and other indicators of sensor feedback, is weighted and calculated; sensor consistency analysis is introduced (such as too large difference between multiple sensor data is determined as abnormal) Specifically, the early warning trigger logic can adopt: rule engine: such as key deviation > threshold A and perception deviation < threshold B, trigger early warning; fuzzy logic controller: realize soft threshold control; Bayesian network: comprehensive multiple risk factors for probability reasoning.

[0041] It can be understood that through the joint analysis of key hidden danger parameters and sensor feedback data, the real-time early warning trigger and risk level division of core risk area are realized. This principle not only combines the accuracy of traditional threshold judgment and statistical deviation analysis, but also introduces the intelligent inference ability of fuzzy logic and Bayesian reasoning, so as to build a high-precision, high-robustness risk judgment system.

[0042] Firstly, in the key deviation value calculation model, the system takes the risk threshold as the core reference and constructs a multi-level deviation calculation system including absolute deviation, standardized deviation and dynamic deviation. The calculation formula of absolute deviation is |current value-risk threshold|, which is used to measure the direct deviation of hidden danger parameters relative to the preset safety boundary and is the first layer sensitive index of risk identification; the standardized deviation is normalized by (current value-average value) / standard deviation, which realizes the unified dimension between different parameter dimensions, so that the system can perform horizontal comparability analysis between multi-dimensional hidden danger parameters; the dynamic deviation is continuously updated based on the time window sliding mechanism, which reflects the time dynamic characteristics of hidden danger evolution, so as to identify potential slow-changing risks and periodic fluctuations. The introduction of this deviation calculation system enables the system not only to detect single abnormal points, but also to perceive the formation trend and cumulative effect of hidden dangers in advance. Secondly, in the perceived deviation value calculation model, the system considers factors such as the confidence of the sensor, the error range and environmental interference, and calculates the overall credibility of the sensor perception data through a weighted fusion algorithm. This model can adaptively adjust the weight of different sensor data, and preferentially adopt the perception results with high confidence and small error. In addition, in order to prevent individual sensors from causing misjudgment due to failure or noise, the system introduces a sensor consistency analysis mechanism: when the feedback of multiple sensors on the same parameter differs by more than a set threshold, the system automatically identifies it as an abnormal perception state and sends a data verification request, thereby improving the stability and accuracy of the overall judgment. In the early warning trigger logic layer, the system adopts a hierarchical decision-making architecture, combining rule engine, fuzzy logic controller and Bayesian network model, to realize a gradual risk judgment process from deterministic logic to probabilistic reasoning. The rule engine layer mainly handles explicit logical conditions, such as "trigger early warning when key deviation value is greater than threshold A and perceived deviation is less than threshold B", which is used to quickly identify explicit risk events; the fuzzy logic controller layer introduces a soft threshold control mechanism, mapping risk levels from low to high as a continuous interval, which is suitable for handling uncertain or transitional risk situations, ensuring smooth transition rather than sudden triggering of the early warning process; the Bayesian network layer further performs probabilistic reasoning based on the joint distribution of multiple risk factors, combining key parameter deviation, sensor confidence and historical risk prior information to calculate the posterior probability of risk level, thereby achieving high-confidence early warning decision-making under complex multi-source data conditions. Through the coordinated operation of the above multi-level logic, the system can realize the whole-process closed-loop control from risk identification, deviation calculation, perception fusion to intelligent reasoning and decision-making. When the core risk area is judged to be a low risk level, the system can automatically adjust the acquisition frequency and monitoring intensity; when it is judged to be a high risk level, the system automatically triggers the early warning and response mechanism, sends alarm information to the management end and starts the emergency disposal process.

[0043] In step S600, a pre-warning strategy of the construction area is determined based on the risk level of the core risk area, and pre-warning information is generated and sent.

[0044] Specifically, when determining the pre-warning strategy of the construction area based on the risk level of the core risk area, if the risk level of the core risk area is a low risk level, the collection frequency of the collected data is adjusted based on the sensor feedback data of the core risk area; if the risk level of the core risk area is a high risk level, pre-warning information is generated and sent based on the area identifier of the construction area and the risk threshold of the key hidden danger parameter.

[0045] Specifically, when adjusting the collection frequency of the collected data based on the sensor feedback data of the core risk area, the adjustment coefficient is determined based on the key deviation value and the perception deviation value; and the adjusted collection frequency is determined based on the adjustment coefficient and the collected collection frequency.

[0046] It can be understood that the pre-warning strategy of the construction area is formulated based on the risk level of the core risk area, and dynamic and safe management is achieved. For the core risk area with a low risk level, the frequency of periodic data collection is intelligently adjusted according to the sensor feedback data. By analyzing the key deviation value and the perception deviation value, the adjustment coefficient is calculated, and the collection frequency is adjusted accordingly, which can optimize the monitoring resources and improve the operation efficiency and data processing capacity on the premise of ensuring the safety monitoring accuracy. Secondly, for the core risk area with a high risk level, specific pre-warning information is generated and sent. The pre-warning information content includes key information such as the area identifier of the construction area, the key hidden danger parameter and its risk threshold, etc., which provides clear safety prompts and countermeasures for the on-site management personnel. Through this mechanism, the high-risk area can be identified and responded in time, and the proactive prevention and control of potential accidents is realized. Thirdly, through the dynamic management strategy driven by the risk level. The low-risk area realizes resource optimization and continuous monitoring by adjusting the collection frequency, and the high-risk area realizes rapid response and decision support by generating pre-warning information. This strategy takes into account the safety and monitoring efficiency of the construction site, so that the monitoring resources can be dynamically allocated according to the risk level, and hierarchical management and intelligent pre-warning are realized. Finally, through the above mechanism, the safety management of the construction site realizes the closed-loop management from risk identification, core area positioning, risk level determination to pre-warning strategy execution. This method can effectively improve the safety monitoring accuracy, response speed and management efficiency of the construction site, and provides scientific, intelligent and operable safety management technical support for railway engineering construction.

[0047] In order to better enable those skilled in the art to fully understand and implement the present application, the specific implementation principles of the present application are further supplemented in the following in combination with a specific application scenario.

[0048] In a railway engineering construction site, the construction area is divided into multiple monitoring sub-zones, each equipped with multi-source sensing devices for collecting multi-modal sensing data. These devices are connected to the data processing center through wired or wireless means, ensuring real-time and accuracy of data. For example, in a certain tunnel construction scene, the construction area includes the tunnel entrance, the middle of the tunnel, and the tunnel exit. According to the division strategy, after calculating the corresponding risk feature values based on the safety risk type, the middle of the tunnel is determined as a high-risk area due to complex geological conditions and frequent construction activities, so the number of monitoring sub-zones in this area is more intensive than the tunnel entrance and exit.

[0049] First, the safety risk type of the construction area is determined. In this tunnel construction scenario, key safety parameters extracted from historical safety data include temperature, humidity, vibration frequency, and noise level indicators. By analyzing the correlation between these parameters, a comprehensive correlation feature value is generated. For example, when the humidity in the middle of the tunnel increases, the probability of rock mass stability decreasing increases, and this correlation is quantified as a comprehensive correlation feature value through a mathematical model. Then, the safety risk classification model is used to classify the comprehensive correlation feature value, and finally determine the overall safety risk type of the construction area as "medium-high risk".

[0050] Next, based on the division strategy, the construction area is divided into several monitoring sub-zones. The middle of the tunnel, as a high-risk area, has significantly more monitoring sub-zones than the tunnel entrance and exit. The boundaries of the monitoring sub-zones are presented in the form of virtual grids, ensuring that data collection within each sub-zone is independent and complete. For example, every 50 meters in the middle of the tunnel sets up a monitoring sub-zone, while every 100 meters in the tunnel entrance and exit sets up a monitoring sub-zone.

[0051] In the method, it is determined whether an abnormal situation exists in the monitoring sub-zone. For example, in a certain time period, the temperature in a certain monitoring sub-zone in the middle of the tunnel continues to rise. The system predicts the temperature in the future preset time period based on the environmental sensing data collected in the target time period, generating predicted environmental sensing data. If the prediction result shows that the temperature will continue to rise, the critical environmental sensing data in the future preset time period, i.e., the safety upper limit of the temperature, is further determined as 35°C. Compare the predicted environmental sensing data with the critical environmental sensing data, if the predicted value exceeds 35°C, it is determined that the monitoring sub-zone has an abnormal situation.

[0052] The above-mentioned target is to determine the core risk area and its key hidden danger parameters and risk threshold. In this tunnel construction scene, based on historical safety data and multi-modal perception data, the standard range of temperature is 15-30°C, and the standard range of humidity is 40-70%. After comparing the multi-modal perception data of each monitoring partition with the standard safety parameter range, it is found that the temperature of a certain monitoring partition in the middle of the tunnel exceeds 30°C and the humidity is less than 40%, so it is determined as the core risk area. At the same time, temperature and humidity are determined as the key hidden danger parameters, and their risk thresholds are set to 35°C and 35% respectively.

[0053] In the embodiment, the trigger condition of the early warning mechanism is further refined. For example, if the temperature in the core risk area is 36°C and its risk threshold is 35°C, the key deviation value is 1°C. At the same time, the perception deviation value is determined based on the sensor feedback data and the preset feedback standard. If both of them exceed the preset threshold, it is determined to trigger the early warning, and the specific early warning strategy is determined according to the risk level of the core risk area.

[0054] Finally, in the embodiment, different early warning strategies are formulated based on the risk level of the core risk area. If the risk level of the core risk area is low risk level, the acquisition frequency of the acquisition is adjusted based on the sensor feedback data. For example, if the temperature in a certain monitoring partition is close to but does not exceed the risk threshold, the acquisition frequency is appropriately increased to discover potential risks in time. If the risk level of the core risk area is high risk level, early warning information is generated and sent based on the region identifier of the construction area and the risk threshold of the key hidden danger parameter. For example, if the temperature in a certain monitoring partition in the middle of the tunnel exceeds 35°C and continues to rise, early warning information containing temperature exceeding standard information is generated and sent to relevant personnel through the communication network.

[0055] In the whole implementation process, the division of the construction area, the distribution of the monitoring partition and the positioning of the core risk area all need to be operated by the schematic diagram. The acquisition and transmission of multi-modal perception data and sensor feedback data rely on multi-source perception devices installed in the monitoring partition. These devices are connected with the data processing center through wired or wireless way, ensuring the real-time and accuracy of the data. The determination of key hidden danger parameters and risk thresholds is based on the comparison results of standard safety parameter range and actual data, ensuring the scientificity and reliability of hidden danger identification. The generation and sending of early warning information are realized through the communication module, ensuring that relevant personnel can receive early warning information in time and take corresponding measures.

[0056] The above scenario is only a preferred embodiment of the present application and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0057] In the above embodiments, by combining historical safety data and multi-modal sensing data of the construction site, the safety risk types of the construction area can be comprehensively analyzed, so that the hidden danger identification not only depends on artificial experience, but also covers various safety risk factors, thereby improving the comprehensiveness and scientificity of risk identification. Secondly, by dividing the construction area into monitoring sub-zones and periodically collecting multi-modal sensing data of each sub-zone, the method realizes real-time dynamic monitoring of the construction site, can timely discover abnormal conditions and locate the core risk area, and improves the timeliness and accuracy of hidden danger identification. Thirdly, by extracting the key hidden danger parameters of the core risk area and its risk threshold, and combining sensor feedback data for judgment and warning, multi-dimensional risk analysis and hierarchical management can be realized, so that the warning result is more accurate, and the construction management personnel can formulate corresponding disposal strategies according to different risk levels. Finally, the warning strategy of the construction area is generated based on the risk level of the core risk area, and the warning information is sent, realizing the closed-loop management from hidden danger monitoring to warning decision, enhancing the intelligentization, initiative and operability of the construction site safety management, which helps to reduce the probability of safety accidents and improve the overall construction safety level.

[0058] Those skilled in the art will appreciate that embodiments of the application can be provided as methods, systems or computer program products. Accordingly, the application can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the application can be embodied in the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk memory, CD-ROMs, optical storage media, etc.) embodying computer usable program code.

[0059] The application is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing apparatus produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 An apparatus for performing each flow or a plurality of flows and / or blocks. Figure 1 An apparatus for performing each flow or a plurality of flows and / or blocks.

[0060] These computer program instructions can also be stored in a computer-readable memory that can direct the computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction apparatus, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1one or more processes and / or blocks Figure 1 the function specified in the one or more blocks.

[0061] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operation steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, so that the instructions executed on the computer or other programmable data processing devices provide processes for implementing the flow Figure 1 one or more processes and / or blocks Figure 1 the steps of the function specified in the one or more blocks.

[0062] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, but not to limit it, although the above embodiments of the present application have been described in detail, those skilled in the art should understand: the specific embodiments of the present application can be modified or replaced by the same, without departing from the spirit and scope of the present application, any modification or equivalent replacement, which should be covered within the scope of protection of the claims of the present application.

Claims

1. A method for identifying potential hazards in railway engineering based on multidimensional feature deconstruction, characterized in that, include: Based on historical safety data from railway construction sites and corresponding multimodal perception data, the types of safety risks for construction areas are determined using machine learning models such as random forest or LSTM. The construction area is divided into several monitoring zones based on the type of safety risk, and multimodal sensing data of any monitoring zone is periodically collected through multi-source sensing devices. Based on the multimodal sensing data collected within the target time period, it is determined whether there are any abnormalities in the corresponding monitoring zones. If there are, the multimodal sensing data and sensor feedback data of each monitoring zone within the target time period are obtained. Based on the multimodal sensing data of each monitoring zone, the core risk areas are identified, and the key hidden danger parameters for the core risk areas and the corresponding risk thresholds are determined. Based on the risk thresholds corresponding to key hazard parameters and sensor feedback data, it is determined whether to trigger an early warning, and when an early warning is triggered, the risk level of the core risk area is determined, including low risk level and high risk level. Based on the risk level of the core risk area, determine the early warning strategy for the construction area and generate and send early warning information.

2. The railway engineering hidden danger identification method based on multi-dimensional feature deconstruction as described in claim 1, characterized in that, When determining early warning strategies for construction areas based on the risk levels of core risk areas, the following should be included: The acquisition frequency is adjusted based on sensor feedback data from core risk areas; Early warning information is generated and sent based on the area identification of the construction area and the risk threshold of key hidden danger parameters.

3. The railway engineering hidden danger identification method based on multi-dimensional feature deconstruction as described in claim 1, characterized in that, When determining the type of safety risk for a construction area based on historical safety data from railway construction sites and corresponding multimodal sensing data, the following are included: Several key safety parameters for the construction area were determined based on historical safety data and corresponding multimodal sensing data. Based on the correlation between key safety parameters, multidimensional feature fusion indicators are determined by principal component analysis (PCA) or correlation calculation. The types of safety risks for the construction area are determined based on comprehensive correlation feature values ​​and a safety risk classification model.

4. The railway engineering hidden danger identification method based on multi-dimensional feature deconstruction as described in claim 1, characterized in that, When dividing a construction area into several monitoring zones based on the type of safety risks in the construction area, the following are included: Determine the corresponding risk characteristic values ​​based on the type of safety risks in the construction area; The number of monitoring zones corresponding to the construction area is determined based on risk characteristic values; The construction area is divided into several monitoring zones based on the number of monitoring zones.

5. The railway engineering hidden danger identification method based on multi-dimensional feature deconstruction as described in claim 1, characterized in that, When determining whether there are anomalies in the corresponding monitoring zone based on multimodal sensing data collected within the target time period, the following are included: Based on the environmental perception data collected within the target time period, the environmental perception data within the future preset time period is predicted to obtain the predicted environmental perception data. Based on the multimodal sensing data collected within the target time period, determine the critical environmental sensing data within the future preset time period; The comparison results between the predicted environmental sensing data and the critical environmental sensing data are used to determine whether there are any abnormalities in the corresponding monitoring zones.

6. The railway engineering hidden danger identification method based on multi-dimensional feature deconstruction as described in claim 5, characterized in that, When determining core risk areas based on multimodal sensing data from each monitoring zone, the following are included: Safety parameters in multimodal sensing data within each monitoring zone that do not conform to the standard safety parameter range are identified as key hidden danger parameters, and risk thresholds corresponding to key hidden danger parameters are determined based on the standard safety parameter range. The standard safety parameter range for the construction area is determined based on historical safety data and corresponding multimodal sensing data of the construction area. The core risk areas were determined based on the comparison results of standard safety parameter ranges and multimodal sensing data from each monitoring zone. The core risk areas are determined based on the number of safety parameters in each monitoring zone that do not conform to the standard safety parameter range.

7. The railway engineering hidden danger identification method based on multi-dimensional feature deconstruction as described in claim 5, characterized in that, When determining whether to trigger an early warning based on the risk threshold corresponding to key hazard parameters and sensor feedback data, and when determining the risk level of the core risk area upon triggering an early warning, the following should be included: The key deviation value is determined based on the risk threshold corresponding to the key hazard parameters and the comparison results of the key hazard parameter data. The perception deviation value is determined based on sensor feedback data and preset feedback standards; Whether to trigger an alert is determined based on key deviation values ​​and perceived deviation values.

8. The railway engineering hidden danger identification method based on multi-dimensional feature deconstruction as described in claim 1, characterized in that, When determining early warning strategies for construction areas based on the risk levels of core risk areas, the following should be included: If the risk level of the core risk area is low, the collection frequency will be adjusted based on the sensor feedback data of the core risk area. If the risk level of the core risk area is high, then an early warning message will be generated and sent based on the area identifier of the construction area and the risk threshold of key hidden danger parameters.

9. The railway engineering hidden danger identification method based on multi-dimensional feature deconstruction according to claim 2, characterized in that, When adjusting the acquisition frequency based on sensor feedback data from core risk areas, the following applies: The adjustment coefficient is determined based on the key deviation value and the perceived deviation value; The adjusted acquisition frequency is determined based on the adjustment coefficient and the acquisition frequency.

10. The railway engineering hidden danger identification method based on multi-dimensional feature deconstruction as described in claim 1, characterized in that, Multimodal sensing data includes environmental parameters, equipment operating status, and personnel activity information.

Images