Rail transit data analysis method and system based on cloud computing

Abstract

The invention discloses a rail transit data analysis method and system based on cloud computing, and relates to the technical field of rail transit data analys.The rail transit data analysis method comprises the steps that global operation data of a target rail transit network is collected, a dynamic panoramic topological graph is constructed based on the global operation data, the dynamic panoramic topological graph is divided into N traffic area sub-graphs, and the N traffic area sub-graphs are used for analyzing the traffic area sub-graphs; configuring elastic cloud computing nodes mapped to the N traffic area sub-graphs, distributing global operation data to the corresponding elastic cloud computing nodes, performing global operation data initialization screening and cloud association conversion through a data clustering algorithm in combination with a cloud computing processing model, determining abnormal data distribution, and after a cloud platform monitors a key event, sending the key event to the cloud platform. Summarizing the abnormal data of each elastic cloud computing node, mining the generality of the abnormal data, determining and predicting the distribution of the abnormal data in combination with the real-time passenger flow evolution trend, obtaining an abnormal distribution diagram, mapping the abnormal distribution diagram in the dynamic panoramic topological graph in real time, and generating global early warning information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rail transit data analysis, and in particular to a rail transit data analysis method and system based on cloud computing. BACKGROUND

[0002] With the continuous expansion of the scale of the rail transit network and the continuous improvement of the complexity of operation, data-driven fine management has become the core demand of the industry development. At present, a large amount of multi-source data is generated in the process of operation of the rail transit system, and the traditional dimensions of multi-source data cover passenger flow distribution, train operation state and equipment working condition, and the public opinion feedback in social media also gradually becomes a key factor affecting the operation efficiency. However, the existing data analysis technology still has significant bottlenecks: on the one hand, the degree of data integration is insufficient, most systems only analyze single type of data in isolation, and cannot realize the deep linkage of multi-source data, resulting in prominent data island phenomenon and difficulty in capturing cross-modal correlation anomalies; on the other hand, the allocation of computing resources is static, which cannot adapt to the passenger flow tide changes in peak and off-peak hours and holiday scenes, and there are often problems of insufficient computing resources in high-load areas and waste of resources in low-load areas; in addition, the abnormal processing is mostly limited to real-time identification, lacks the prediction ability of abnormal diffusion path, and the warning information mostly only includes basic location and type without integrating decision support dimensions such as economic impact evaluation, resulting in insufficient response and passenger experience optimization of the operation party.

[0003] At present, the Chinese invention with the application number CN120589070A discloses a rail transit remote monitoring method and system based on artificial intelligence, which realizes the positioning and remote monitoring of equipment failure through artificial intelligence technology, and assists maintenance and disposal combined with augmented reality technology. The above technology cannot meet the fine analysis needs of the whole rail transit operation: firstly, the data coverage dimension is limited, focusing on the monitoring and fault diagnosis of equipment operation data, without including social media public opinion data in the analysis framework, which cannot capture the cross-modal collaborative anomalies of public opinion heat, passenger flow fluctuation and equipment load; secondly, there is a lack of dynamic adaptive computing power scheduling mechanism, which does not perform regional segmentation and elastic allocation of computing power based on the real-time changes of operation data, making it difficult to cope with the instantaneous data processing pressure brought by passenger flow tide; thirdly, the abnormal response is limited to fault disposal after the event, without building a prediction model combining passenger flow evolution trend and abnormal common characteristics, which cannot predict the potential risks of congestion diffusion and fault chain reaction in advance; fourthly, the warning output dimension is single, without realizing the visualization mapping and multi-end synchronous pushing of abnormal distribution, and also lacks the quantitative evaluation of operation loss, making it difficult to support the precise decision of the operation party and the travel guidance of passengers. SUMMARY

[0004] The technical problems solved by the present application are that the prior art is difficult to integrate passenger flow, equipment and public opinion data, cannot dynamically configure computing power to adapt to tidal passenger flow, lacks abnormal diffusion prediction ability, and the early warning lacks economic evaluation and multi-end pushing, and it is difficult to support global fine operation decision-making of rail transit.

[0005] To solve the above technical problems, the present application provides the following technical solutions: a rail transit data analysis method based on cloud computing, comprising the following steps: Step S1, collecting global operation data of a target rail transit network, and constructing a dynamic panoramic topology graph based on the global operation data; Step S2, dividing the dynamic panoramic topology graph into N traffic area subgraphs; Step S3, configuring elastic cloud computing nodes mapped to the N traffic area subgraphs, recording the service traffic area subgraph range, the upper limit of computing power and the data receiving port of each elastic cloud computing node, and monitoring the passenger flow changes and public opinion heat of each traffic area subgraph in real time; Step S4, distributing the global operation data to the corresponding elastic cloud computing nodes, combining a cloud computing processing model, and performing global operation data initialization filtering and cloud association conversion through a data clustering algorithm to determine the abnormal data distribution; Step S5, after the cloud platform detects a key event, the abnormal data of each elastic cloud computing node is summarized, the commonality of the abnormal data is mined, the predicted abnormal data distribution is determined combined with the real-time passenger flow evolution trend, and the abnormal distribution graph is obtained; Step S6, real-time mapping the abnormal distribution graph in the dynamic panoramic topology graph to generate global early warning information.

[0006] Preferably, the step S1 comprises the following sub-steps: Step S101, collecting global operation data of a target rail transit network in real time, the global operation data comprising passenger flow data, train operation state data, equipment working condition data and social media public opinion data; Step S102, constructing an initial panoramic topology graph of the rail transit network based on the global operation data; Step S103, dynamically adjusting the initial panoramic topology graph according to the real-time flow of the global operation data to obtain a dynamic panoramic topology graph.

[0007] Preferably, the step S2 comprises the following sub-steps: Step S201, under the constraint of a preset principle tolerance, combining real-time global operation data to preliminarily divide the dynamic panoramic topology graph to obtain N initial traffic area subgraphs; The preset principle tolerance comprises a passenger flow peak balance principle and a weak dependence principle; The passenger flow peak balance principle comprises: The initial traffic area subgraph passenger flow load fluctuation threshold is within a first preset range; The weight adjustment of the passenger flow peak or public opinion early warning period to the passenger flow peak balancing principle is to the first percentage; The weak dependence principle includes that the initial traffic area subgraph boundary division preferentially maintains the integrity of the same line continuous station, and the initial traffic area subgraph inter-data interaction frequency is within a second preset range; Step S202, according to the real-time passenger flow change and the update of the dynamic panoramic topology graph, the coverage range of the initial traffic area subgraph is adaptively optimized, and a traffic area subgraph is obtained.

[0008] Preferably, the logic of adaptively optimizing the coverage range of the initial traffic area subgraph includes: Real-time monitoring of the passenger flow density of each initial traffic area subgraph, if the passenger flow of one of the initial traffic area subgraphs is greater than a preset load threshold, the coverage range of the one of the initial traffic area subgraphs is reduced and the surrounding area is divided to the adjacent initial traffic area subgraph; If the load difference between the initial traffic area subgraphs is greater than a second percentage due to passenger flow tide, the initial traffic area subgraph boundary is adjusted; If the public opinion heat of one of the initial traffic area subgraphs is greater than a first preset warning line, the coverage range of the one of the initial traffic area subgraphs is preferentially ensured to completely match the station involved in the public opinion.

[0009] Preferably, the step S3 includes the following sub-steps: Step S301, evaluating the service bearing demand of each traffic area subgraph, specifically including: Combining the real-time passenger flow load, data processing amount and public opinion correlation degree of the traffic area subgraph, the computing power benchmark value of the cloud computing node required by each traffic area subgraph is determined; Step S302, according to the computing power benchmark value configuration corresponding to the elastic cloud computing node, the dynamic mapping relationship between the elastic cloud computing node and the traffic area subgraph is established, the service traffic area subgraph range, the computing power upper limit and the data receiving port of each elastic cloud computing node are recorded, and stored to the elastic cloud computing node mapping library of the cloud platform; Step S303, real-time monitoring of the passenger flow change and the public opinion heat of each traffic area subgraph, specifically including: If the traffic area subgraph load is greater than a third percentage of the computing power benchmark value, the corresponding elastic cloud computing node is expanded; If the traffic area subgraph load is less than a fourth percentage of the computing power benchmark value, the elastic cloud computing node computing power is reduced.

[0010] Preferably, the step S4 includes the following sub-steps: In step S401, the global operation data of multi-period rail transit is read simultaneously through a cloud computing processing model, abnormal features are mined, a pre-check database accessed in parallel by multiple elastic cloud computing nodes is built, the pre-check database updates thresholds according to seasonal passenger flow and public opinion hotspots, and the abnormal features include passenger flow fluctuation thresholds, equipment parameter thresholds and negative public opinion sentiment thresholds. The cloud computing processing model includes a distributed preprocessing layer and a cloud conversion layer, the preprocessing layer integrates a data cleaning module, and the cloud conversion layer embeds a data clustering algorithm based on machine learning; In step S402, the global operation data is distributed to each elastic cloud computing node according to the correspondence relationship of the traffic area subgraph, the pre-check database is called from the preprocessing layer, and the passenger flow data, equipment working condition data and social media public opinion data are traversed for feature correction in parallel by multiple elastic cloud computing nodes, so as to screen out abnormal data of rail transit matching the abnormal features; In step S403, the screened abnormal data is transmitted into the cloud conversion layer, classified according to the main data mode through the machine learning clustering algorithm, the intra-modal data correlation rules and cross-modal correlation rules are established, and the abnormal data distribution is determined, the main data mode including passenger flow congestion, equipment failure and public opinion early warning.

[0011] Preferably, the step S5 includes the following sub-steps: In step S501, the cloud platform monitors key events in the rail transit network in real time, and starts a data aggregation process once a key event is detected, when the data aggregation process is started, the cloud platform will notify all elastic cloud computing nodes corresponding to the traffic area subgraph, so that the elastic cloud computing nodes upload the abnormal data processed by themselves to the cloud platform, and the elastic cloud computing nodes preferentially upload the abnormal data of the event-associated area; The key events include: The third percentile of the real-time passenger flow of the station exceeding the preset passenger flow threshold, the train failure and stagnation, and the negative public opinion sentiment index of the rail transit related on the social media being greater than the second preset warning line; The event-associated area includes: When the train fails, the abnormal data of the stations to be passed through by the failed train is preferentially uploaded; When the public opinion early warning, the abnormal data of the key stations mentioned in the public opinion is preferentially uploaded; In step S502, after the cloud platform aggregates the abnormal data uploaded by each elastic cloud computing node, the correlation strength between the elastic cloud computing nodes is learned according to the machine learning algorithm, the common features of the abnormal data are analyzed, and the correlation graph of the abnormal data and the key events is constructed; The common features include intra-modal common features and cross-modal common features; The intra-modal common features include the propagation path features of passenger flow congestion and the chain reaction types of equipment failure. The cross-modal common feature includes a cooperative change rule of public opinion heat and passenger flow congestion degree. In step S503, according to the association graph, in combination with a real-time passenger flow evolution trend, an abnormal distribution is predicted to obtain predicted abnormal data, real-time abnormal data and predicted abnormal data are integrated, and an abnormal distribution graph is generated. The real-time passenger flow evolution trend includes passenger flow direction and passenger flow growth rate.

[0012] Preferably, the logic of predicting the abnormal distribution comprises: The passenger flow direction is fused with the spatial feature, specifically including: The passenger flow direction is converted into a spatial adjacency weight matrix, the weight of the station pointed by the flow direction is a first spatial connection weight, and the weight of the station pointed by the opposite direction of the flow direction is a second spatial connection weight. The first spatial connection weight and the second spatial connection weight are taken as the spatial input of the prediction model, and the first spatial connection weight is greater than the second spatial connection weight; The passenger flow growth rate is fused with the time feature, specifically including: The passenger flow growth rate is taken as the time sequence input of the prediction model. When the passenger flow growth rate is greater than a third preset range, it is predicted that the abnormality covers adjacent first number of stations in a first unit of time. If the passenger flow growth rate is less than the third preset range, it is predicted that the abnormality covers adjacent second number of stations in a second unit of time, and the first number is greater than the second number. The spatial feature is a station and line association feature, and the time feature is a passenger flow evolution feature. The common feature in the association graph is introduced as a constraint condition of the prediction model. If it is an intra-modal common feature, the diffusion range is corrected according to the propagation path feature of passenger flow congestion and the chain reaction type of equipment failure. If it is a cross-modal common feature, the prediction result is adjusted in combination with real-time public opinion data. When predicting by the prediction model, the stations in the event association area are preferentially included in the priority coverage range; The stations and lines covered by the abnormality are predicted in stages according to a preset time period, and the predicted abnormal data of each time period is generated.

[0013] Preferably, the step S6 comprises the following sub-steps: In step S601, a coordinate mapping relationship between the abnormal distribution graph and the dynamic panoramic topology graph is established, and the real-time abnormal area and the predicted abnormal area in the abnormal distribution graph are respectively matched with the corresponding station and line positions in the dynamic panoramic topology graph. The matching mode includes: using differential visual identification according to the abnormal type, marking passenger flow congestion with red, marking equipment failure with yellow, and marking public opinion warning with orange. In step S602, global early warning information is generated based on the mapping result of the coordinate mapping relationship between the abnormal distribution map and the dynamic panoramic topology map, the early warning information including an abnormal type, an abnormal occurrence position, an abnormal influence range, and an economic influence evaluation; The economic influence evaluation includes an increase in per capita commuting cost caused by train delay and an estimation of peripheral business revenue loss caused by station congestion; The early warning levels are divided according to the abnormal influence range, and are divided into a first-level early warning, a second-level early warning, and a third-level early warning in descending order; The first-level early warning is a global associated abnormality, the second-level early warning is a cross-line abnormality, and the third-level early warning is a single-station abnormality; In step S603, the global early warning information is synchronized to a cloud platform visual display interface, and is pushed to a rail transit operation management system, a station broadcasting system, and a passenger service APP through an interface, and the early warning information is refreshed in real time according to the update of the dynamic panoramic topology map.

[0014] A rail transit data analysis system based on cloud computing includes a construction module, a segmentation module, a configuration module, a processing module, a monitoring module, and an early warning module. The construction module is configured to collect global operation data of a target rail transit network in real time, construct an initial panoramic topology map based on the global operation data, and dynamically adjust the initial panoramic topology map according to real-time update data to obtain a dynamic panoramic topology map. The segmentation module is configured to segment the dynamic panoramic topology map into N traffic area subgraphs under the constraint of a preset principle tolerance. The configuration module is configured to configure elastic cloud computing nodes mapped to the N traffic area subgraphs. The processing module is configured to distribute the global operation data to corresponding elastic cloud computing nodes, combine a cloud computing processing model, and perform monitoring data initialization filtering and cloud association conversion through a data clustering algorithm to determine abnormal data distribution. The monitoring module is configured to monitor key events in real time, aggregate abnormal data of each elastic cloud computing node, mine commonalities of the abnormal data, determine a predicted abnormal distribution in combination with a real-time passenger flow evolution trend, and obtain an abnormal distribution map. The early warning module is configured to map the abnormal distribution map in the dynamic panoramic topology map to generate global early warning information.

[0015] The beneficial effects of the present application: by integrating passenger flow, train operation state, equipment working condition and social media public opinion data, combining the space-time synchronization characteristics of dynamic panoramic topological graph, effectively solving the traditional multi-source data island problem, improving data integrity, improving cross-modal anomaly recognition rate, accurately capturing the chain correlation from the rise of public opinion heat to the intensification of passenger congestion to the overload of equipment load, avoiding the one-sidedness of single data analysis. At the same time, based on the elastic cloud computing node scheduling mechanism of the traffic area subgraph load, realize the dynamic adaptation of computing power, improve the utilization rate of computing power resources, reduce the waste of invalid computing power, and shorten the expansion response time of computing power in peak period, which not only avoids the data processing delay of high-load area, but also reduces the energy consumption in off-peak period, balances efficiency and economy. Using spatiotemporal graph neural network combined with common constraint prediction algorithm, the abnormal treatment is upgraded from passive response to active prediction, improving the accuracy of abnormal prediction, which can predict the abnormal diffusion range and time in advance, shorten the emergency response time, and the correlation graph can intuitively present the linkage of events, abnormalities and affected areas, helping the operator to quickly locate the core problem. The multi-end collaborative early warning system constructed can synchronize the early warning information to the operation management system, station broadcast and passenger service APP, shorten the delay of multi-end synchronization, speed up the passengers to obtain abnormal information, improve passenger satisfaction, and the operator can also use economic impact evaluation to quantify the loss, decision-making is more accurate, which provides core support for fine operation of large-scale rail transit network. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A step flow chart of a rail transit data analysis method based on cloud computing provided for an embodiment of the present application; Figure 2 A basic flow diagram of a rail transit data analysis system based on cloud computing provided for an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all embodiments.

[0018] Embodiment 1, refer to Figure 1 , provides a rail transit data analysis method based on cloud computing, characterized in that it comprises the following steps: Step S1, collecting global operation data of the target rail transit network, and constructing a dynamic panoramic topological graph based on the global operation data.

[0019] Step S2, dividing the dynamic panoramic topological graph into N traffic area subgraphs.

[0020] Step S3, configure elastic cloud computing nodes mapped to N traffic area subgraphs, record the service traffic area subgraph range, upper limit of computing power and data receiving port of each elastic cloud computing node, and monitor the passenger flow change and public opinion heat of each traffic area subgraph in real time.

[0021] Step S4, distribute the global running data to the corresponding elastic cloud computing nodes, combine the cloud computing processing model, and perform global running data initialization filtering and cloud association conversion through a data clustering algorithm to determine the abnormal data distribution.

[0022] Step S5, after the cloud platform detects a key event, the abnormal data of each elastic cloud computing node is summarized, the commonality of the abnormal data is mined, the predicted abnormal data distribution is determined combined with the real-time passenger flow evolution trend, and the abnormal distribution map is obtained.

[0023] Step S6, real-time mapping of the abnormal distribution map in the dynamic panoramic topology map to generate global early warning information.

[0024] Step S1 includes the following sub-steps: Step S101, real-time collection of global running data of the target rail transit network, the global running data including passenger flow data, train running state data, equipment working condition data and social media public opinion data, realizing multi-source data collection and unified aggregation at a second level, improving data integrity and avoiding information gap problems of traditional single data collection.

[0025] Step S102, based on the global running data, an initial panoramic topology map of the rail transit network is constructed. The initial topology map can intuitively present the line correlation, providing a precise spatial basis for subsequent regional segmentation, and the topology construction efficiency is improved compared with traditional CAD drawing. The initial panoramic topology map is constructed by using the graph structure modeling tool NetworkX, setting the station as the node and the line as the edge, and marking the node attributes and edge attributes. The node attributes include the station type and the transfer level, and the edge attributes include the line length and the design speed.

[0026] The calculation formula of the node weight is: W n =C t ×C d ; Wherein, W n is the node weight, C t is the transfer level coefficient, and C d为 is the daily passenger flow coefficient.

[0027] The calculation formula of the edge weight is: W e =(L / V d )×F; Wherein, W e is the edge weight, L represents the line length, and Vd Representative design speed, F is the train passing frequency.

[0028] Step S103, according to the real-time flow of the global operation data, the initial panoramic topology is dynamically adjusted, and the dynamic panoramic topology is obtained. The dynamic panoramic topology is identified based on real-time train scheduling and passenger flow direction. The directed identification is dynamically updated according to the scheduling instruction, train running speed and passenger flow boarding and alighting.

[0029] The real-time updating rule of the dynamic directed identification includes: According to the scheduling instruction trigger, if the train temporary speed limit instruction is received, the directed identification color of the corresponding line changes from green to yellow; Linkage with train speed, when the train speed is less than 20km / h, it is entering station or failure, the line width increases, when the train speed is greater than 80km / h, it is normal driving, the line width returns to default; If the number of passengers boarding the platform is greater than the number of passengers alighting, it indicates that the passenger flow is concentrated, the station node flashes red, otherwise it flashes green.

[0030] The operation personnel can dynamically identify the line through the initial panoramic topology Figure 1 The dynamic information acquisition efficiency of the initial panoramic topology is improved compared with the traditional static topology, and the space-time synchronous topology basis is provided for subsequent initial traffic area subgraph segmentation and abnormal mapping.

[0031] Step S2 includes the following substeps: Step S201, under the constraint of the preset principle tolerance, combining real-time global operation data, the dynamic panoramic topology is preliminarily segmented to obtain N initial traffic area subgraphs.

[0032] The preset principle tolerance includes the passenger flow peak balance principle and the weak dependence principle.

[0033] The passenger flow peak balance principle includes: The initial traffic area subgraph passenger flow load fluctuation threshold is within the first preset range.

[0034] The weight of the passenger flow peak balance principle is improved by the first percentage during the passenger flow peak or public opinion early warning period, so as to ensure that the processing pressure of each initial traffic area subgraph tends to be balanced during the high load period.

[0035] The calculation formula of the initial traffic area subgraph load balance degree is: L0=1-(F 0max -F 0min ) / F0; Wherein, L0 is the initial traffic area subgraph load balance degree, F 0max is the maximum passenger flow load of the initial traffic area subgraph, F 0minF0 is the minimum passenger flow load of the initial traffic area subgraph.

[0036] The weak dependence principle includes that the data interaction frequency between the initial traffic area subgraphs is within a second preset range, and the boundary division of the initial traffic area subgraph preferentially maintains the integrity of the same line continuous station.

[0037] In step S202, the coverage range of the initial traffic area subgraph is adaptively optimized according to real-time passenger flow changes and the updating situation of the dynamic panoramic topology graph, and a traffic area subgraph is obtained.

[0038] The logic of adaptively optimizing the coverage range of the initial traffic area subgraph includes: Real-time monitoring of the passenger flow density of each initial traffic area subgraph, if the passenger flow of one of the initial traffic area subgraphs is greater than a preset load threshold, the coverage range of one of the initial traffic area subgraphs is reduced and the surrounding area is divided to the adjacent initial traffic area subgraph.

[0039] If the load difference between the initial traffic area subgraphs is greater than a second percentage due to passenger flow tide, the boundary of the initial traffic area subgraph is adjusted.

[0040] If the heat of public opinion associated with one of the initial traffic area subgraphs is greater than a preset warning line, the coverage range of one of the initial traffic area subgraphs is preferentially ensured to completely match the station involved in public opinion.

[0041] After optimization, the load balancing degree of the initial traffic area subgraph is improved, the response speed of public opinion associated with abnormalities is improved, and the problem that the traditional static segmentation cannot adapt to passenger flow tide and public opinion burst is solved.

[0042] Step S3 includes the following sub-steps: Step S301, evaluating the business bearing demand of each traffic area subgraph, specifically including: Combining the real-time passenger flow load of the traffic area subgraph, the data processing amount and the public opinion correlation degree, the computing power benchmark value of the cloud computing node required by each traffic area subgraph is determined.

[0043] The computing power benchmark value is calculated by using a multiple linear regression model.

[0044] The calculation formula of the computing power benchmark value is: C=α× +β× +γ× ; Wherein, C is the computing power benchmark value, is the passenger flow data processing amount, and a is the passenger flow data weight, is the equipment data processing amount, and β is the equipment data weight, γ is the weight of public opinion data.

[0045] The error of computing power evaluation is small, and the accuracy is significantly improved compared to the traditional fixed computing power configuration.

[0046] In step S302, the corresponding elastic cloud computing node is configured according to the computing power reference value, a dynamic mapping relationship between the elastic cloud computing node and the traffic area subgraph is established, the service traffic area subgraph range, the computing power upper limit and the data receiving port of each elastic cloud computing node are recorded, and stored in the elastic cloud computing node mapping library of the cloud platform.

[0047] The elastic cloud computing node is configured by using the Kubernetes container orchestration technology, and the elastic cloud computing node mapping library stores node information by using a Redis cache database. The time consumption of elastic cloud computing node configuration is shortened from hours to minutes, and the response speed of elastic cloud computing node mapping library query is improved to milliseconds, supporting subsequent fast data distribution.

[0048] In step S303, the passenger flow change and public opinion heat of each traffic area subgraph are monitored in real time, specifically including: If the traffic area subgraph load is greater than the third percentage of the computing power reference value, the corresponding elastic cloud computing node is expanded.

[0049] If the traffic area subgraph load is less than the fourth percentage of the computing power reference value, the computing power of the elastic cloud computing node is reduced.

[0050] Step S4 includes the following sub-steps: In step S401, the global operation data of multi-period rail transit is read simultaneously by a cloud computing processing model, the isolated forest algorithm is used to mine the abnormal features in the historical data, the pre-check database of multi-elastic cloud computing node parallel access is built, and the pre-check database updates the threshold value with seasonal passenger flow and public opinion hotspots. The abnormal features include passenger flow fluctuation threshold, equipment parameter threshold and negative emotion threshold of public opinion.

[0051] The cloud computing processing model includes a distributed preprocessing layer and a cloud conversion layer. The preprocessing layer integrates a data cleaning module, and the cloud conversion layer embeds a data clustering algorithm based on machine learning.

[0052] In step S402, the global operation data is distributed to each elastic cloud computing node according to the corresponding relationship of the traffic area subgraph, the pre-check database is called from the preprocessing layer, and the passenger flow data, equipment working condition data and social media public opinion data are traversed respectively by multiple elastic cloud computing nodes in parallel to perform feature correction, and the abnormal data of rail transit matched with the abnormal features are screened out.

[0053] Step S403, the screened abnormal data is transmitted to the cloud conversion layer, classified according to the main data mode by the machine learning clustering algorithm for parallel processing, the data correlation conversion efficiency is improved, the intra-modal data correlation rules and cross-modal correlation rules are established, the abnormal data distribution is determined, and the main data mode includes passenger flow congestion, equipment failure and public opinion early warning.

[0054] Step S5 includes the following sub-steps: Step S501, the cloud platform monitors the key events in the rail transit network in real time, and starts the data aggregation process as soon as a key event is monitored. When the data aggregation process is started, the cloud platform will notify all the elastic cloud computing nodes corresponding to the traffic area subgraph, so that the elastic cloud computing nodes upload the abnormal data processed by themselves to the cloud platform. The elastic cloud computing nodes preferentially upload the abnormal data of the event correlation area. The cloud platform monitors the key events in real time, quickly focuses on the core area of the problem, and reduces the delay caused by irrelevant data transmission.

[0055] The key events include: The third percentile of the real-time passenger flow at the station exceeds the preset passenger flow threshold, the train fails to stop, and the negative sentiment index of the rail transit related public opinion on the social media is greater than the second preset warning line.

[0056] The event correlation area includes: When the train fails, the abnormal data of the stations to be passed through by the failed train is preferentially uploaded.

[0057] When the public opinion early warning, the abnormal data of the key stations mentioned in the public opinion is preferentially uploaded.

[0058] Step S502, after the cloud platform aggregates the abnormal data uploaded by each elastic cloud computing node, the correlation strength between the elastic cloud computing nodes is analyzed according to the machine learning algorithm to learn the common characteristics of the abnormal data, and the correlation relationship diagram of the abnormal data and the key events is constructed.

[0059] The combination algorithm of graph neural network combined with spectral clustering and association rule learning is used as the machine learning algorithm, and the specific logic includes: The elastic cloud computing node is regarded as a node in the graph neural network, the data interaction frequency between the nodes and the probability of abnormal data appearing together are regarded as the weight of the edge, the correlation strength between the elastic cloud computing nodes is learned through the message passing mechanism of the graph neural network, the neighbor node features are aggregated by the graph neural network model, the connection weight between the nodes is dynamically updated, and the abnormal propagation correlation of different traffic area subgraphs is quantified.

[0060] For the aggregated abnormal data, the spectral clustering algorithm is used to cluster the abnormal data with high similarity into a class.

[0061] The intra-modal commonality features are clustered based on passenger flow density features and propagation speed features, and a propagation path feature of diffusion along a transfer route is extracted.

[0062] The cross-modal commonality features are clustered by feature fusion, specifically including mapping public opinion heat and passenger flow growth speed to the same feature space, and obtaining a synergistic rule that public opinion heat rising leads to passenger flow congestion intensifying.

[0063] The association rules of abnormal data and key events are mined by a frequent pattern growth algorithm, and a correlation graph is constructed with the key events as the core, the abnormal data as the nodes, and the correlation strength as the edge weight, to intuitively present the linkage logic of events, abnormalities, and affected areas.

[0064] The commonality features include intra-modal commonality and cross-modal commonality.

[0065] The intra-modal commonality includes a propagation path feature of passenger flow congestion and a chain reaction type of device failure.

[0066] The cross-modal commonality includes a synergistic change rule of public opinion heat and passenger flow congestion degree.

[0067] In step S503, according to the correlation graph, the abnormal distribution is predicted in combination with a real-time passenger flow evolution trend, to obtain predicted abnormal data, and the real-time abnormal data and the predicted abnormal data are integrated to generate an abnormal distribution graph. The real-time passenger flow evolution trend includes passenger flow direction and passenger flow growth speed.

[0068] The logic of predicting the abnormal distribution includes: The abnormal distribution is predicted by a spatiotemporal graph neural network and a multi-factor constraint regression prediction model, and the specific prediction logic is as follows: The spatiotemporal graph neural network is used as the core to fuse a time series prediction module and a rule constraint layer. The spatiotemporal graph neural network is used to capture the linkage features of abnormalities in space and time, the time series prediction module is used to process the dynamic changes of passenger flow growth speed, and the rule constraint layer is used to integrate the prior knowledge of commonality features. The linkage features in space are station and line correlation features, and the linkage features in time are passenger flow evolution features.

[0069] The passenger flow direction is fused with the spatial features, specifically including: The passenger flow direction is converted into a spatial adjacency weight matrix. The weight of a station pointed by the flow direction is a first spatial connection weight, and the weight of a station in the opposite direction pointed by the flow direction is a second spatial connection weight. The first spatial connection weight and the second spatial connection weight are taken as the spatial input of the spatiotemporal graph neural network. The first spatial connection weight is greater than the second spatial connection weight, guiding the prediction model to preferentially learn the propagation rule of abnormalities along the flow direction. If the passenger flow of the fault station mainly flows to the east transfer station, the prediction model improves the abnormal probability prediction weight of the east station.

[0070] The passenger flow growth rate is fused with the time characteristics, specifically including: The passenger flow growth rate is taken as the time sequence input of the space-time graph neural network. When the passenger flow growth rate is greater than a third preset range, the time decay coefficient of the combined model is reduced, the time gradient calculation of the abnormal expansion is accelerated, and it is predicted that the abnormality covers adjacent first number of stations in a first unit of time. When the passenger flow growth rate is less than the third preset range, the time decay coefficient of the combined model is increased, the abnormal expansion speed is delayed, and it is predicted that the abnormality covers adjacent second number of stations in a second unit of time, and the first number is greater than the second number.

[0071] If there is a modal intra-common feature of the propagation path feature of the passenger flow congestion in the association relationship graph, additional weights are given to the inter-station connection of the propagation path in the spatial attention layer of the space-time graph neural network, and the prediction model is forced to preferentially predict abnormal expansion along the propagation path.

[0072] If the interlocking reaction type of the equipment failure has a modal intra-common feature of the interlocking reaction type of the equipment failure, the third number of signal associated stations upstream and downstream of the failure point are marked as high-risk nodes, and the initial value of the abnormal probability is increased.

[0073] If there is a cross-modal commonality in the association relationship graph that the passenger flow congestion range expands by a fifth percentage every time the public opinion heat increases by a fourth percentage, the real-time public opinion heat is converted into a correction coefficient to magnify the prediction range output by the space-time graph neural network.

[0074] The event associated area is marked as a seed node, a higher initial abnormal probability is given in the initial feature vector of the space-time graph neural network, and the neighborhood influence of the seed node is preferentially calculated in the spatial propagation, so as to ensure that the prediction result is inclined to the core associated area.

[0075] Step S6 includes the following substeps: Step S601, establish the coordinate mapping relationship between the abnormal distribution map and the dynamic panoramic topology map, and match the real-time abnormal area and the predicted abnormal area in the abnormal distribution map with the corresponding station and line position in the dynamic panoramic topology map respectively.

[0076] The matching mode includes: using differentiated visual identification according to the abnormal type, marking passenger flow congestion with red, marking equipment failure with yellow, and marking public opinion early warning with orange.

[0077] Step S602, generate global early warning information based on the mapping result of the coordinate mapping relationship between the abnormal distribution map and the dynamic panoramic topology map, and the early warning information includes the abnormal type, the abnormal occurrence position, the abnormal influence range and the economic influence evaluation.

[0078] The economic impact assessment includes the estimation of the increase in the cost of commuting per capita caused by train delays and the loss of revenue of surrounding businesses caused by station congestion.

[0079] The warning levels are divided according to the abnormal influence range, and are divided into a first-level warning, a second-level warning and a third-level warning in descending order.

[0080] The first-level warning is a network-wide associated anomaly, the second-level warning is a cross-line anomaly, and the third-level warning is a single-station anomaly.

[0081] In step S603, the global warning information is synchronized to the cloud platform visual display interface, supporting the operation personnel to view the abnormal details and evolution trend, and at the same time, the interface is pushed to the rail transit operation management system, the station broadcasting system and the passenger service APP, realizing the multi-end synchronization of the warning information, and according to the update of the dynamic panoramic topology map, the warning information is refreshed in real time, ensuring that the warning information is consistent with the actual running state.

[0082] Embodiment 2, refer to Figure 2 provides a rail transit data analysis system based on cloud computing, comprising a construction module, a segmentation module, a configuration module, a processing module, a monitoring module and a warning module.

[0083] The construction module is used for collecting real-time global operation data of a target rail transit network, constructing an initial panoramic topology map based on the global operation data, and dynamically adjusting the initial panoramic topology map according to real-time update data to obtain a dynamic panoramic topology map.

[0084] The segmentation module is used for segmenting the dynamic panoramic topology map into N traffic area subgraphs under the constraint of a preset principle tolerance.

[0085] The configuration module is used for configuring elastic cloud computing nodes mapped to the N traffic area subgraphs.

[0086] The processing module is used for distributing the global operation data to the corresponding elastic cloud computing nodes, combining a cloud computing processing model, and performing monitoring data initialization filtering and cloud association conversion through a data clustering algorithm to determine an abnormal data distribution.

[0087] The monitoring module is used for real-time monitoring of key events, aggregating abnormal data of each elastic cloud computing node, mining commonalities of the abnormal data, determining a predicted abnormal distribution in combination with a real-time passenger flow evolution trend, and obtaining an abnormal distribution map.

[0088] The warning module is used for real-time mapping of the abnormal distribution map in the dynamic panoramic topology map to generate global warning information.

[0089] The application improves the efficiency of rail transit operation, data processing capability, response time of abnormality and passenger experience in all aspects. By integrating passenger flow, train operation status, equipment working condition and social media public opinion data, combining the time and space synchronization characteristics of dynamic panoramic topological map, the traditional multi-source data island problem is effectively solved, the data integrity is improved, the cross-modal abnormality recognition rate is improved, the chain correlation from the rise of public opinion heat to the intensification of passenger flow congestion to the overload of equipment load can be accurately captured, and the one-sidedness of single data analysis is avoided. At the same time, based on the elastic cloud computing node scheduling mechanism of the traffic area subgraph load, the computing power is dynamically adapted, the utilization rate of computing power resources is improved, the waste of invalid computing power is reduced, and the response time of computing power expansion during peak period is shortened, which avoids the data processing delay of high load area and reduces the energy consumption during off-peak period, balances the efficiency and economy. The prediction algorithm using spatiotemporal graph neural network combined with common constraints upgrades the abnormality processing from passive response to active prediction, improves the accuracy of abnormality prediction, can predict the abnormality diffusion range and time in advance, shortens the emergency response time, and the correlation graph can intuitively present the linkage of events, abnormalities and influence areas, helping the operator to quickly locate the core problem. The multi-end collaborative early warning system can synchronize the early warning information to the operation management system, station broadcast and passenger service APP, shorten the multi-end synchronization delay, speed up the passengers to obtain abnormal information, improve the passenger satisfaction, and the operator can also quantify the loss by means of economic impact evaluation, make more accurate decisions, and provide core support for fine operation of large-scale rail transit network.

[0090] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, which should be covered by the scope of the claims of the present application.

Claims

1. A cloud computing-based rail transit data analysis method, characterized in that, The method comprises the following steps: Step S1, collecting global operation data of a target rail transit network, and constructing a dynamic panoramic topology graph based on the global operation data; Step S2, dividing the dynamic panoramic topology graph into N traffic area subgraphs; Step S3, configuring elastic cloud computing nodes mapped to the N traffic area subgraphs, recording the service traffic area subgraph range, upper limit of computing power and data receiving port of each elastic cloud computing node, and monitoring passenger flow changes and public opinion heat of each traffic area subgraph in real time; Step S4, distributing the global operation data to the corresponding elastic cloud computing nodes, combining a cloud computing processing model, and performing global operation data initialization filtering and cloud association conversion through a data clustering algorithm to determine an abnormal data distribution; Step S5, after the cloud platform detects a key event, the abnormal data of each elastic cloud computing node is summarized, the commonness of the abnormal data is mined, the predicted abnormal data distribution is determined in combination with the real-time passenger flow evolution trend, and an abnormal distribution graph is obtained; Step S6, real-time mapping the abnormal distribution graph in the dynamic panoramic topology graph to generate global early warning information.

2. The rail transit data analysis method based on cloud computing according to claim 1, wherein, The step S1 comprises the following sub-steps: Step S101, collecting global operation data of a target rail transit network in real time, wherein the global operation data comprises passenger flow data, train operation state data, equipment working condition data and social media public opinion data; Step S102, constructing an initial panoramic topology graph of the rail transit network based on the global operation data; Step S103, dynamically adjusting the initial panoramic topology graph according to the real-time flow of the global operation data to obtain a dynamic panoramic topology graph.

3. The cloud computing-based rail transit data analysis method of claim 2, wherein, The step S2 comprises the following sub-steps: Step S201, under the constraint of a preset principle tolerance, combining real-time global operation data, and preliminarily dividing the dynamic panoramic topology graph to obtain N initial traffic area subgraphs; The preset principle tolerance comprises a passenger flow peak balance principle and a weak dependence principle; The passenger flow peak balance principle comprises: The initial traffic area subgraph passenger flow load fluctuation threshold is within a first preset range; The weight of the passenger flow peak balance principle is adjusted to a first percentage during a passenger flow peak or public opinion early warning period; The weak dependence principle comprises that the data interaction frequency between the initial traffic area subgraphs is within a second preset range, and the initial traffic area subgraph boundary division preferentially maintains the integrity of the same line continuous station; Step S202, adaptively optimizing the coverage range of the initial traffic area subgraph according to the real-time passenger flow changes and the update of the dynamic panoramic topology graph to obtain a traffic area subgraph.

4. The rail transit data analysis method based on cloud computing according to claim 3, wherein, The logic of adaptively optimizing the coverage range of the initial traffic area subgraph comprises: Monitoring the passenger flow density of each initial traffic area subgraph in real time, if the passenger flow of one of the initial traffic area subgraphs is greater than a preset load threshold, the coverage range of the one of the initial traffic area subgraphs is reduced and the surrounding area is divided to the adjacent initial traffic area subgraph; If the load difference between the initial traffic area subgraphs is greater than a second percentage due to passenger flow tide, the initial traffic area subgraph boundary is adjusted. If the public opinion heat degree associated with one of the initial traffic area subgraphs is greater than the first preset warning line, the coverage range of the one of the initial traffic area subgraphs is preferentially ensured to completely match the site involved in the public opinion.

5. The cloud computing-based rail transit data analysis method of claim 4, wherein, The step S3 includes the following substeps: Step S301, evaluating the service carrying demand of each traffic area subgraph, specifically including: Determine the computing power benchmark value of the cloud computing node required by each traffic area subgraph in combination with the real-time passenger flow load, data processing amount and public opinion correlation of the traffic area subgraph; Step S302, configuring the corresponding elastic cloud computing node according to the computing power benchmark value, establishing a dynamic mapping relationship between the elastic cloud computing node and the traffic area subgraph, recording the service traffic area subgraph range, computing power upper limit and data receiving port of each elastic cloud computing node, and storing to the elastic cloud computing node mapping library of the cloud platform; Step S303, real-time monitoring of passenger flow changes and public opinion heat degree of each traffic area subgraph, specifically including: If the traffic area subgraph load is greater than the third percentage of the computing power benchmark value, the corresponding elastic cloud computing node is expanded; If the traffic area subgraph load is less than the fourth percentage of the computing power benchmark value, the computing power of the elastic cloud computing node is reduced.

6. The cloud computing-based rail transit data analysis method of claim 5, wherein, The step S4 includes the following substeps: Step S401, simultaneously reading the global operation data of multi-period rail transit through the cloud computing processing model, mining abnormal features, building a pre-check database for parallel access of multiple elastic cloud computing nodes, the pre-check database updating the threshold value according to seasonal passenger flow and public opinion hotspots, the abnormal features including passenger flow fluctuation threshold, equipment parameter threshold and negative public opinion emotion threshold; The cloud computing processing model includes a distributed preprocessing layer and a cloud conversion layer, the preprocessing layer integrates a data cleaning module, and the cloud conversion layer embeds a data clustering algorithm based on machine learning; Step S402, distributing the global operation data to each elastic cloud computing node according to the corresponding relationship of the traffic area subgraph, calling the pre-check database from the preprocessing layer, traversing the passenger flow data, equipment working condition data and social media public opinion data through multiple elastic cloud computing nodes in parallel, and respectively performing feature correction, and screening out abnormal data of rail transit matching the abnormal features; Step S403, passing the screened abnormal data into the cloud conversion layer, classifying according to the main data mode through the machine learning clustering algorithm, establishing intra-modal data correlation rules and cross-modal correlation rules, and determining the distribution of abnormal data, the main data mode including passenger flow congestion, equipment failure and public opinion warning.

7. The cloud computing-based rail transit data analysis method of claim 6, wherein, The step S5 includes the following substeps: Step S501, the cloud platform real-time monitors the key events in the rail transit network, and once a key event is monitored, the data aggregation process is started, when the data aggregation process is started, the cloud platform will notify all elastic cloud computing nodes corresponding to the traffic area subgraph, so that the elastic cloud computing nodes upload the abnormal data processed by themselves to the cloud platform, and the elastic cloud computing nodes preferentially upload the abnormal data of the event associated area; The key events include: Real-time passenger flow of the site exceeding the third percentage of the preset passenger flow threshold, train failure and stagnation, and negative public opinion emotion index of rail transit related on social media being greater than the second preset warning line; The event correlation region comprises: When a train fails, the abnormal data of the stations to be passed through by the failed train is preferentially uploaded; When public opinion is warned, the abnormal data of the key stations mentioned in the public opinion is preferentially uploaded; In step S502, after the cloud platform aggregates the abnormal data uploaded by each elastic cloud computing node, the correlation strength between the elastic cloud computing nodes is learned according to a machine learning algorithm, the common characteristics of the abnormal data are analyzed, and a correlation graph of the abnormal data and key events is constructed; The common characteristics comprise intra-modal common characteristics and cross-modal common characteristics; The intra-modal common characteristics comprise a propagation path characteristic of passenger flow congestion and a chain reaction type of equipment failure; The cross-modal common characteristics comprise a cooperative variation law of public opinion heat and passenger flow congestion degree; In step S503, according to the correlation graph, in combination with a real-time passenger flow evolution trend, an abnormal distribution is predicted, predicted abnormal data is obtained, real-time abnormal data and the predicted abnormal data are integrated, and an abnormal distribution graph is generated, wherein the real-time passenger flow evolution trend comprises a passenger flow direction and a passenger flow growth speed.

8. The cloud computing-based rail transit data analysis method of claim 7, wherein, The logic of predicting the abnormal distribution comprises: The passenger flow direction is fused with spatial characteristics, specifically including: The passenger flow direction is converted into a spatial adjacency weight matrix, a weight of a station pointed by the passenger flow direction is preset as a first spatial connection weight, a weight of a station in the opposite direction pointed by the passenger flow direction is preset as a second spatial connection weight, the first spatial connection weight and the second spatial connection weight are taken as spatial inputs of a prediction model, and the first spatial connection weight is greater than the second spatial connection weight; The passenger flow growth speed is fused with time characteristics, specifically including: The passenger flow growth speed is taken as a time sequence input of the prediction model, when the passenger flow growth speed is greater than a third preset range, it is predicted that the abnormality covers adjacent first number of stations in a first unit time, and when the passenger flow growth speed is less than the third preset range, it is predicted that the abnormality covers adjacent second number of stations in a second unit time, wherein the first number is greater than the second number; The spatial characteristics are station and line correlation characteristics, and the time characteristics are passenger flow evolution characteristics; The common characteristics in the correlation graph are introduced as constraint conditions of the prediction model, if the common characteristics are intra-modal common characteristics, the diffusion range is corrected according to the propagation path characteristic of passenger flow congestion and the chain reaction type of equipment failure, and if the common characteristics are cross-modal common characteristics, the prediction result is adjusted in combination with real-time public opinion data; When the prediction model is used for prediction, stations in the event correlation region are preferentially included in a preferential coverage range; The stations and lines covered by the abnormality are predicted in stages according to preset time periods, and prediction abnormal data of each time period is generated.

9. The cloud computing-based rail transit data analysis method of claim 8, wherein, The step S6 comprises the following sub-steps: In step S601, a coordinate mapping relationship between the abnormal distribution graph and a dynamic panoramic topology graph is established, real-time abnormal regions and predicted abnormal regions in the abnormal distribution graph are matched with corresponding station and line positions in the dynamic panoramic topology graph, respectively; The matching mode comprises: adopting differential visual identification according to the abnormal types, marking passenger flow congestion with red, marking equipment failure with yellow, and marking public opinion warning with orange. Step S602, generating global early warning information based on the mapping result of the coordinate mapping relationship between the abnormal distribution map and the dynamic panoramic topology map, the early warning information including abnormal type, abnormal occurrence position, abnormal influence range and economic impact assessment; The economic impact assessment includes per capita commuting cost increase caused by train delay and surrounding business revenue loss estimation caused by station congestion; According to the abnormal influence range, the early warning levels are divided in descending order into first-level early warning, second-level early warning and third-level early warning; The first-level early warning is a global associated abnormality, the second-level early warning is a cross-line abnormality, and the third-level early warning is a single station abnormality; Step S603, synchronizing the global early warning information to the cloud platform visual display interface, and pushing it to the rail transit operation management system, the station broadcasting system and the passenger service APP through the interface, and refreshing the early warning information in real time according to the update of the dynamic panoramic topology map.

10. A cloud computing-based rail transit data analysis system applied to a cloud computing-based rail transit data analysis method according to any one of claims 1-9, characterized in that, It includes a construction module, a segmentation module, a configuration module, a processing module, a monitoring module and an early warning module. The construction module is used for real-time acquisition of global operation data of a target rail transit network, construction of an initial panoramic topology map based on the global operation data, and dynamic adjustment of the initial panoramic topology map according to real-time update data to obtain a dynamic panoramic topology map. The segmentation module is used for segmenting the dynamic panoramic topology map into N traffic area subgraphs under the constraint of a preset principle tolerance. The configuration module is used for configuring elastic cloud computing nodes mapped to the N traffic area subgraphs. The processing module is used for distributing the global operation data to corresponding elastic cloud computing nodes, combining a cloud computing processing model, and performing monitoring data initialization filtering and cloud association conversion through a data clustering algorithm to determine abnormal data distribution. The monitoring module is used for real-time monitoring of key events, summarizing abnormal data of each elastic cloud computing node, mining commonalities of abnormal data, determining a predicted abnormal distribution in combination with a real-time passenger flow evolution trend, and obtaining an abnormal distribution map. The early warning module is used for real-time mapping of the abnormal distribution map in the dynamic panoramic topology map to generate global early warning information.

Images