Railway infrastructure monitoring and safety level evaluation method based on cross-service data fusion of engineering power supply

Abstract

This invention discloses a method for monitoring and assessing the safety level of railway infrastructure based on cross-business data fusion of engineering and power supply. It constructs a two-layer quantitative indicator system covering both objective conditions and external environmental risks; employs a bidirectional cross-attention mechanism to achieve deep semantic feature fusion of engineering, power supply, and environmental data; collaboratively identifies the status of key components and outputs indicator values ​​based on a shared backbone network and multiple semantic detection heads; utilizes a combination of the analytic hierarchy process (AHP) and entropy weighting method for weighting, and dynamically adjusts the weights using a context-aware mechanism to calculate a comprehensive safety evaluation value, thereby classifying safety levels and triggering early warnings; ultimately achieving hierarchical processing from millisecond-level instinctive reflexes to long-cycle knowledge evolution. This invention realizes deep semantic fusion and coupled feature extraction of cross-business data, establishes a dynamically adaptive safety assessment model, and completes hierarchical intelligent early warning and response, improving the intelligent level of railway infrastructure operation and maintenance and enhancing traffic safety assurance capabilities.

Description

Technical Field

[0001] This invention belongs to the field of railway intelligent monitoring and safety assessment technology, and in particular relates to a method for monitoring and assessing the safety level of railway infrastructure based on cross-business data fusion of engineering and power supply. Background Technology

[0002] Currently, safety monitoring of railway infrastructure mainly relies on separate monitoring systems for track maintenance, signaling, and environmental aspects. While these systems have achieved basic automation to some extent, their inherent technical architecture has gradually revealed three major systemic flaws in practice. First, each business system uses independent data standards and acquisition protocols, resulting in structural barriers between track geometry data from track maintenance, equipment status data from signaling, and multi-source heterogeneous data from the environment, such as wind speed and icing, leading to severe information silos. This data-level fragmentation makes it difficult to conduct cross-business collaborative analysis and data verification when comprehensively assessing the safety status of the entire line, greatly restricting the overall and forward-looking nature of operation and maintenance decisions.

[0003] Secondly, existing detection algorithms, especially image recognition-based feature extraction methods, exhibit significantly reduced adaptability and stability when facing complex environmental conditions such as extreme cold, strong winds, rain, and snow. For example, under strong wind interference, the visual measurement accuracy of dynamic swaying of the overhead contact line is greatly reduced; under icing conditions, the identification of surface defects on components is prone to misjudgment. This directly leads to unreliable perception results of the critical infrastructure status, creating hidden dangers for subsequent risk assessments.

[0004] Finally, and most critically, the existing security assessment mechanisms suffer from a significant drawback. Currently prevalent methods rely excessively on pre-set static thresholds and fixed-weight indicators. This rigid assessment model fails to capture the dynamic evolution of infrastructure conditions and, more importantly, fails to effectively quantify the coupling effects between multiple system risk sources. For example, during strong winds, the dynamic response risk of the overhead contact system should increase dramatically, but the static assessment model cannot automatically and promptly adjust the weights of related indicators, resulting in assessment results that lag far behind actual risk changes. This limits early warning effectiveness, making it difficult to issue accurate and timely signals before risks truly escalate. Solving these problems presents significant challenges, stemming from the need to simultaneously overcome three technical hurdles: deep semantic fusion of cross-protocol data, robust and accurate intelligent sensing in complex environments, and the establishment of dynamic assessment models capable of responding in real-time to changes in internal and external conditions. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a method for monitoring and assessing the safety level of railway infrastructure based on cross-business data fusion of engineering and power supply, thereby resolving the issues existing in the prior art.

[0006] Firstly, to achieve the above objectives, this invention provides a method for monitoring and assessing the safety level of railway infrastructure based on cross-business data fusion of engineering and power supply, comprising the following steps: S1. Collect multi-source data across engineering and power supply services, and perform spatiotemporal alignment and quality repair processing on the multi-source data, which includes engineering data, power data and environmental data. S2. Input the processed multi-source data into the feature fusion model built on the bidirectional cross-attention mechanism to perform deep semantic fusion across business data and generate a unified fusion feature vector. S3. Input the fused feature vector into the shared backbone network to extract general semantic features, and use multiple semantic detection heads to perform parallel decoding of the general semantic features to identify the status of key infrastructure components and output quantitative indicator values ​​corresponding to preset objective indicators and environmental indicators. S4. The quantitative index values ​​are normalized, and weights are assigned by a combination of the analytic hierarchy process and the entropy weight method. The index weights are dynamically adjusted by combining the scenario awareness mechanism to calculate the comprehensive safety evaluation value. S5. Based on the preset threshold range where the comprehensive security evaluation value is located, classify the security level and trigger the corresponding early warning response; wherein, the process from S2 to S5 is deployed and executed in a layered manner based on the edge, fog and cloud collaborative computing architecture.

[0007] Optionally, in S3, the multi-semantic detection head includes a target detection head, a semantic segmentation head, and a key point detection head; the target detection head is used to identify rail defects and fastener status, the semantic segmentation head is used to segment the track bed slab compaction area and the contact wire icing area, and the key point detection head is used to locate key points of the contact wire to calculate the sway.

[0008] Optionally, in S4, the combined weighting method of the analytic hierarchy process and the entropy weighting method specifically includes: calculating the static benchmark weight of the indicator based on the expert judgment matrix; calculating the dynamic data weight of the indicator based on the distribution of monitoring data; and weighting and synthesizing the static benchmark weight and the dynamic data weight.

[0009] Optionally, in S4, the process of dynamically adjusting the indicator weights by the scenario awareness mechanism includes: real-time monitoring of environmental sensor and equipment status signals; when the monitoring signal matches a preset risk scenario rule, amplifying the weight adjustment coefficient of the corresponding indicator according to the rule; and performing a dot product operation between the weight adjustment coefficient and the weight obtained by combined weighting to obtain the final comprehensive weight.

[0010] Optionally, the layered deployment and execution of the edge, fog, and cloud collaborative computing architecture specifically includes: running a lightweight model at the edge layer to reflect extreme risks at the millisecond level and trigger the highest level response; running the feature fusion model, shared backbone network, multi-semantic detection head, and dynamic security assessment model at the fog computing layer to complete fusion, identification, assessment, and hierarchical early warning decisions at the second or minute level; and performing model optimization and knowledge base updates at the cloud computing layer.

[0011] Optionally, prior to S1, a security assessment indicator system is also constructed, which includes an objective indicator layer for quantifying the health status of the infrastructure itself, and an environmental indicator layer for characterizing the dynamic risks of external natural conditions.

[0012] Optionally, the objective indicator layer includes track geometry deterioration degree, track bed slab compaction visual index, catenary dynamic parameters, rail defect index, fastener integrity rate, signal equipment fault identification rate, automatic fault diagnosis accuracy, and early warning information generation timeliness.

[0013] Optionally, the environmental indicator layer includes the comprehensive wind damage risk value, ice adhesion risk index, lightning overvoltage frequency, earthquake early warning P-wave response time, and heavy rainfall roadbed water damage risk index.

[0014] Secondly, the present invention also provides a computer terminal device, comprising: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the railway infrastructure monitoring and safety level assessment method based on cross-business data fusion of engineering power supply in the first aspect above.

[0015] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the railway infrastructure monitoring and safety level assessment method based on cross-business data fusion of engineering and power supply in the first aspect described above.

[0016] Compared with the prior art, the present invention has the following advantages and technical effects: This invention provides a method for monitoring and assessing the safety level of railway infrastructure based on cross-business data fusion of engineering and power supply. Through cross-business data fusion and intelligent assessment mechanisms, it achieves accurate monitoring and dynamic early warning of railway infrastructure status. This invention realizes deep semantic fusion of multi-source data across businesses, breaking through the information silo limitations of traditional monitoring systems. By fully exploring the coupling relationships between data through a bidirectional cross-attention mechanism, it improves the accuracy of status identification and adaptability in complex environments. This invention constructs a dynamically adaptive safety assessment system. Through hierarchical analysis, entropy weight combination, and context-aware mechanisms, it achieves real-time dynamic adjustment of assessment weights according to the environment and equipment status, improving the scientific rigor and timeliness of assessment results. This invention implements hierarchical intelligent early warning and response. Based on an edge, fog, and cloud collaborative computing architecture, it supports full-process risk management from millisecond-level instinctive reflexes to long-cycle knowledge evolution, ensuring the timeliness of early warnings and the accuracy of responses. Ultimately, this invention significantly improves the intelligence level of railway infrastructure operation and maintenance, possessing the characteristics of full automation, scalability, and continuous optimization, enhancing traffic safety assurance capabilities and operation and maintenance decision-making efficiency. Attached Figure Description

[0017] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating the railway infrastructure monitoring and safety level assessment method based on cross-business data fusion of engineering and power supply according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a cross-business feature fusion model based on a bidirectional cross-attention mechanism according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the security level assessment model and emergency response plan for an embodiment of the present invention; Figure 4 This is a schematic diagram of a hierarchical intelligent evaluation and response system based on an edge, fog, and cloud collaborative architecture, according to an embodiment of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0020] like Figure 1As shown, this embodiment provides a method for monitoring and assessing the safety level of railway infrastructure based on cross-business data fusion of engineering and power supply, including: S1. Collect multi-source data across engineering and power supply services, and perform spatiotemporal alignment and quality repair processing on the multi-source data, which includes engineering data, power data and environmental data. S2. Input the processed multi-source data into the feature fusion model built on the bidirectional cross-attention mechanism to perform deep semantic fusion across business data and generate a unified fusion feature vector. S3. Input the fused feature vector into the shared backbone network to extract general semantic features, and use multiple semantic detection heads to perform parallel decoding of the general semantic features to identify the status of key infrastructure components and output quantitative indicator values ​​corresponding to preset objective indicators and environmental indicators. S4. The quantitative index values ​​are normalized, and weights are assigned by a combination of the analytic hierarchy process and the entropy weight method. The index weights are dynamically adjusted by combining the scenario awareness mechanism to calculate the comprehensive safety evaluation value. S5. Based on the preset threshold range where the comprehensive security evaluation value is located, classify the security level and trigger the corresponding early warning response; wherein, the process from S2 to S5 is deployed and executed in a layered manner based on the edge, fog and cloud collaborative computing architecture.

[0021] Furthermore, prior to S1, a security assessment indicator system is constructed, which includes an objective indicator layer for quantifying the health status of the infrastructure itself, and an environmental indicator layer for characterizing the dynamic risks of external natural conditions.

[0022] Furthermore, the objective indicator layer includes track geometry deterioration degree, track bed slab visual index, catenary dynamic parameters, rail defect index, fastener integrity rate, signal equipment fault identification rate, fault automatic diagnosis accuracy, and early warning information generation timeliness.

[0023] Furthermore, the environmental indicator layer includes the comprehensive wind damage risk value, ice adhesion risk index, lightning overvoltage frequency, earthquake early warning P-wave response time, and heavy rainfall roadbed water damage risk index.

[0024] Specifically, the implementation process of this embodiment includes: like Figure 1 As shown, this embodiment includes: constructing an indicator system, defining evaluation dimensions and quantitative targets; cross-business data fusion perception to obtain the raw state data required for the indicators; dynamic data evaluation to output security levels and early warning instructions; and layered execution in the edge, fog, and cloud architecture to achieve real-time operation and continuous optimization of the process.

[0025] Construction of a safety assessment indicator system: Construction of objective indicators: It directly reflects the health and performance status of the infrastructure itself (engineering and electrical systems), and its data comes entirely from the direct output of the subsequent intelligent sensing modules. Specific indicators are shown in Table 1.

[0026] Table 1

[0027] Environmental indicator construction: This represents the real-time, dynamic risk posed by the external environment to infrastructure, with data derived from environmental sensors or calculations fused with environmental data. Specific indicators are shown in Table 2.

[0028] Table 2

[0029] Furthermore, cross-business data fusion perception and evaluation: Data acquisition and standardized preprocessing: Based on the requirements of the indicator system (Tables 1 and 2), heterogeneous data from multiple sources, including civil engineering, electrical engineering, and environmental data, were collected simultaneously. Standardization operations such as spatiotemporal alignment and quality checks were performed on the raw data to form a spatiotemporally consistent multi-source data stream. See Table 3 for details.

[0030] Table 3

[0031] Feature-level fusion based on bidirectional cross-attention mechanism: like Figure 2 As shown, to achieve deep semantic interaction among heterogeneous data from multiple sources such as engineering, electrical, and environmental data, this embodiment abandons traditional feature splicing or early fusion schemes and innovatively introduces a bidirectional cross-attention mechanism. This mechanism is the core algorithmic guarantee for achieving deep cross-business correlation perception in this embodiment. Traditional fusion methods only perform physical data merging and fail to uncover the inherent semantic relationships between data. The attention mechanism adopted in this embodiment enables dynamic interaction of features.

[0032] Specifically, this embodiment generates three sets of representations for each service's characteristics: , , Vectors. During fusion, the query vector of each business sequentially identifies the key vectors of all other businesses. By calculating similarity, a set of dynamic, normalized attention weights is obtained. These weights determine how much information should be extracted from the feature values ​​of each other business when generating new features for the current business. Pairwise, mutual attention calculations are performed between all business features, and the original business features are enhanced and modified by relevant information from all other businesses. The calculation process is as follows: ; in, For query vector; The key vector; It is a value vector; is the scaling factor, representing the dimension of the key vector.

[0033] Each business's new features, weighted by the attention of all other businesses, are then residually joined with its original features and normalized. Next, all the new features from all businesses are concatenated along the channel dimension, and then dimensionality-reduced and fused through a fully connected layer to ultimately generate a unified feature. Fuse feature vectors.

[0034] Furthermore, in S3, the multi-semantic detection head includes a target detection head, a semantic segmentation head, and a key point detection head; the target detection head is used to identify rail defects and fastener status, the semantic segmentation head is used to segment the track bed slab compaction area and the contact wire icing area, and the key point detection head is used to locate key points of the contact wire to calculate the sway.

[0035] Specifically, the implementation process of this embodiment includes: Collaborative identification and index quantification based on fusion features: The fused feature vector The input is combined with a shared backbone network and multiple semantic detection heads to perform component state recognition. The shared backbone network... Higher-level general semantic features are extracted. Based on these general features, the multi-semantic detection head decodes the specific quantization results in parallel. Finally, a series of real-time raw values ​​of quantization indicators that conform to the definition of the indicator system are obtained (such as "swaying amount: 150mm", "caking rate: 0.15"). The functional descriptions of the shared backbone network and the multi-semantic detection head are shown in Table 4.

[0036] Table 4

[0037] Furthermore, dynamic security assessment and early warning generation: Data normalization: The calculation engine outputs the raw values ​​of all secondary objective indicators in real time. All objective indicators are then processed into dimensionless, comparable numerical values, resulting in a normalized vector of objective indicators. , where n is the total number of objective indicators.

[0038] Benefit-oriented indicators: the higher the value, the safer; ; Cost-related indicators: the higher the value, the more dangerous it is; ; in and These are the historical maximum and minimum values ​​of the indicator, or the theoretically allowed limit.

[0039] Furthermore, in S4, the combined weighting method of the analytic hierarchy process and the entropy weighting method specifically includes: calculating the static benchmark weight of the indicator based on the expert judgment matrix; calculating the dynamic data weight of the indicator based on the distribution of monitoring data; and weighting and synthesizing the static benchmark weight and the dynamic data weight.

[0040] Furthermore, in S4, the process of dynamically adjusting the indicator weights by the scenario awareness mechanism includes: real-time monitoring of environmental sensor and equipment status signals; when the monitoring signal matches a preset risk scenario rule, amplifying the weight adjustment coefficient of the corresponding indicator according to the rule; and performing a dot product operation between the weight adjustment coefficient and the weight obtained by combined weighting to obtain the final comprehensive weight.

[0041] Specifically, the implementation process of this embodiment includes: Dynamic weight calculation: a. Calculate the static benchmark weights : Static baseline weights are calculated using the Analytic Hierarchy Process (AHP). The calculation expert directly performs pairwise comparisons on all n secondary objective indicators, forming a judgment matrix A. Let there be a total of n secondary objective indicators participating in the evaluation, denoted as... We invited k domain experts to conduct pairwise importance comparisons of all secondary indicators using the Saaty 1-9 scale shown in Table 5.

[0042] Table 5

[0043] No. The judgments of the experts form an n×n positive-negative judgment matrix. : ; Among them, matrix elements satisfy: , ,and This matrix accurately depicts any two indicators in the experts' minds. and The relative importance of the two.

[0044] To synthesize the opinions of multiple experts, this embodiment employs the geometric mean method for aggregated decision-making. The consistency ratio (CR) of each decision matrix is ​​calculated to eliminate severely inconsistent judgments. Then, the geometric mean of the corresponding elements of the decision matrices that pass the consistency test is used as the element of the group decision matrix A. ; Wherein, the index subscript of the expert, l=1,2,…,k, indicates the number of experts. One expert; This represents the total number of domain experts who participated in the evaluation.

[0045] Calculate the geometric mean of the elements in each row: ; in, This refers to the total number of secondary objective indicators participating in the evaluation.

[0046] right Normalization is performed to obtain the weights: ; in, Let be the final static benchmark weight of the i-th indicator.

[0047] Obtain the static baseline weight vector: .

[0048] b. Calculate dynamic data weights : To overcome the limitation of static weights in reflecting real-time system state changes, this embodiment introduces the entropy weight method, calculating dynamic weights based on the real-time distribution characteristics of monitoring data. The higher the dispersion of an indicator, the smaller its entropy value, the greater the amount of information it provides, and the higher its weight should be assigned in the evaluation. Assume the evaluation system collects data from m effective monitoring units (e.g., different mileage sections along the route) within a certain analysis period (e.g., the past 24 hours), and the evaluation system contains n secondary objective indicators. After normalizing all monitoring data according to indicator type, the following data matrix is ​​formed: ; in, This represents the normalized value of the i-th monitoring unit on the j-th evaluation index. This matrix X comprehensively depicts the overall state distribution of all monitored objects within the current time window.

[0049] Convert each indicator value into a weighting form, and calculate the characteristic weighting of the i-th monitoring unit under the j-th indicator. : ; Where m is the number of valid monitoring unit data collected; Based on the definition of information entropy, calculate the information entropy value of the j-th indicator. : ; in, The information entropy coefficient is used to ensure... .when At that time, it was stipulated .

[0050] Define the coefficient of difference for the j-th indicator. : ; Among them, the coefficient of difference This reflects the degree of difference in the value of the j-th indicator among the various monitoring units. The larger the value, the more significant the indicator's role in distinguishing the status of monitoring units.

[0051] The difference coefficients are normalized to obtain the dynamic weight of the j-th indicator. : ; Finally, the dynamic data weight vector is obtained. .

[0052] c. Determine the scenario adjustment factor : To enable the safety assessment system to proactively respond to extreme operating conditions and emergencies, this embodiment designs a scenario-adaptive weight adjustment mechanism. This mechanism dynamically identifies specific risk scenarios by real-time analysis of multi-source information from environmental sensors, equipment monitoring units, and the scheduling system, and generates a weight adjustment coefficient vector accordingly. This allows for instantaneous and precise control over the weights of evaluation indicators.

[0053] make Adjust the coefficient vector for the scenario, where The adjustment coefficient corresponding to the j-th secondary objective indicator. The value is determined by a predefined scenario response rule base. This embodiment continuously monitors a set of key scenario trigger signals. This includes: extreme environmental signals, such as instantaneous wind speed. Exceeding the safety threshold, the earthquake early warning system is triggered. Etc.; Key equipment status signals: such as failure rate of critical components Sudden changes, sharp increases in specific monitoring indicators (such as catenary sway), etc. The rule base adopts an "IF-THEN" production rule structure, mapping the input scenario signal S to specific weight adjustment strategies. Some core rules are as follows: Rule R1 (High Wind Scenario): IF( THEN; Among them, λ_wind hazard comprehensive risk value = 1.5, λ_contact network dynamic parameters = 1.3, and λ_other indicators = 1.0. Explanation: When strong wind weather is identified, the decisive role of indicators directly related to wind hazards is significantly enhanced.

[0054] Rule R2 (Earthquake Early Warning Scenario): IF( THEN; Among them, λ earthquake early warning P-wave response time = 2.0, λ bridge / tunnel deformation rate = 1.8, and λ other indicators = 0.9. Explanation: Under earthquake early warning, emergency response speed and structural stability become the core concerns, and their weights are greatly amplified, while the weights of other indicators are appropriately reduced.

[0055] Rule R3 (Foreign Object Intrusion Alarm): IF( THEN; Among them, λearing information generation timeliness = 1.7, λrail defect index = 1.2, and λother indicators = 1.0. Explanation: For sudden intrusion events, the system response speed and the real-time assessment of the track body status are given higher priority.

[0056] The system monitors real-time signal streams. The matching process involves executing the assignment operation corresponding to a specific rule when it is triggered, thereby generating an adjustment coefficient vector. If multiple rules are triggered simultaneously, the maximum value of the adjustment coefficients for the same indicator will be used to ensure a full response to the most severe risks. If no rules are triggered, the system will output the default value. (i.e., a vector in which all elements are 1), at this point the overall weights remain unchanged and no scenario adjustment is made.

[0057] Calculate the final composite weight : ; ; in, , It is a hyperparameter that balances expert experience and data patterns, and can be initially set to 0.6 and 0.4. It is the Hadamard product (vector dot product), representing Each element in the multiplier The corresponding elements are used to instantly amplify or reduce the weight of specific indicators.

[0058] Objective safety evaluation value : Normalized objective index vector Final comprehensive weight Input, and obtain the objective evaluation value. : ; in, The final comprehensive weight of the j-th indicator; is the normalized value of the j-th index.

[0059] Security level classification: like Figure 3 As shown, after calculating the comprehensive security evaluation value S, this embodiment automatically determines the security level of the current infrastructure based on the preset security affiliation level range and triggers a standardized early warning and response process linked to it. The determination thresholds, technical meanings, and response measures for each level are detailed in Table 6.

[0060] Table 6

[0061] Furthermore, the layered deployment and execution of the edge, fog, and cloud collaborative computing architecture specifically includes: running a lightweight model at the edge layer to reflect extreme risks at the millisecond level and trigger the highest level response; running the feature fusion model, shared backbone network, multi-semantic detection head, and dynamic security assessment model at the fog computing layer to complete fusion, identification, assessment, and hierarchical early warning decisions at the second or minute level; and performing model optimization and knowledge base updates at the cloud computing layer.

[0062] Specifically, the implementation process of this embodiment includes: Hierarchical evaluation framework and computational deployment: like Figure 4 As shown, considering the real-time nature and continuous optimization of security assessment, this embodiment proposes a collaborative perception and assessment method based on a layered architecture of edge, fog, and cloud, combining subjective and objective factors. The core of this method intelligently distributes a deep learning model that combines a unified shared backbone with multi-semantic detection heads across different computational layers, tightly coupling it with a dynamic subjective-objective fusion security assessment model to form a complete closed loop from perception to cognition. The feature recognition task is mainly completed by the shared backbone and multi-semantic detection heads, performed at the fog computation layer. The roles of each layer in the assessment framework and the feature recognition roles of the shared backbone and multi-semantic detection heads will be described separately.

[0063] Edge layer: The edge layer is deployed within the field data acquisition unit, and its core function is to achieve millisecond-level instinctive response to extreme risks. This layer receives pre-processed cross-service fused feature streams and runs an ultra-lightweight target detection model. Its decision-making logic is entirely based on pre-defined deterministic expert rules, matching and identifying only a few extremely dangerous patterns (such as large foreign object intrusion). Once triggered, it directly bypasses the upper-layer decision-making link and executes the highest-level (red) emergency response to ensure bottom-line safety in the worst-case scenarios. Data that does not trigger extreme alerts and preliminary results are synchronously uploaded to the fog computing layer.

[0064] Fog Computing Layer: The fog computing layer is the core of this framework, responsible for the deep fusion of subjective and objective data at the second and minute levels, and for optimal early warning decisions. At this layer, a model combining a complete shared backbone and multiple semantic detection heads performs deep analysis of the uploaded fusion features. Based on the recognition results and the built-in indicator engine, it calculates objective security indicators in real time. The fog computing layer calculates indicator weights through a comprehensive weighting model that integrates AHP static weights and entropy weighting dynamic weights, and dynamically adjusts them in conjunction with a context-aware mechanism. Subsequently, the weighted objective evaluation value is fused with the subjective evaluation value based on fuzzy comprehensive evaluation to generate a comprehensive security evaluation value. Finally, based on the preset security level range, it automatically triggers four levels of early warning: "blue, yellow, orange, and red," completing the intelligent decision-making closed loop from multi-source data to hierarchical response.

[0065] Cloud computing layer: The cloud computing layer primarily enables the collaborative evolution of subjective and objective knowledge, which occurs on a cycle of hours or days.

[0066] The updating of subjective knowledge relies on expert review and feedback on uncertain cases within a fog-like environment, forming labeled gold samples. These are used to update the integrated perception model, as well as for correction and... Enriching the subjective rules and prior distributions in the knowledge base enables the expert knowledge system to be iteratively optimized.

[0067] In this embodiment, a computer terminal device is provided, including: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the above-described method for monitoring and assessing the safety level of railway infrastructure based on cross-business data fusion of engineering and power supply.

[0068] In this embodiment, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the above-described method for monitoring and assessing the safety level of railway infrastructure based on cross-business data fusion of engineering and power supply.

[0069] This invention provides a method for monitoring and assessing the safety level of railway infrastructure based on cross-business data fusion of engineering and power supply. Through cross-business data fusion and intelligent assessment mechanisms, it achieves accurate monitoring and dynamic early warning of railway infrastructure status. This invention realizes deep semantic fusion of multi-source data across businesses, breaking through the information silo limitations of traditional monitoring systems. By fully exploring the coupling relationships between data through a bidirectional cross-attention mechanism, it improves the accuracy of status identification and adaptability in complex environments. This invention constructs a dynamically adaptive safety assessment system. Through hierarchical analysis, entropy weight combination, and context-aware mechanisms, it achieves real-time dynamic adjustment of assessment weights according to the environment and equipment status, improving the scientific rigor and timeliness of assessment results. This invention implements hierarchical intelligent early warning and response. Based on an edge, fog, and cloud collaborative computing architecture, it supports full-process risk management from millisecond-level instinctive reflexes to long-cycle knowledge evolution, ensuring the timeliness of early warnings and the accuracy of responses. Ultimately, this invention significantly improves the intelligence level of railway infrastructure operation and maintenance, possessing the characteristics of full automation, scalability, and continuous optimization, enhancing traffic safety assurance capabilities and operation and maintenance decision-making efficiency.

[0070] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring and assessing the safety level of railway infrastructure based on cross-business data fusion of engineering and power supply, characterized in that, Includes the following steps: S1. Collect multi-source data across engineering and power supply services, and perform spatiotemporal alignment and quality repair processing on the multi-source data, which includes engineering data, power data and environmental data. S2. Input the processed multi-source data into the feature fusion model built on the bidirectional cross-attention mechanism to perform deep semantic fusion across business data and generate a unified fusion feature vector. S3. Input the fused feature vector into the shared backbone network to extract general semantic features, and use multiple semantic detection heads to perform parallel decoding of the general semantic features to identify the status of key infrastructure components and output quantitative indicator values ​​corresponding to preset objective indicators and environmental indicators. S4. The quantitative index values ​​are normalized, and weights are assigned by a combination of the analytic hierarchy process and the entropy weight method. The index weights are dynamically adjusted by combining the scenario awareness mechanism to calculate the comprehensive safety evaluation value. S5. Based on the preset threshold range where the comprehensive security evaluation value is located, classify the security level and trigger the corresponding early warning response; wherein, the process from S2 to S5 is deployed and executed in a layered manner based on the edge, fog and cloud collaborative computing architecture.

2. The method according to claim 1, characterized in that, In S3, the multi-semantic detection head includes a target detection head, a semantic segmentation head, and a key point detection head; the target detection head is used to identify rail defects and fastener status, the semantic segmentation head is used to segment the track bed slab compaction area and the contact wire icing area, and the key point detection head is used to locate key points of the contact wire to calculate the sway.

3. The method according to claim 1, characterized in that, In S4, the combined weighting method of the analytic hierarchy process and the entropy weighting method specifically includes: calculating the static benchmark weight of the index based on the expert judgment matrix; calculating the dynamic data weight of the index based on the distribution of monitoring data; and weighting and synthesizing the static benchmark weight and the dynamic data weight.

4. The method according to claim 3, characterized in that, In S4, the process of dynamically adjusting the indicator weights by the scenario awareness mechanism includes: real-time monitoring of environmental sensor and equipment status signals; when the monitoring signal matches a preset risk scenario rule, amplifying the weight adjustment coefficient of the corresponding indicator according to the rule; and performing a dot product operation between the weight adjustment coefficient and the weight obtained by combined weighting to obtain the final comprehensive weight.

5. The method according to claim 1, characterized in that, The layered deployment and execution of the edge, fog, and cloud collaborative computing architecture specifically includes: running a lightweight model at the edge layer to reflect extreme risks at the millisecond level and trigger the highest level response; running the feature fusion model, shared backbone network, multi-semantic detection head, and dynamic security assessment model at the fog computing layer to complete fusion, identification, assessment, and hierarchical early warning decisions at the second or minute level; and performing model optimization and knowledge base updates at the cloud computing layer.

6. The method according to claim 1, characterized in that, Prior to S1, a security assessment indicator system is also constructed, which includes an objective indicator layer for quantifying the health status of the infrastructure itself, and an environmental indicator layer for characterizing the dynamic risks of external natural conditions.

7. The method according to claim 6, characterized in that, The objective indicator layer includes track geometry deterioration, track bed slab visual index, catenary dynamic parameters, rail defect index, fastener integrity rate, signal equipment fault identification rate, automatic fault diagnosis accuracy, and early warning information generation timeliness.

8. The method according to claim 6, characterized in that, The environmental indicator layer includes the comprehensive risk value of wind damage, the risk index of ice adhesion, the frequency of lightning overvoltage, the response time of P-wave earthquake early warning, and the risk index of roadbed water damage caused by heavy rainfall.

9. A computer terminal device, characterized in that, include: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors perform the steps of the method as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-8.

Images