Heavy-load train control safety evaluation method and system and electronic equipment

Abstract

The invention relates to the field of heavy-load train safety, and discloses a heavy-load train control safety evaluation method and system and electronic equipment. The method comprises the following steps: collecting monitoring data of the heavy-load train, wherein the monitoring data comprises slow change data and time sequence data; performing flexible section division on the operation line based on line features, and performing feature extraction on data acquired by each flexible section unit to obtain a total feature set corresponding to each unit; constructing a feature relationship multi-graph for representing the relationship between the features based on the correlation analysis result of the total feature set; analyzing the feature relationship multi-graph to obtain an optimal feature set of each flexible section unit; and on the basis of the optimal feature set of each flexible section unit, abnormal monitoring is performed on the operation state through a preset context sensing model, and operation safety is evaluated according to a monitoring result. The system construction and maintenance cost is reduced, analysts can be supported to carry out deeper insight, and the intelligent degree of multi-dimensional monitoring data analysis is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of heavy-haul train safety, and particularly to a method, a system, and an electronic device for evaluating the safety of heavy-haul train operation. Background Art

[0002] Under the related technology, in the actual operation of large heavy-haul transportation channels, there are some key technical problems with large marshalling heavy-haul freight cars, such as: large longitudinal vehicle body, coupler fracture, uncoupling, accidental emergency stop slowdown, etc. These problems seriously affect the safety and efficiency of transportation.

[0003] In addition, in view of the problems in braking performance caused by some heavy-haul train derailment accidents, it shows that there are also defects in the existing heavy-haul trains in terms of braking systems, coupler devices, and wheel-rail relationships. Therefore, it is necessary to conduct more comprehensive system monitoring and comprehensive big data analysis to further improve the stability and safety of marshalling heavy-haul trains.

[0004] In the related technology, there are defects in the following aspects:

[0005] Data singularity problem: Most of the analyses in the existing technology are limited to a single dimension. A single data source usually can only provide information from a specific angle, such as coupler force, wheel-rail force, etc. Although these data are important, it is difficult to reveal the interaction between different factors, thus limiting the comprehensive judgment ability for complex scenarios. The complex scenarios of heavy-haul railways involve factors such as lines, vehicles, environment, and people. A single data source usually cannot comprehensively represent the multi-dimensional dynamic relationships in these complex scenarios. At the same time, due to the lack of sufficient historical data and context information, it is impossible to effectively model potential non-linear correlation relationships.

[0006] Insufficient analysis accuracy: In the existing technology, certain progress has been made in the separate monitoring of braking systems, coupler devices, and the study of wheel-rail forces. However, these technologies usually fail to achieve comprehensive multi-dimensional data analysis, so it is difficult to comprehensively evaluate the comprehensive performance of the system.

[0007] Cost problem: The existing technology often requires multiple sets of independent monitoring systems and data analysis tools, which increases the construction and maintenance costs of the system. The incompatibility of data between different systems also causes additional integration costs. Due to the lack of a comprehensive analysis platform, the existing technology has low efficiency in dealing with complex working conditions and dynamic environments, and cannot provide real-time feedback on the system status, resulting in a long cycle for problem discovery and solution.

[0008] Insufficient depth of data analysis: The data analysis process pays more attention to limited information such as maximum values and overlimit values, and lacks the understanding, extraction, and analysis of deeper-level features in various time domains, frequency domains, time series, spatio-temporal dimensions, etc. Summary of the Invention

[0009] During the railway transportation process, the braking performance and safety of trains are directly related to the safety and efficiency of operation. However, the operating safety of trains is often affected by various factors, including dynamic parameters such as line conditions, train formation, braking characteristics of locomotives and rolling stock, and operating status. The interaction of these factors may lead to serious consequences such as insufficient braking, coupler fracture, and vehicle rollover. However, the existing safety monitoring and analysis processes for heavy-haul trains often focus on single factors, and the evaluation methods are mostly traditional qualitative or simple quantitative evaluations based on threshold judgments, which can only handle and analyze after accidents and can no longer meet the dynamic analysis and evaluation requirements facing multiple, non-linear, uncertain interferences, and massive detection data.

[0010] In short, in terms of data sources, there are numerous, diverse, widely distributed, and rapidly evolving factors affecting the operating safety state of heavy-haul railways. It is technically complex and difficult to fuse data to accurately, real-time, comprehensively, and rapidly obtain state information; in terms of analysis objectives, the causes of unsafe operating states of trains are complex, the operating conditions are variable, and external interferences are severe, posing great challenges for highly accurate abnormal state diagnosis and safety state evaluation.

[0011] Relying on the research background of the comprehensive experiment of long heavy-haul trains with multiple monitoring sections and multiple formation modes carried out for the first time in China, and based on the big data acquisition of the safety-critical information of heavy-haul railways operating under real working conditions, the present invention needs to solve the following technical problems:

[0012] 1. Acquisition and integration of multi-source data:

[0013] How to break through the limitations of a single data source, obtain multi-dimensional and multi-type data, and provide a multi-faceted and multi-angle data source for data analysis is the prerequisite for accurate analysis of the operating safety of heavy-haul trains.

[0014] 2. Fusion and processing of multi-source heterogeneous data:

[0015] The analysis of the operating safety of heavy-haul trains involves various data sources, such as line conditions, track conditions, braking system pressure, coupler force, wheel-rail force, vehicle vibration, environmental conditions (temperature, humidity, wind force), etc. These data have multiple dimensions, heterogeneous sources, and rapid dynamic changes. How to uniformly process these multi-source heterogeneous data, extract effective and key features from them, and synthesize multi-dimensional information to form systematic safety evaluation indicators is an important technical problem to be solved by the present invention.

[0016] 3. Spatiotemporal dynamic characteristics:

[0017] During the train operation process, the safety state is affected by spatio-temporal changes. For example, during long-distance operation, the changes in slopes and curve radii affect the safety risks brought about by train operation. How to process dynamic data with time series characteristics and comprehensively evaluate the train operation safety at different geographical locations (spatial dimension) is the technical problem to be solved by the present invention.

[0018] 4. Identification and Modeling of Safety Critical Variables:

[0019] The operation safety involves the interaction of multiple key variables. For example, the influence of brake force distribution and train length on derailment risk. Identifying and modeling the specific impacts of these variables on safety is challenging. How to identify and quantify the safety critical variables during the operation process and how to establish the correlation model between variables for evaluating the overall operation safety are the technical problems to be solved by the present invention.

[0020] 5. Abnormal State Detection:

[0021] During the operation process, abnormal conditions (such as brake failure, train derailment, etc.) or high-risk states may occur, which need to be detected and warned in time. How to establish a safety threshold model to judge whether there is a risk of exceeding the safety domain in the current state is one of the technical problems to be solved by the present invention.

[0022] The purpose of the present invention is to provide at least one heavy-haul train operation safety evaluation method, system and electronic device. Based on the multi-source monitoring data collected by the multi-dimensional perception integrated monitoring platform for heavy-haul trains, the data is stored in the data center, and the data content includes line conditions and locomotive monitoring parameters and vehicle monitoring parameters with unified spatio-temporal tags. The line conditions include information such as curves, slopes, stations, phase separators, etc.; the locomotive monitoring parameters include train formation information, locomotive traction and braking states, the air pressure state of the end-of-train device, the end-of-train exhaust information, vehicle running speed and mileage, driver operation information, etc.; the vehicle monitoring parameters include brake shoe pressure, brake shoe temperature, the pressure of each component of the braking system, the vibration of the main brake valve, vehicle vibration acceleration, three-way coupler force, wheel-rail force, and environmental temperature and humidity.

[0023] The present invention ensures the accuracy and applicability of the analysis results by fully considering the background and environmental factors of the monitoring data during the analysis process - including line information, driver operation information, environmental temperature and humidity, etc. Specifically, it includes:

[0024] Cost-effectiveness: By integrating different data analysis functions, the cost of system construction and maintenance is reduced, and the overall efficiency is improved;

[0025] Deeper data analysis: By incorporating continuously monitored data into the analysis and making full use of experimental data from multiple different environments and driver operations, the analysis method and anomaly detection method can reflect the complex process under the combined action of multiple factors during the operation of heavy-haul railways, supporting analysts to gain deeper insights;

[0026] The present invention can help users more comprehensively understand and analyze the operation information of heavy-haul trains, thereby providing a comprehensive safety assessment of heavy-haul train operation. At the same time, it improves the intelligence level of multi-dimensional monitoring data analysis to help users better enhance the operation safety of heavy-haul trains and guide the optimization of heavy-haul train operation;

[0027] The anomaly detection method for heavy-haul train operation based on context awareness and window-level loss function enhances the model's sensitivity to spatial anomalies by modeling the characteristics of spatially segmented sections; by combining time-space context, it can simultaneously capture the time dependence in the time series and the correlation relationship between spatially segmented sections; in addition, it also has the characteristics of strong flexibility and can adjust the segmentation strategy and feature extraction method according to different application scenarios.

[0028] To solve the above technical problems, at least one embodiment of the present application provides a safety evaluation method for heavy-haul train operation, and the method includes:

[0029] Collect the monitoring data of the heavy-haul train, and classify the monitoring data based on the change situation of the monitoring data to obtain slow-varying data and time-series data; wherein, the monitoring data includes at least one of the following dimension data: driver operation information, heavy-haul train system, operation line, and external environment;

[0030] Based on the line characteristics of the heavy-haul train operation line, perform flexible section division on the operation line to obtain flexible section units, and respectively perform feature extraction on the data collected for each flexible section unit to obtain the total feature set corresponding to each flexible section unit;

[0031] Perform correlation analysis on the total feature sets of each flexible section unit, and construct a feature relationship multi-graph for characterizing the relationship between the features in the total feature set based on the correlation analysis results;

[0032] Select the first features whose centrality meets the preset first condition from the feature relationship multi-graphs of each flexible section unit by the node centrality method as key features; use the minimum spanning tree method to analyze the feature relationship multi-graphs of each flexible section unit, retain the association structure that meets the preset second condition between the key features and remove redundant features to obtain second features; merge the first features and the second features to obtain the preferred feature set of each flexible section unit;

[0033] Based on the preferred feature sets of each flexible section unit, the operating state of the heavy-haul train is monitored for anomalies through a preset context-aware model, and the operating safety of the heavy-haul train is evaluated according to the monitoring results.

[0034] At least one embodiment of the present application further provides a heavy-haul train operation safety evaluation system, including:

[0035] A data acquisition module, configured to acquire monitoring data of the heavy-haul train, and classify the monitoring data based on the change situation of the monitoring data to obtain slow-varying data and time-series data; wherein, the monitoring data includes data in at least one of the following dimensions: driver operation information, heavy-haul train system, operating line, and external environment;

[0036] A feature set determination module, configured to perform flexible section division on the operating line based on the line features of the heavy-haul train operating line to obtain flexible section units, and perform feature extraction on the data collected for each flexible section unit to obtain the total feature set corresponding to each flexible section unit;

[0037] A feature relationship multi-graph construction module, configured to perform correlation analysis on the total feature sets of each flexible section unit, and construct a feature relationship multi-graph for characterizing the relationships between the features in the total feature set based on the correlation analysis results;

[0038] A preferred feature set determination module, configured to screen out the first features whose centrality meets a preset first condition from the feature relationship multi-graphs of each flexible section unit as key features; use the minimum spanning tree method to analyze the feature relationship multi-graphs of each flexible section unit, retain the association structures between the key features that meet a preset second condition and remove redundant features to obtain second features; merge the first features and the second features to obtain the preferred feature sets of each flexible section unit;

[0039] A monitoring and evaluation module, configured to monitor the operating state of the heavy-haul train for anomalies through a preset context-aware model based on the preferred feature sets of each flexible section unit, and evaluate the operating safety of the heavy-haul train according to the monitoring results.

[0040] At least one embodiment of the present application further provides an electronic device, including a memory, a processor, and a computer program stored on the memory, and the processor executes the computer program to implement the steps of the method as described above.

[0041] At least one embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method as described above are implemented.

[0042] At least one embodiment of the present application further provides a computer program product, including a computer program, which implements the steps of the above-described method when executed by a processor.

[0043] The overload train operation safety evaluation method, system and electronic device provided by the embodiments of the present application, compared with the prior art, fully consider the background and environmental factors of the monitoring data during the analysis process - including line information, driver operation information, environmental temperature and humidity, etc., to ensure the accuracy and applicability of the analysis results. Specifically, it includes:

[0044] Cost - effectiveness: By integrating different data analysis functions, the cost of system construction and maintenance is reduced, and the overall efficiency is improved;

[0045] Deeper data analysis: By incorporating continuously monitored data into the analysis and making full use of experimental data under multiple different environments and driver operations, the analysis method and anomaly detection method can reflect the complex process under the combined action of multiple factors during the operation of heavy - haul railways, supporting analysts to gain deeper insights;

[0046] The present invention can help users more comprehensively understand and analyze the operation information of heavy - haul trains, thereby providing a comprehensive safety assessment of heavy - haul train operation, while improving the intelligence level of multi - dimensional monitoring data analysis to help users better enhance the operation safety of heavy - haul trains and guide the optimization of heavy - haul train operation;

[0047] The heavy - haul train operation anomaly detection method based on context awareness and window - level loss function enhances the model's sensitivity to spatial anomalies by modeling the features of spatially segmented sections; by jointly considering time - space context, it can simultaneously capture the temporal dependence in the time series and the correlation between spatially segmented sections; in addition, it has the characteristics of strong flexibility and can adjust the segmentation strategy and feature extraction method according to different application scenarios.

[0048] In some alternative embodiments, the collection of the monitoring data of the heavy - haul train includes:

[0049] Based on the research of historical literature, the operation mechanism of heavy - haul trains and the hierarchical task analysis of heavy - haul train drivers, as well as the analysis of historical accidents and relevant reports of heavy - haul trains, obtain the influencing factors of heavy - haul train operation safety;

[0050] Classify the influencing factors of operation safety according to a preset classification method to obtain the classification result of the influencing factors of operation safety;

[0051] Based on the influencing factors of operation safety and the classification result of the influencing factors of operation safety, and based on historical data, conduct data source mining to obtain the classification of data sources for the influencing factors of operation safety;

[0052] Collect the monitoring data of the heavy-haul train according to the classification of the data sources of the manipulation safety influencing factors.

[0053] In some alternative embodiments, the preset classification method includes:

[0054] The influencing factors of the driver and operation behavior, the influencing factors of the train system, the influencing factors of the line conditions, and the influencing factors of the external environment. For such complex systems as heavy-haul railways, based on the analysis objective of the operation safety of heavy-haul trains, the elements participating in the operation of the railway system and affecting its safety are decomposed into four core dimensions: "human" (driver), "vehicle" (train system), "road" (line conditions), and "environment" (external environment). Systematically guide the identification of manipulation safety influencing factors and the mining and classification of relevant data sources based on the "human-vehicle-road-environment" framework.

[0055] In some alternative embodiments, the slowly varying data includes data with slow changes or completely static data; the time-series data includes data with dynamic changes.

[0056] In some alternative embodiments, time and line kilometer post information are associated in the time-series data; a piece of data is recorded at the start and end of each trip of the heavy-haul train and at preset acquisition positions respectively, so as to expand the slowly varying data into data with spatio-temporal information that remains unchanged for a long time.

[0057] In some alternative embodiments, the step of extracting features from the data collected for each flexible section unit respectively to obtain the total feature set corresponding to each flexible section unit includes:

[0058] Perform time-domain feature extraction and frequency-domain feature extraction on the data collected for each flexible section unit respectively; where:

[0059] For the slowly varying data collected for each flexible section unit, extract the trend characteristics through linear regression, extract the periodic component characteristics through Fourier transform, and generate a first feature set based on the time-domain statistical characteristics, the trend characteristics, and the periodic component characteristics of the slowly varying data;

[0060] For the time-series data collected for each flexible section unit, perform time-domain analysis and frequency-domain analysis respectively to reflect the change characteristics of the data in time and the change characteristics in frequency, and generate a second feature set based on the change characteristics in time and the change characteristics in frequency;

[0061] Take the set of the first feature set and the second feature set as the total feature set.

[0062] In some alternative embodiments, the steps of performing correlation analysis on the total feature set include:

[0063] Through Pearson correlation coefficient, Spearman correlation coefficient and lag correlation calculation, the time correlation characteristics between different features are analyzed, and through Granger causality test, cointegration analysis and vector autoregression, the causal relationship, cointegration relationship and dynamic relationship between different features are analyzed;

[0064] The spatial correlation between different features is analyzed through spatial distribution pattern, spatial correlation and spatial lag analysis tools to obtain the spatial dependence relationship between different features.

[0065] In some optional embodiments, in the feature relationship multi-graph:

[0066] Nodes represent data features;

[0067] Multiple edges between variables represent different types of correlations, and different types of correlations are distinguished by different colors;

[0068] The weight and direction of the edges are used to represent the strength of the association and the causal relationship respectively. Brief Description of the Drawings

[0069] One or more embodiments are exemplarily illustrated by the pictures in the corresponding drawings. These exemplary illustrations do not limit the embodiments. Elements with the same reference numerals in the drawings are represented as similar elements. Unless otherwise stated, the drawings in the drawings do not constitute a proportional limitation.

[0070] Figure 1 It is a flowchart of a method for evaluating the operation safety of a heavy-haul train provided by an embodiment of the present disclosure;

[0071] Figure 2 It is a schematic diagram of a feature relationship multi-graph provided by an embodiment of the present disclosure. Detailed Description of the Embodiments

[0072] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be described in detail below with reference to the drawings. However, those of ordinary skill in the art can understand that in the embodiments of the present invention, many technical details are proposed for the better understanding of the present invention by readers. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions required to be protected by the present invention can be achieved.

[0073] In the prior art, certain progress has been made in the separate monitoring of braking systems, the monitoring of coupler devices, and the study of wheel-rail forces. However, these technologies generally fail to achieve multi-dimensional comprehensive data analysis, making it difficult to comprehensively evaluate the overall performance of the system. For example, although the performance improvement of the braking system can enhance braking efficiency, without considering the interaction effects of the coupler device and wheel-rail forces, it may lead to problems in other aspects. The prior art often requires multiple independent monitoring systems and data analysis tools, which increases the construction and maintenance costs of the system. The incompatibility of data between different systems also incurs additional integration costs. Due to the lack of a comprehensive analysis platform, the prior art has low efficiency in dealing with complex working conditions and dynamic environments, unable to provide real-time feedback on the system status, resulting in a long cycle for problem discovery and resolution.

[0074] Aiming at the deficiencies of the prior art, the purpose of the embodiments of the present invention is, relying on the research background of the comprehensive experiment of long heavy-haul trains with multiple monitoring sections and multiple formation modes carried out for the first time in the country, and based on the big data acquisition of the safety-critical information of heavy-haul railways operating under real working conditions, the main technical problems to be solved include:

[0075] Insufficient data comprehensiveness: Most of the analyses in the prior art are limited to a single dimension and cannot achieve a comprehensive analysis of braking systems, coupler devices, and wheel-rail forces;

[0076] High integration cost: The prior art requires multiple independent systems for data acquisition and analysis, resulting in relatively high overall system construction and maintenance costs;

[0077] Low efficiency: A single analysis method cannot quickly respond to complex dynamic environments and changes in working conditions, and the data analysis efficiency is low.

[0078] The present invention ensures the accuracy and applicability of the analysis results by fully considering the background and environmental factors of the monitoring data during the analysis process, including line information, driver operation information, environmental temperature and humidity, etc. Specifically, it includes:

[0079] Cost-effectiveness: By integrating different data analysis functions, the construction and maintenance costs of the system are reduced, and the overall efficiency is improved;

[0080] Deeper data analysis: By incorporating continuously monitored data into the analysis and making full use of test data from multiple different environments and driver operations, the analysis method and anomaly detection method can reflect the complex process under the combined action of multiple factors during the operation of heavy-haul railways, supporting analysts to gain deeper insights;

[0081] The present invention can help users comprehensively understand and analyze the operation information of heavy-haul trains, thereby providing a comprehensive safety assessment for the operation of heavy-haul trains. At the same time, it improves the intelligence level of multi-dimensional monitoring data analysis to help users better enhance the operation safety of heavy-haul trains and guide the optimization of heavy-haul train operation;

[0082] The heavy-haul train operation anomaly detection method based on context awareness and window-level loss function enhances the model's sensitivity to spatial anomalies by modeling the features of spatially segmented sections; by jointly considering time-space context, it can simultaneously capture the temporal dependence in the time series and the correlation between spatially segmented sections; in addition, it has the characteristics of strong flexibility and can adjust the segmentation strategy and feature extraction method according to different application scenarios.

[0083] Embodiment 1:

[0084] The implementation manner of the present invention relates to a heavy-haul train operation safety evaluation method.

[0085] The following specifically describes the implementation details of the heavy-haul train operation safety evaluation method of this implementation manner. The following content is only the implementation details provided for convenient understanding and is not necessary for implementing this solution.

[0086] The heavy-haul train operation safety evaluation method of this embodiment can be applied to an electronic device with communication, computing, and data storage capabilities.

[0087] In this embodiment, based on the multi-source monitoring data collected by the heavy-haul train multi-dimensional perception integrated monitoring platform, the data is stored in the data center, and the data content includes line conditions and locomotive monitoring parameters and vehicle monitoring parameters with unified spatio-temporal tags. The line conditions include information such as curves, slopes, stations, and phase separations; the locomotive monitoring parameters include train formation information, locomotive traction and braking states, the air pressure state of the end-of-train device, the end-of-train exhaust information, vehicle running speed and mileage, driver operation information, etc.; the vehicle monitoring parameters include brake shoe pressure, brake shoe temperature, the pressure of each component of the braking system, the vibration of the brake main valve, vehicle vibration acceleration, three-way coupler force, wheel-rail force, and environmental temperature and humidity.

[0088] As Figure 1 shown, the heavy-haul train operation safety evaluation method provided in this embodiment includes the following steps:

[0089] Step 110, collect the monitoring data of the heavy-haul train, and classify the monitoring data based on the change situation of the monitoring data to obtain slowly varying data and time series data; among them, the monitoring data includes data in at least one of the following dimensions: driver operation information, heavy-haul train system, operating line, and external environment.

[0090] Among them, the data of the relevant operating line can be obtained from an external system.

[0091] The present invention targets complex systems such as heavy-haul railways. Based on the analysis goal of heavy-haul train operation safety, the present invention decomposes the factors involved in the operation of the railway system and affecting its safety into four core dimensions: "people" (drivers), "cars" (train systems), "roads" (line conditions), and "environment" (external environment). Based on the framework of "people-cars-roads-environment", the present invention systematically guides the identification of factors affecting operation safety and the mining and classification of relevant data sources. Specifically, it includes the following parts:

[0092] 1. Collection and identification of factors affecting heavy-load train operation safety:

[0093] Based on extensive literature research, heavy-haul train operation mechanisms, layered task analysis of heavy-haul train drivers, and analysis of heavy-haul train accidents and related reports, factors affecting heavy-haul train operating safety are identified and collected.

[0094] 2. Classify the collected influencing factors according to the following classification method:

[0095] (1) Driver and operating behavior (human) influencing factors:

[0096] Physiological state: The impact of fatigue, distraction, and health status on manipulative behavior.

[0097] Psychological state: emotional fluctuations, emergency response ability and psychological stress.

[0098] Skill level: driving experience, proficiency in equipment operation and mastery of rules and regulations.

[0099] Decision-making ability: judgment and response time to emergencies.

[0100] Safety hazards: fatigue driving, misoperation, and slow emergency response.

[0101] (2) Train system (machine) influencing factors:

[0102] Braking system: performance, response speed and reliability of braking equipment.

[0103] Power system: stability of train traction and acceleration performance.

[0104] Control system: Controls the operating status of the system.

[0105] Vehicle status: The health of key components such as wheel wear, bearing failures, and cabin connections.

[0106] Safety hazards: brake failure, system failure, and aging of key components.

[0107] (3) Line conditions (road) influencing factors:

[0108] Track status: track wear, joint condition, track straightness.

[0109] Line conditions: gradient, curvature, switches, etc.

[0110] Other systems: signal lights, stations, line conditions at neutral section passing points.

[0111] Safety hazards: emergency braking caused by track failures, signal malfunctions, and line congestion.

[0112] (4) External environment (surroundings) influencing factors:

[0113] Weather conditions: rain, snow, hail, etc., causing track slipperiness or reduced visibility.

[0114] Natural disasters: interference to train operation caused by earthquakes, landslides, and floods.

[0115] External interference: foreign object intrusion.

[0116] Safety hazards: track slipperiness, blocked vision, and intrusion of external foreign objects.

[0117] 3. Mining of data sources related to safety influencing factors in heavy-haul train operation:

[0118] Based on the collection and classification results of safety influencing factors in heavy-haul train operation, combined with the investigation of available data, information systems, and the collection of data available through public channels on the Internet, the classification and collation of data sources related to safety influencing factors in heavy-haul train operation are as follows:

[0119] (1) People:

[0120] None.

[0121] (2) Vehicles:

[0122] Braking system: types and performance of locomotive and vehicle braking equipment.

[0123] Power system: locomotive power, traction force, electric braking force, traction characteristics, etc.

[0124] Vehicle status: vehicle formation, health status of key components such as wheel wear, braking system, and car body connectors.

[0125] (3) Tracks:

[0126] Line conditions: gradient, curvature, switches, bridges, tunnels.

[0127] Other systems: signal lights, stations, neutral section passing points.

[0128] (4) Surroundings:

[0129] Weather conditions: weather conditions during vehicle operation.

[0130] 4. Sort out the data collected by the monitoring platform based on the "human-vehicle-railway-environment" framework:

[0131] Human: Driver number, driver operation information;

[0132] Vehicle: Brake shoe pressure, brake shoe temperature, pressure of each component of the braking system, vibration of the main brake valve, vehicle vibration acceleration, three-way coupler force, wheel-rail force;

[0133] Railway: The line on which the vehicle is running, and the kilometer post of the line associated with the monitoring data;

[0134] Environment: Ambient temperature, ambient humidity.

[0135] 5. Integration and expansion of data sources:

[0136] Integrate the data sources in the above-mentioned Part 3 and Part 4 to form all data sources available for comprehensive data analysis. The overall "human-vehicle-railway-environment" framework is still adopted. Under each dimension of "human", "vehicle", "railway", and "environment", two categories of "monitoring information" and "influence factors" are distinguished. The data sources sorted out in 4 are put into the "monitoring information" category, and the data sources sorted out in 3 are put into the "influence factors" category.

[0137] With the accumulation of system data, some data can be based on historical data and become new data sources through statistics, analysis, and summary. For example, through the analysis of the operation data of a driver multiple times, the data on his skill level and decision-making ability can be statistically analyzed.

[0138] Step 120. Divide the operation line into flexible section units based on the line characteristics of the heavy-haul train operation line, and extract features from the data collected for each flexible section unit to obtain the total feature set corresponding to each flexible section unit.

[0139] The data collected in the above Step 110 can be divided into two categories according to their change situations: slowly changing or completely static slowly varying data, and time-series data with dynamic changes. In practical applications, all trains running from the starting point to the ending point form an operation trip. All time-series data is associated with time and line kilometer post information. Since the slowly varying data is statically stored or only collected a limited number of times for each trip, it can be expanded into data with spatio-temporal information that remains unchanged for a long time by recording a piece of data at the starting and ending points of each trip and at the position of each collection.

[0140] To include all available data in the analysis, the slowly varying data is combined with the time-series data with dynamic changes in a horizontal fusion manner for data analysis through the above method.

[0141] Step 130: Conduct correlation analysis on the total feature sets of each flexible section unit, and construct a feature relationship multi-graph for characterizing the relationships among the features in the total feature sets based on the correlation analysis results (for reference, see Figure 2 ).

[0142] In the heavy-haul railway system, the multi-source data of "human", "vehicle", "road", and "environment" have strong temporal and spatial correlations. Understanding and mining the spatio-temporal correlation laws of the data will lay a foundation for subsequent data fusion and operation safety analysis. Correlation analysis can help extract useful features, optimize the model input structure, select an appropriate model, and ultimately improve the accuracy, efficiency, and interpretability of the anomaly monitoring model.

[0143] Step 140: Screen out the first features whose centrality meets the preset first condition from the feature relationship multi-graph of each flexible section unit by the node centrality method as key features; analyze the feature relationship multi-graph of each flexible section unit by the minimum spanning tree method, retain the association structure that meets the preset second condition among the key features, and remove redundant features to obtain second features; merge the first features and the second features to obtain the preferred feature set of each flexible section unit.

[0144] Among them, the preset first condition can be set as: the highest centrality; the preset second condition can be set as: the minimum association structure.

[0145] Specifically:

[0146] Feature selection is carried out based on the feature relationship multi-graph (Multilayer Graph) constructed in Step 130. Combining node centrality and the minimum spanning tree (MST) for feature selection can remove redundant features while retaining representative key features.

[0147] In the feature relationship multi-graph, the centrality of a node can measure its importance. Therefore, the features with the highest centrality can be screened out by the node centrality method, which are the key features for subsequent analysis. The centrality method calculates the centrality index of each node, including degree centrality, betweenness centrality, pagerank centrality, etc. After normalizing multiple centrality indexes, the sum of the squared values is taken as the total centrality value, and ranking is performed according to this value, and the top N items are selected as important features.

[0148] Use the minimum spanning tree method (MST) to analyze the feature relationship multi-graph. The MST removes redundant features by retaining the minimum association structure among the key features. Merge the features screened by the MST and the node centrality method to ensure that the feature selection covers key features and effectively avoids redundant features.

[0149] Step 150: Based on the preferred feature sets of each flexible section unit, use a preset context-aware model to perform anomaly monitoring on the operating status of the heavy-haul train, and evaluate the operation safety of the heavy-haul train according to the monitoring results.

[0150] Since the monitoring data during the train operation process is large in quantity and abnormal data is very rare, it means that the representativeness of abnormal situations in the dataset is very low. Due to the lack of a comprehensive description of the time series context, the traditional point-by-point error method makes it difficult for the model to capture complex time dependencies and abnormal patterns across time steps.

[0151] For the typical scenario of sparse abnormal distribution and multi-dimensional time series monitoring data in the operation of heavy-haul trains, by using context-aware modeling to extract complex dependencies between time series, and at the same time evaluating the abnormal patterns of the overall window through sequence-level loss, rather than being limited to single-point errors.

[0152] The method disclosed in this embodiment can, compared with the prior art, more comprehensively evaluate the actual state of train operation. By fully considering the background and environmental factors of the monitoring data - including line information, driver operation information, environmental temperature and humidity, etc. - during the analysis process, the accuracy and applicability of the analysis results are ensured. Moreover, it can also help users more comprehensively understand and analyze the operation information of heavy-haul trains, so as to provide a comprehensive safety evaluation of heavy-haul train operation, while improving the intelligence level of multi-dimensional monitoring data analysis to help users better improve the operation safety of heavy-haul trains and guide the optimization of heavy-haul train operation. Further, the heavy-haul train operation anomaly detection method based on context awareness and window-level loss function enhances the model's sensitivity to spatial anomalies through feature modeling of spatially segmented sections; by combining time-space context, it can simultaneously capture the time dependencies in the time series and the correlation relationships between spatially segmented sections; in addition, it has the characteristics of strong flexibility and can adjust the segmentation strategy and feature extraction method according to different application scenarios.

[0153] Embodiment 2:

[0154] Based on the above embodiment, this embodiment further explains and illustrates the heavy-haul train operation safety evaluation method provided in the above embodiment.

[0155] In step 110: Step 110, collect the monitoring data of the heavy-haul train, and classify the monitoring data based on the change situation of the monitoring data to obtain slow-changing data and time series data; wherein, the monitoring data includes data of at least one of the following dimensions: driver operation information, heavy-haul train system, operation line, and external environment.

[0156] In some embodiments, collecting the monitoring data of the heavy-haul train includes:

[0157] Based on the research of historical documents, the operation mechanism of heavy-haul trains, the hierarchical task analysis of heavy-haul train drivers, and the analysis of historical accidents and related reports of heavy-haul trains, the influencing factors of heavy-haul train operation safety are obtained;

[0158] Classify the influencing factors of operation safety according to the preset classification method to obtain the classification result of the influencing factors of operation safety;

[0159] Based on the influencing factors of operation safety and the classification result of the influencing factors of operation safety, and based on historical data, data source mining is carried out to obtain the data source classification of the influencing factors of operation safety;

[0160] Collect the monitoring data of heavy-haul trains according to the data source classification of the influencing factors of operation safety.

[0161] In some embodiments, the preset classification method includes:

[0162] Driver and operation behavior influencing factors, train system influencing factors, line condition influencing factors, and external environment influencing factors.

[0163] In some embodiments, the slowly varying data includes data with slow changes or completely static data;

[0164] Time series data includes data with dynamic changes.

[0165] In some embodiments, time and line kilometer post information are associated in the time series data;

[0166] Record a piece of data at the starting point and the ending point of each operation trip of the heavy-haul train and at the preset collection positions respectively, so as to expand the slowly varying data into data with spatio-temporal information that remains unchanged for a long time.

[0167] The present invention aims at complex systems such as heavy-haul railways. Based on the analysis objective of heavy-haul train operation safety, the elements participating in the operation of the railway system and affecting its safety are decomposed into four core dimensions: "human" (driver), "vehicle" (train system), "road" (line condition), "environment" (external environment). Based on the "human-vehicle-road-environment" framework, it systematically guides the identification of influencing factors of operation safety and the mining and classification of related data sources. Specifically, it includes the following parts:

[0168] 1. Collection and identification of influencing factors of heavy-haul train operation safety:

[0169] Based on extensive literature research, the operation mechanism of heavy-haul trains, the hierarchical task analysis of heavy-haul train drivers, and the analysis of heavy-haul train accidents and related reports, identify and collect the influencing factors related to heavy-haul train operation safety.

[0170] 2. Classify the collected influencing factors according to the following classification method:

[0171] (1) Driver and operating behavior (human) influencing factors:

[0172] Physiological state: The influence of fatigue, distraction, and health status on operating behavior.

[0173] Psychological state: Emotional fluctuations, emergency handling ability, and psychological pressure.

[0174] Skill level: Driving experience, proficiency in equipment operation, and mastery of rules and regulations.

[0175] Decision-making ability: Judgment of unexpected situations and response time.

[0176] Safety hazards: Fatigue driving, misoperation, and slow emergency response.

[0177] (2) Train system (machine) influencing factors:

[0178] Braking system: Performance, response speed, and reliability of braking equipment.

[0179] Power system: Stability of train traction and acceleration performance.

[0180] Control system: Operating state of the control system.

[0181] Vehicle state: Health status of key components such as wheel wear, bearing failure, and carbody connectors.

[0182] Safety hazards: Brake failure, system failure, and aging of key components.

[0183] (3) Track conditions (road) influencing factors:

[0184] Track state: Track wear, joint condition, and track straightness.

[0185] Track conditions: Gradient, curvature, switches, etc.

[0186] Other systems: Signal lights, stations, and track conditions at neutral section passing positions.

[0187] Safety hazards: Track failure, signal failure, and emergency braking caused by line congestion.

[0188] (4) External environment (environment) influencing factors:

[0189] Weather conditions: Rain, snow, hail, etc. causing track slipperiness or reduced visibility.

[0190] Natural disasters: Interference to train operation caused by earthquakes, landslides, and floods.

[0191] External interference: Intrusion of foreign objects.

[0192] Safety hazards: slippery tracks, blocked vision, intrusion of external objects.

[0193] 3. Mining of data sources related to safety influencing factors in the operation of heavy-haul trains:

[0194] Based on the collection and classification results of safety influencing factors in the operation of heavy-haul trains, combined with the investigation of available data, information systems, and the collection of data available through public channels on the Internet, the data sources related to safety influencing factors in the operation of heavy-haul trains are classified and sorted as follows:

[0195] (1) People:

[0196] None.

[0197] (2) Vehicles:

[0198] Braking system: types and performance of locomotive and vehicle braking equipment.

[0199] Power system: locomotive power, traction force, electric braking force, traction characteristics, etc.

[0200] Vehicle status: vehicle formation, health status of key components such as wheel wear, braking system, and car body connectors.

[0201] (3) Tracks:

[0202] Track conditions: gradient, curvature, switches, bridges, tunnels.

[0203] Other systems: signal lights, stations, neutral section positions.

[0204] (4) Environment:

[0205] Weather conditions: weather conditions during vehicle operation.

[0206] 4. Sorting out the data collected by the monitoring platform based on the "people-vehicle-track-environment" framework:

[0207] People: driver number, driver operation information;

[0208] Vehicles: brake shoe pressure, brake shoe temperature, pressures of various components of the braking system, vibration of the brake main valve, vehicle vibration acceleration, three-way coupler force, wheel-rail force;

[0209] Tracks: the track where the vehicle is running, kilometer post of the monitoring data associated with the track;

[0210] Environment: ambient temperature, ambient humidity.

[0211] 5. Integration and expansion of data sources:

[0212] Integrate the data sources in the above-mentioned Part 3 and Part 4 to form all the data sources available for comprehensive data analysis. The overall framework still adopts the "human-vehicle-road-environment" framework. Under each dimension of "human", "vehicle", "road", and "environment", two categories of "monitoring information" and "influence factors" are distinguished. The data sources sorted out in 4 are put into the "monitoring information" category, and the data sources sorted out in Part 3 are put into the "influence factors" category.

[0213] As the system data accumulates, some data can be based on historical data and become new data sources through statistics, analysis, and summary. For example, through the analysis of the operation data of a driver multiple times, the data on his skill level and decision-making ability can be statistically analyzed.

[0214] Step 120: Divide the operation line into flexible section units based on the line characteristics of the heavy-haul train operation line, and extract the characteristics of the data collected for each flexible section unit respectively to obtain the total feature set corresponding to each flexible section unit. Specifically, it includes the following processing steps:

[0215] 1. Processing of multi-source heterogeneous data:

[0216] The data collected in the above Step 110 can be divided into two categories according to their change situations: slowly changing data that changes slowly or is completely static, and time-series data that changes dynamically. In practical applications, all trains running from the starting point to the ending point form an operation trip. All time-series data is associated with time and line kilometer post information. Since the slowly changing data is stored statically or only collected a limited number of times per trip, it can be expanded into data with spatio-temporal information that remains unchanged for a long time by recording a piece of data at the starting point and ending point of each trip and at the position of each collection.

[0217] To incorporate all available data into the analysis, the slowly changing data is combined with the dynamically changing time-series data in a horizontal fusion manner for data analysis through the above method.

[0218] 2. Flexible section division based on line characteristics:

[0219] The monitoring data is highly affected by line characteristics. To intuitively decompose and extract the line-related characteristics in the monitoring time-series data, so as to more accurately capture the coupling relationship between the line environment and the train operation state, a flexible section division based on line characteristics is proposed.

[0220] As can be seen from Part 3 of the above Step 110, the line-related characteristic data includes: curves, gradients, bridges, tunnels, stations, switches, signal lights, and neutral section passing.

[0221] First, divide the line into straight and curve sections with different curvatures according to the curvature characteristics. Each section contains information about its starting point, ending point, and curve radius.

[0222] Based on the above section division, according to the slope characteristics, the lines with different slopes within a section are further divided, and each section is appended with its slope-related information.

[0223] By analogy, all features such as bridges, tunnels, stations, switches, signal machines, and over-equipment are incorporated into the division conditions, and the overall line is divided into flexible sections with different line characteristics.

[0224] 3. Data feature extraction based on flexible unit division:

[0225] In some embodiments, feature extraction is respectively performed on the data collected for each flexible section unit to obtain the total feature set corresponding to each flexible section unit, including:

[0226] Time-domain feature extraction and frequency-domain feature extraction are respectively performed on the data collected for each flexible section unit; where:

[0227] For the slowly varying data collected for each flexible section unit, the trend characteristics are extracted through linear regression, the periodic component characteristics are extracted through Fourier transform, and a first feature set is generated based on the time-domain statistical characteristics, trend characteristics, and periodic component characteristics of the slowly varying data;

[0228] For the time-series data collected for each flexible section unit, time-domain analysis and frequency-domain analysis are respectively performed to reflect the change characteristics of the data in time and in frequency, and a second feature set is generated based on the change characteristics in time and in frequency;

[0229] The set of the first feature set and the second feature set is used as the total feature set.

[0230] Specifically, according to the line kilometer post information associated with the monitoring data and the flexible section division result of the second step in this procedure, time-domain and frequency-domain feature extraction are respectively performed on the monitoring data within each flexible section according to the data category.

[0231] (1) For the slowly varying data, its data changes slowly, has strong trend, and less noise, such as weather, environmental temperature and humidity, etc., and its time-domain statistical characteristics, trend, and periodic variation are mainly concerned. The time-domain related features mainly include: mean, variance, extreme value, peak-to-peak value, skewness, kurtosis, root mean square value, etc. Linear regression is used to extract its trend features, and Fourier transform is used to extract its periodic components.

[0232] (2) For dynamic monitoring data, due to the characteristics such as non-stationarity of the operation data of heavy-haul trains in a complex operating environment, time-frequency domain analysis is used to reflect the variation characteristics of the signal in time and frequency. The present invention uses the ensemble empirical mode decomposition (EEMD) method to adaptively decompose the dynamic monitoring data into multiple intrinsic mode functions (IMFs) and a residual. Principal component analysis (PCA) is performed on the extracted IMFs to reduce the dimension and extract the most representative features, thereby improving the generalization ability of the model.

[0233] Statistical features such as mean, variance, skewness, kurtosis, energy, entropy, etc. are extracted from the dimension-reduced IMFs as their time-domain features; Fourier transform (FFT) is performed on each IMF to obtain its frequency spectrum, and frequency-domain features such as power spectral density, frequency center, bandwidth, etc. are extracted from it; the time-domain features and frequency-domain features extracted from different IMFs are combined as the total feature set of the dynamic monitoring data.

[0234] Step 130: Perform correlation analysis on the total feature sets of each flexible section unit, and construct a feature relationship multi-graph (reference can be made to Figure 2 ) based on the results of the correlation analysis to characterize the relationships between the features in the total feature set.

[0235] In the heavy-haul railway system, the multi-source data of "human", "vehicle", "road", and "environment" have strong temporal and spatial correlations. Understanding and mining the spatio-temporal correlation laws of the data will lay a foundation for subsequent data fusion and operation safety analysis. Correlation analysis can help extract useful features, optimize the input structure of the model, select a suitable model, and ultimately improve the accuracy, efficiency, and interpretability of the anomaly monitoring model.

[0236] In some embodiments, the steps of performing correlation analysis on the total feature set include:

[0237] Through the calculation of Pearson correlation coefficient, Spearman correlation coefficient, and lag correlation, analyze the temporal correlation characteristics between different features, and perform causal relationship, cointegration relationship, and dynamic relationship analysis between different features through Granger causality test, cointegration analysis, and vector autoregression;

[0238] Analyze the spatial correlation between different features through spatial distribution pattern, spatial correlation, and spatial lag analysis tools to obtain the spatial dependence relationship between different features.

[0239] In some embodiments, in the feature relationship multi-graph:

[0240] Nodes represent data features;

[0241] Multiple edges between variables represent different types of correlations, and different types of correlations are distinguished by different colors;

[0242] The weights and directions of the edges are used to represent the strength of the association and the causal relationship, respectively.

[0243] Specifically, it includes the following processing steps:

[0244] 1. Temporal correlation analysis:

[0245] Through the calculation of Pearson correlation coefficient, Spearman correlation coefficient, and lag correlation, the temporal correlation characteristics between different features can be analyzed, specifically including: uncorrelated, linearly correlated, non-linearly correlated, or lag correlated. Further, through Granger causality test, cointegration analysis, and vector autoregression, the causal relationship, cointegration relationship, and dynamic relationship between features can be analyzed.

[0246] 2. Spatial correlation analysis:

[0247] The spatial correlation of the data is analyzed through spatial distribution pattern, spatial correlation, and spatial lag analysis tools. The spatial distribution pattern calculates the spatial distribution pattern of a single feature through methods such as global spatial autocorrelation (Moran's I), hotspot analysis (Getis-Ord Gi* statistic), and local spatial autocorrelation, providing a basis for multi-feature analysis. Spatial correlation analyzes the numerical change correlation of two features in spatial positions through covariance and correlation coefficient. Spatial regression analysis can obtain the spatial dependence relationship between features.

[0248] 3. Association representation based on the multi-graph structure:

[0249] Based on the analysis results of the first two steps in this step, considering that there may be various associations between data features (such as temporal association, spatial association, causal association, non-linear association, etc.), the association relationship between variables is represented based on the multi-graph. As shown in the following figure, nodes are used to represent data features, and multiple edges between variables represent different types of associations. Different colors are used to distinguish different types of associations. The weights and directions of the edges can be used to represent the strength of the association and the causal relationship.

[0250] Step 140: Screen out the first features whose centrality meets the preset first condition from the multi-graph of feature relationships of each flexible section unit as key features; analyze the multi-graph of feature relationships of each flexible section unit by using the minimum spanning tree method, retain the association structure that meets the preset second condition between the key features and remove redundant features to obtain the second features; merge the first features and the second features to obtain the preferred feature set of each flexible section unit.

[0251] Feature selection is performed on the Multilayer Graph of feature relationships constructed in step 130. Combining node centrality and the Minimum Spanning Tree (MST) for feature selection can remove redundant features while retaining representative key features.

[0252] In the Multilayer Graph of feature relationships, the centrality of a node can measure its importance. Therefore, the features with the highest centrality can be screened out through the node centrality method, which are the key features for subsequent analysis. The centrality method calculates the centrality index of each node, including degree centrality, betweenness centrality, pagerank centrality, etc. After normalizing multiple centrality indexes, the sum of their squared values is taken as the total centrality value, and ranking is performed according to this value, and the top N items are selected as important features.

[0253] The Minimum Spanning Tree method (MST) is used to analyze the Multilayer Graph of feature relationships. The MST removes redundant features by retaining the minimum association structure between key features. The features screened by the MST and the node centrality method are combined to ensure that feature selection covers key features and effectively avoids redundant features.

[0254] Step 150: Based on the optimal feature set of each flexible section unit, the operating state of the heavy-haul train is monitored for anomalies through a preset context-aware model, and the operation safety of the heavy-haul train is evaluated according to the monitoring results.

[0255] Since the amount of monitoring data during the train operation process is large and abnormal data is very rare, it means that the representativeness of abnormal situations in the dataset is very low. The traditional point-by-point error method is difficult for the model to capture complex time-dependent relationships and abnormal patterns across time steps due to the lack of a comprehensive description of the time series context.

[0256] For the typical scenario of sparse abnormal distribution and multi-dimensional time series monitoring data in the operation of heavy-haul trains, complex dependencies between time series are extracted by using context-aware modeling, and at the same time, the abnormal patterns of the overall window are evaluated through sequence-level loss, rather than being limited to single-point errors. The specific implementation steps include:

[0257] (1) Data preprocessing:

[0258] The multi-dimensional spatio-temporal sequence features selected in step 140 have been divided into sections in the spatial dimension according to the line features in the second step of step 120. In the data preprocessing stage of anomaly detection, the division in the time dimension needs to be considered. Since the division of the spatial flexible unit sections is usually long and of different lengths, considering factors such as the vehicle running speed, the sampling frequency of monitoring data, and the impact of time step division on modeling, the data is further subdivided in the time dimension, and the data is divided into windows with fixed event steps, which can be expressed as The form (where N is the number of samples, T is the length of the time window, and F is the dimension of the feature sequence) is used as the model input data, that is, each time step T contains F features.

[0259] (2) Construct a preset context-aware model:

[0260] The present invention uses a transformer as the preset context-aware model. In order to adapt to the data structure after spatial segmentation, it is necessary to model the dependencies in both the time dimension and the spatial dimension simultaneously. By introducing a two-layer self-attention mechanism, the dependencies in the time dimension and the spatial dimension are processed separately.

[0261] (3) Construction of the sequence-level loss function:

[0262] The time series is reconstructed through the model output, and the time reconstruction error Lt can be obtained by calculating the error between it and the true value. Similarly, the spatial features are reconstructed through the model output, and the spatial reconstruction error Ls can be obtained by calculating the error between it and the true value. The comprehensive reconstruction loss Lreconstruction = αLt + βLs can be obtained in a weighted cumulative manner, where α and β are hyperparameters used to balance the reconstruction importance of time and spatial features.

[0263] (4) Abnormal detection method process:

[0264] Abnormal detection includes two stages: training and detection. In the training stage, normal data is used to train the anomaly monitoring model, minimizing the above joint loss function, and at the same time recording the comprehensive loss distribution of each window. In the detection stage, the joint loss value of each window is calculated for the newly input data, and anomalies are detected simultaneously in the time dimension and the spatial dimension: based on the time dimension, it can be determined whether a certain time window is abnormal, and based on the spatial dimension, it can be located whether a specific segmented section is abnormal.

[0265] The method disclosed in this embodiment can evaluate the actual state of train operation more comprehensively compared with the prior art. By fully considering the background and environmental factors of the monitoring data during the analysis process - including line information, driver operation information, environmental temperature and humidity, etc., the accuracy and applicability of the analysis results are ensured. Moreover, it can also help users understand and analyze the operation information of heavy-haul trains more comprehensively, so as to provide a comprehensive safety assessment of heavy-haul train operation, and at the same time improve the intelligence level of multi-dimensional monitoring data analysis to help users better improve the operation safety of heavy-haul trains and guide the optimization of heavy-haul train operation. Further, the heavy-haul train operation anomaly detection method based on context awareness and window-level loss function enhances the model's sensitivity to spatial anomalies by modeling the features of spatial segmented sections; by jointly considering time-space context, it can simultaneously capture the time dependencies in the time series and the correlation relationships between spatial segmented sections; in addition, it has the characteristics of strong flexibility and can adjust the segmentation strategy and feature extraction method according to different application scenarios.

[0266] Embodiment 3:

[0267] Another embodiment of the present application relates to a heavy-haul train operation safety evaluation system.

[0268] The implementation details of the heavy-haul train operation safety evaluation system of this embodiment are specifically described below. The following content is only the implementation details provided for convenient understanding and is not necessary for implementing this solution. The heavy-haul train operation safety evaluation system provided in this embodiment includes:

[0269] A data acquisition module, configured to collect the monitoring data of the heavy-haul train, and classify the monitoring data based on the change situation of the monitoring data to obtain slow-changing data and time-series data; wherein, the monitoring data includes at least one of the following dimension data: driver operation information, heavy-haul train system, operation line, and external environment;

[0270] A feature set determination module, configured to perform flexible section division on the operation line based on the line features of the heavy-haul train operation line to obtain flexible section units, and perform feature extraction on the data collected for each flexible section unit to obtain the total feature set corresponding to each flexible section unit;

[0271] A feature relationship multi-graph construction module, configured to perform correlation analysis on the total feature sets of each flexible section unit, and construct a feature relationship multi-graph for characterizing the relationships between the features in the total feature set based on the correlation analysis results;

[0272] A preferred feature set determination module, configured to screen out the first features whose centrality meets the preset first condition from the feature relationship multi-graphs of each flexible section unit as key features through the node centrality method; analyze the feature relationship multi-graphs of each flexible section unit by using the minimum spanning tree method, retain the correlation structures between the key features that meet the preset second condition and remove redundant features to obtain second features; merge the first features and the second features to obtain the preferred feature set of each flexible section unit.

[0273] A monitoring and evaluation module, configured to perform abnormal monitoring on the running state of the heavy-haul train based on the preferred feature sets of each flexible section unit through a preset context awareness model, and evaluate the operation safety of the heavy-haul train according to the monitoring results.

[0274] In some alternative embodiments, collecting the monitoring data of the heavy-haul train includes:

[0275] Based on the research of historical literature, the operation mechanism of heavy-haul trains and the hierarchical task analysis of heavy-haul train drivers, as well as the analysis of historical accidents and related reports of heavy-haul trains, obtain the influencing factors of heavy-haul train operation safety;

[0276] Classify the factors affecting operation safety according to a preset classification method to obtain the classification result of the factors affecting operation safety;

[0277] Mine data sources based on the factors affecting operation safety and the classification result of the factors affecting operation safety and historical data to obtain the classification of data sources of the factors affecting operation safety;

[0278] Collect the monitoring data of heavy-haul trains according to the classification of data sources of the factors affecting operation safety.

[0279] In some alternative embodiments, the preset classification method includes:

[0280] Driver and operation behavior influencing factors, train system influencing factors, line condition influencing factors, and external environment influencing factors.

[0281] In some alternative embodiments, the slowly varying data includes data with slow changes or completely static data; the time series data includes data with dynamic changes.

[0282] In some alternative embodiments, time and line kilometer post information are associated in the time series data; a piece of data is recorded at the start and end points of each operation trip of the heavy-haul train and at preset collection positions respectively to expand the slowly varying data into data with spatio-temporal information that remains unchanged for a long time.

[0283] Among them, the preset collection positions can be set according to actual requirements.

[0284] In some alternative embodiments, the feature set determination module includes: an extraction unit, a first generation unit, a second generation unit, and a merging unit; among them,

[0285] The extraction unit is used to perform time-domain feature extraction and frequency-domain feature extraction on the data collected by each flexible section unit respectively; among them:

[0286] The first generation unit is used to extract the trend characteristics of the slowly varying data collected by each flexible section unit through linear regression, extract the periodic component characteristics through Fourier transform, and generate a first feature set based on the time-domain statistical characteristics, trend characteristics, and periodic component characteristics of the slowly varying data;

[0287] The second generation unit is used to perform time-domain analysis and frequency-domain analysis on the time series data collected by each flexible section unit respectively to reflect the change characteristics of the data in time and the change characteristics in frequency, and generate a second feature set based on the change characteristics in time and the change characteristics in frequency;

[0288] The merging unit is used to use the set of the first feature set and the second feature set as the total feature set.

[0289] In some alternative embodiments, the feature relationship multi-graph construction module includes: a first analysis unit and a second analysis unit; wherein,

[0290] The first analysis unit is configured to perform time correlation characteristic analysis between different features through Pearson correlation coefficient, Spearman correlation coefficient and lag correlation calculation, and perform causal relationship, cointegration relationship and dynamic relationship analysis between different features through Granger causality test, cointegration analysis and vector autoregression;

[0291] The second analysis unit is configured to analyze the spatial correlation between different features through spatial distribution pattern, spatial correlation and spatial lag analysis tools, and obtain the spatial dependence relationship between different features.

[0292] In some alternative embodiments, in the feature relationship multi-graph:

[0293] Nodes represent data features;

[0294] Multiple edges between variables represent different types of correlations, and different types of correlations are distinguished by different colors;

[0295] The weight and direction of the edge are used to represent the strength of the association and the causal relationship respectively.

[0296] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of each module in this heavy-haul train operation safety evaluation can refer to the corresponding processes in the foregoing method embodiments, and this embodiment will not be repeated here.

[0297] It is worth mentioning that each module involved in this embodiment is a logical module. In actual applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, units not closely related to solving the technical problems proposed in this application are not introduced in this embodiment, but this does not mean that there are no other units in this embodiment.

[0298] Embodiment 4:

[0299] Another embodiment of the present application relates to an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the heavy-haul train operation safety evaluation method in the foregoing embodiments.

[0300] Among them, the memory and the processor are connected in a bus manner. The bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and the memory together. The bus may also connect various other circuits together, such as peripheral devices, voltage regulators, and power management circuits, etc., which are well known in the art, and thus will not be further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver may be a component or multiple components, such as multiple receivers and transmitters, and provides a unit for communicating with various other devices on the transmission medium. The data processed by the processor is transmitted on the wireless medium through the antenna. Further, the antenna also receives data and transmits the data to the processor.

[0301] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. The memory can be used to store the data used by the processor when executing operations.

[0302] Embodiment Five:

[0303] Another embodiment of the present application relates to a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the above method embodiments are implemented.

[0304] That is, those skilled in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by instructing relevant hardware through a program. The program is stored in a storage medium, including several instructions for causing a device (which may be a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs, etc., which can store program codes.

[0305] In some embodiments of the present application, a computer program product is further provided, including a computer program, and when the computer program is executed by a processor, the steps of the method described in the above embodiments are implemented.

[0306] Those of ordinary skill in the art can understand that the above embodiments are specific embodiments for implementing the present application, and in practical applications, various changes can be made in form and details without departing from the spirit and scope of the present application.

Claims

1. A heavy-load train operation safety evaluation method, characterized in that: include: Collect monitoring data of heavy-load trains, and classify the monitoring data based on changes in the monitoring data to obtain slowly changing data and time series data; wherein the monitoring data includes data of at least one of the following dimensions: driver operation information, heavy-load train system, operating line, and external environment; Based on the line characteristics of the heavy-load train operation line, the operation line is divided into flexible sections to obtain flexible section units, and feature extraction is performed on the data collected from each flexible section unit to obtain a total feature set corresponding to each flexible section unit; Performing association analysis on the total feature set of each flexible segment unit, and constructing a feature relationship multigraph for characterizing the relationship between each feature in the total feature set based on the association analysis result; A first feature whose centrality satisfies a preset first condition is selected from the feature relationship multigraph of each flexible segment unit by using a node centrality method, as a key feature; a minimum spanning tree method is used to analyze the feature relationship multigraph of each flexible segment unit, and the association structure between key features that satisfies a preset second condition is retained and redundant features are removed to obtain a second feature; the first feature and the second feature are merged to obtain a preferred feature set of each flexible segment unit; Based on the preferred feature set of each flexible section unit, the running state of the heavy-load train is monitored for abnormality through a preset context perception model, and the operating safety of the heavy-load train is evaluated according to the monitoring results.

2. The method according to claim 1, characterized in that The collecting of monitoring data of heavy-load trains includes: Based on the historical literature research, the operation mechanism of heavy-haul trains and the hierarchical task analysis of heavy-haul train drivers, as well as the analysis of historical accidents and related reports of heavy-haul trains, the factors affecting the safety of heavy-haul train operation are obtained; Classify the factors affecting operation safety according to a preset classification method to obtain the classification results of the factors affecting operation safety; According to the manipulation safety influencing factors and the classification results of the manipulation safety influencing factors and based on historical data, data source mining is performed to obtain a classification of manipulation safety influencing factor data sources; The monitoring data of the heavy-load train is collected and classified according to the data source of the operation safety influencing factors.

3. The method according to claim 2, characterized in that The preset classification method includes: Factors affecting driver and operating behavior, factors affecting the train system, factors affecting line conditions, and factors affecting the external environment.

4. The method according to claim 1, characterized in that: The slowly varying data includes data that changes slowly or is completely static; The time series data includes dynamically changing data.

5. The method according to claim 4, characterized in that The time and route kilometer mark information are associated with the time series data; A piece of data is recorded at the starting point and the end point of the heavy-load train trip and the preset collection position respectively, so as to expand the slowly varying data into data with time and space information that remains unchanged for a long time.

6. The method according to claim 1, characterized in that The data collected from each flexible segment unit are respectively subjected to feature extraction to obtain a total feature set corresponding to each flexible segment unit, including: The data collected by each flexible segment unit are respectively subjected to time domain feature extraction and frequency domain feature extraction; wherein: For the slowly varying data collected by each flexible segment unit, the trend characteristics are extracted by linear regression, the periodic component characteristics are extracted by Fourier transform, and a first feature set is generated based on the time domain statistical characteristics of the slowly varying data, the trend characteristics and the periodic component characteristics; For the time series data collected by each flexible segment unit, time domain analysis and frequency domain analysis are respectively performed to reflect the time variation characteristics and frequency variation characteristics of the data, and a second feature set is generated based on the time variation characteristics and the frequency variation characteristics; The set of the first feature set and the second feature set is taken as the total feature set.

7. The method according to claim 1, characterized in that The step of performing association analysis on the total feature set comprises: Through the calculation of Pearson correlation coefficient, Spearman correlation coefficient and lagged correlation, the time correlation characteristics between different features are analyzed, and the causal relationship, cointegration relationship and dynamic relationship between different features are analyzed through Granger causality test, cointegration analysis and vector autoregression; The spatial correlation between different features is analyzed through spatial distribution pattern, spatial correlation and spatial hysteresis analysis tools to obtain the spatial dependence relationship between different features.

8. The method according to claim 1, characterized in that In the feature relation multigraph: Nodes represent data features; Multiple edges between variables represent different types of associations, and different types of associations are distinguished by different colors; The weight and direction of the edge are used to represent the strength and causality of the association, respectively.

9. A heavy-load train operation safety evaluation system, characterized in that: include: A data acquisition module, used to collect monitoring data of heavy-load trains, and classify the monitoring data based on changes in the monitoring data to obtain slowly changing data and time series data; wherein the monitoring data includes data of at least one of the following dimensions: driver operation information, heavy-load train system, operating line and external environment; A feature set determination module, used to perform flexible segment division on the operating line based on the line characteristics of the heavy-load train operating line to obtain flexible segment units, and perform feature extraction on the data collected from each flexible segment unit to obtain a total feature set corresponding to each flexible segment unit; A feature relationship multigraph construction module, used for performing association analysis on the total feature set of each flexible segment unit, and constructing a feature relationship multigraph for characterizing the relationship between each feature in the total feature set based on the association analysis result; The preferred feature set determination module is used to select a first feature whose centrality meets a preset first condition from the feature relationship multigraph of each flexible segment unit by using a node centrality method, as a key feature; analyze the feature relationship multigraph of each flexible segment unit by using a minimum spanning tree method, retain the association structure between key features that meets a preset second condition and remove redundant features to obtain a second feature; merge the first feature and the second feature to obtain a preferred feature set of each flexible segment unit; A monitoring and evaluation module is used to perform abnormal monitoring of the running status of the heavy-load train based on the preferred feature set of each flexible section unit through a preset context perception model, and evaluate the operating safety of the heavy-load train according to the monitoring results.

10. An electronic device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method according to any one of claims 1 to 8.

Images