Urban public transportation industry safety supervision system and method based on risk map

Abstract

The invention relates to the technical field of urban traffic safety management, in particular to an urban public transportation industry safety supervision system and method based on a risk map, and the method comprises the following steps: S1, collecting multi-source data from a vehicle end, a city end and a personnel end; s2, writing the multi-source data into a time sequence database and a space database respectively, and establishing a space-time retrieval index; s3, calculating risk values D = D (L, E, C) of the candidate risk points based on an LEC method, and classifying the risk points according to a preset threshold to generate a risk map layer; and S4, performing space-time comparison on the position of the real-time driving vehicle and the risk map, and when the vehicle enters the radius of the risk point or the driving behavior is triggered to be abnormal, triggering a real-time alarm and issuing a disposal suggestion. According to the invention, the problems of data islands, prediction lag, low efficiency of collaboration and the like in a traditional scheme are effectively overcome, and the real-time performance, accuracy and controllability of safety supervision in the public transportation industry are improved.

Description

Technical Field

[0001] This invention relates to a risk map-based safety supervision system and method for the urban public transport industry, and particularly to a risk map-based safety supervision system and method for the urban public transport industry, belonging to the field of urban traffic safety management technology. Background Technology

[0002] Urban public transport safety supervision has widely adopted on-board and city-level sensing devices (such as OBD, GPS, on-board cameras, passenger flow / door control sensors, roadside cameras, and traffic police accident databases) to collect operation and accident data, and uses time-series databases and spatial databases (such as time-series storage + PostGIS-type spatial databases) for storage and retrieval. Static or semi-dynamic "risk maps" are generated based on empirical rules or L·E·C (probability L, exposure E, consequence C) and used for post-event analysis and alerts. The dispatch system can issue voice / interface prompts to drivers. Some systems have begun to try to introduce machine learning models for anomaly detection and hotspot assessment to assist decision-making. Existing solutions are mostly passive / semi-manual driven, facing challenges such as difficulties in fusing multi-source heterogeneous data, low efficiency in spatiotemporal alignment and indexing, insufficient predictive capabilities due to scarce event labels, difficulty in modeling road network topology and time-varying relationships, and conflicts between real-time performance and scalability. In addition, cross-departmental handling processes rely heavily on manual communication, lacking automated work orders and closed-loop acceptance. Privacy and data silos make it difficult to share models among multiple operating entities, and the lack of an immutable audit chain to support accountability leads to delayed risk warnings, low handling efficiency, insufficient predictive reliability, and poor regulatory transparency. Therefore, there is an urgent need to improve the urban public transport industry safety supervision system and methods based on risk maps to solve the above-mentioned problems. Summary of the Invention

[0003] The purpose of this invention is to provide a safety supervision system and method for the urban public transportation industry based on risk maps, in order to solve the problems faced by existing solutions, which are mostly passive / semi-manually driven, and suffer from difficulties in fusion of multi-source heterogeneous data, low efficiency in spatiotemporal alignment and indexing, insufficient predictive ability due to the scarcity of event labels, difficulty in modeling road network topology and time-varying relationships, and conflicts between real-time performance and scalability. In addition, cross-departmental handling processes rely heavily on manual communication, lacking automated work orders and closed-loop acceptance; privacy and data silos make it difficult to share models among multiple operating entities, and there is a lack of an immutable audit chain to support accountability. The above problems lead to delayed risk warnings, low handling efficiency, insufficient predictive credibility, and poor regulatory transparency.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A risk map-based safety supervision system and method for the urban public transport industry includes the following steps: S1. Collect multi-source data from vehicle, city, and personnel ends; S2. Write the multi-source data into the time-series database and the spatial database respectively, and establish a spatiotemporal retrieval index; S3. Calculate the risk value D=D(L,E,C) of candidate risk points based on the LEC method, and generate a risk map layer by classifying the risk points according to a preset threshold. S4. Compare the real-time vehicle location with the risk map in time and space. When the vehicle enters the risk point radius or triggers abnormal driving behavior, trigger a real-time alarm and issue a handling suggestion. S5. Based on historical spatiotemporal accident data, real-time traffic congestion, weather, construction / event calendar, and recent driving behavior anomaly density, a spatiotemporal learning model is used to predict the probability of increased risk in the short term of road sections. S6. Simulate multiple handling strategies for predicted or identified high-risk scenarios in a digital twin environment and evaluate the impact of each handling strategy on the risk value. S7. When the risk point involves municipal or facility issues that require external handling, automatically generate a cross-departmental work order containing evidence packages and assign response SLAs according to risk level, and track the handling until acceptance. S8. Without sharing the original sensitive trajectory data, conduct collaborative training and updates of models among multiple operating entities through federated learning or parameter aggregation, and retain tamper-proof audit logs for key operations to achieve post-event traceability.

[0005] Preferably, in S1, the data collected by the vehicle includes, but is not limited to: OBD data, GPS positioning data, in-vehicle camera video / images, ADAS events, CAN bus data, door opening and closing status, passenger flow sensor data, and in-vehicle voice reporting data.

[0006] Preferably, in S2, the time-series database is used to store high-frequency sensing time-series data, and the spatial database is used to store risk points, trajectories, and geographical features, and a joint spatiotemporal retrieval capability is established through spatial index and time-series index.

[0007] Preferably, in S3, the risk classification adopts the following thresholds: D≥320 is major risk, 160≤D<320 is relatively high risk, 70≤D<160 is moderate risk, and D<70 is low risk; and the above thresholds can be dynamically adjusted according to region and time period.

[0008] Preferably, in S4, real-time alarms are sent to the vehicle terminal in the form of voice prompts and interface pop-ups, and are simultaneously pushed to the dispatch center and the monitoring platform; for major risks, the system is also configured to issue mandatory handling suggestions and record the driver's response behavior for post-event assessment.

[0009] Preferably, in S5, the spatiotemporal learning model is a model structure that combines graph neural networks and temporal networks, used to model the road network topology and the propagation of spatiotemporal risks. The model output is the probability and confidence level of risk increase in the near future.

[0010] Preferably, in S6, the digital twin environment evaluates the risk mitigation effect of the response strategy based on the current road network topology, vehicle / pedestrian flow simulation and weather conditions, and outputs priority recommendations through Monte Carlo simulation or strategy optimization-based simulation methods.

[0011] Preferably, in S7, the automatically generated cross-departmental work orders include work order number, responsible unit, risk point ID, evidence package link, suggested handling measures and expected response time limit, and support multi-department joint approval and closed-loop acceptance records.

[0012] Preferably, in S8, federated learning adopts a framework of local model updates and central aggregation, and performs differential privacy processing or encryption processing on the updates before aggregation to protect the data privacy of the participants; and the system records tamper-proof audit information for key operations such as model updates, risk point changes and work order handling for traceability.

[0013] Preferably, it includes: a data acquisition module, used to collect and standardize multi-source data from vehicle, city, and personnel ends; The spatiotemporal storage module is used to store time-series data and spatial elements separately and provide a spatiotemporal retrieval interface; The real-time stream processing module is used to detect the spatiotemporal relationship between vehicles and risk points and trigger real-time alarms; The risk assessment module is used to calculate risk values ​​and generate risk map layers based on LEC rules; The prediction module is used to predict short-term risks based on a spatiotemporal learning model. The digital twin simulation module is used to simulate response strategies and assess their impact on risks; The work order management module is used to automatically generate, issue, and track cross-departmental work orders and record acceptance. The federated learning module is used to collaboratively train predictive models among multiple operating entities without sharing raw data; The auditing and permissions module is used to record immutable logs of critical operations and control data and function access permissions.

[0014] The present invention has at least the following beneficial effects: This invention constructs a method and system for safety supervision of the urban public transport industry based on risk maps. This achieves intelligent supervision throughout the entire process, from multi-source sensing data collection, unified spatiotemporal storage, risk identification and prediction, digital twin simulation, to cross-departmental closed-loop handling. By introducing a spatiotemporal learning model combining graph neural networks and temporal networks, it can accurately predict the probability of short-term risk increases in road sections while considering road topology and temporal evolution characteristics. Digital twin simulation is used to evaluate handling strategies, helping to select the optimal response plan in advance and reduce secondary risks. Federated learning and differential privacy mechanisms enable collaborative model training among multiple operating entities without sharing sensitive data. Combined with tamper-proof audit logs and an automated work order system, it significantly improves cross-departmental collaboration efficiency and regulatory traceability, effectively overcoming problems such as data silos, prediction lag, and inefficient collaboration in traditional solutions, thus enhancing the real-time performance, accuracy, and controllability of public transport industry safety supervision. Attached Figure Description

[0015] Figure 1 This is a flowchart of the urban public transport industry safety supervision method based on risk maps according to the present invention; Figure 2 This is a system block diagram of the urban public transport industry safety supervision system based on risk maps according to the present invention. Detailed Implementation

[0016] The following will describe in detail the implementation of this application with reference to the accompanying drawings and embodiments, so that the implementation process of how this application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0017] like Figure 1-2 As shown in this embodiment, the urban public transport industry safety supervision system and method based on risk maps includes the following steps: S1. Collect multi-source data from vehicle, city, and personnel ends; S2. Write the multi-source data into a time-series database and a spatial database respectively, and establish a spatiotemporal retrieval index; S3. Calculate the risk value D=D(L,E,C) of candidate risk points based on the LEC method, and generate a risk map layer by classifying the risk points according to a preset threshold; S4. Perform spatiotemporal comparison between the real-time vehicle position and the risk map, and trigger a real-time alarm and issue handling suggestions when a vehicle enters the radius of a risk point or triggers abnormal driving behavior; S5. Based on historical spatiotemporal accident data, real-time traffic congestion, weather, construction / event calendar, and recent abnormal driving behavior density... The system employs a spatiotemporal learning model to predict the probability of increased risk in short-term road sections; S6, in a digital twin environment, it simulates multiple handling strategies for predicted or identified high-risk situations and evaluates the impact of each strategy on the risk value; S7, when the risk point involves municipal or facility issues requiring external handling, it automatically generates cross-departmental work orders containing evidence packages and assigns response SLAs according to risk levels, tracking the handling until acceptance; and S8, without sharing original sensitive trajectory data, it conducts collaborative model training and updates among multiple operating entities through federated learning or parameter aggregation, and retains tamper-proof audit logs for key operations to achieve post-event traceability. By acquiring information from multiple sources (vehicles / cities / personnel) and uniformly building and indexing databases in the spatiotemporal dimensions, this method can form a complete data foundation, supporting closed-loop management of risk identification, classification, early warning, and disposal. LEC-based risk classification makes risks quantifiable and classifiable for disposal. Combined with real-time comparison and short-term prediction, it can transform passive post-event handling into proactive prevention and in-event intervention, thereby reducing the accident rate and improving emergency response efficiency. Federated learning and tamper-proof auditing ensure the privacy and traceability of cross-entity collaboration, improving inter-industry collaboration capabilities and regulatory transparency.

[0018] As a further implementation of the present invention, the data collected by the vehicle includes, but is not limited to: OBD data, GPS positioning data, in-vehicle camera video / images, ADAS events, CAN bus data, door opening and closing status, passenger flow sensor data, and in-vehicle voice reporting data. The rich vehicle-side perception (structured time-series data and unstructured multimedia) improves the signal-to-noise ratio and event evidence capability of risk detection, and can more accurately identify driving abnormalities, passenger flow abnormalities, and equipment failures, supporting post-event review and responsibility judgment, thereby enhancing the ability to locate the source of failure and formulate targeted handling measures. As a further implementation of the present invention, the time-series database is used to store high-frequency sensing time-series data, and the spatial database is used to store risk points, trajectories, and geographical features. A joint spatiotemporal retrieval capability is established through spatial index and time-series index. High-frequency time-series data and geographical features are stored separately and spatiotemporal joint retrieval is achieved through joint index. This not only ensures high-throughput writing and time-series analysis capabilities, but also supports complex spatial queries (such as radius retrieval and trajectory intersection). This significantly improves the efficiency and accuracy of spatiotemporal event location, historical backtracking, and real-time alarms, and facilitates the construction of an scalable operation and analysis platform. As a further implementation of the present invention, the risk classification adopts the following thresholds: D≥320 is a major risk (red), 160≤D<320 is a relatively large risk (orange), 70≤D<160 is a general risk (yellow), and D<70 is a low risk (blue). Furthermore, it supports dynamic adjustment of the above thresholds by region and time period. The clear threshold system facilitates unified judgment standards and handling procedures, and is conducive to rapid resource allocation and SLA determination. Simultaneously, it allows dynamic adjustment of thresholds by region / time period, enabling risk assessment to adapt to spatiotemporal characteristics (such as peak / night / construction period), improving the model's flexibility and practicality, thereby achieving more refined safety supervision. As a further implementation of the present invention, the real-time alarm is sent to the vehicle terminal in the form of voice prompts and interface pop-ups, and simultaneously pushed to the dispatch center and the monitoring platform. For major risks, the system is also configured to issue mandatory handling suggestions and record the driver's response behavior for post-event assessment. The multi-channel real-time alarm (voice + interface + platform) ensures that the driver and dispatcher obtain risk information and take action in the shortest possible time. The mandatory suggestion and response recording mechanism for major risks can not only reduce the probability of risk occurrence in time, but also provide objective basis for post-event evaluation and driver training, and improve the standardization of driving behavior and the ability to trace responsibility. As a further implementation of the present invention, the spatiotemporal learning model is a model structure combining graph neural networks and time-series networks. It is used to model the road network topology and the propagation of spatiotemporal risks. The model output is the probability and confidence level of risk escalation in the short term (e.g., 1-24 hours). The combination of GNN and time-series model can simultaneously capture the influence of road topology and the time evolution law, making the prediction of risk spread and related areas more accurate. At the same time, the output confidence level helps the decision-making level to distinguish between high-confidence predictions and low-confidence predictions, thereby reasonably arranging simulation verification or manual review and improving the reliability of prediction-driven response. As a further implementation of the present invention, the digital twin environment, based on the current road network topology, vehicle / pedestrian flow simulation and weather conditions, evaluates the mitigation effect of the disposal strategy on the risk through Monte Carlo simulation or strategy optimization-based simulation methods and outputs priority suggestions. By verifying multiple disposal strategies (such as speed limits, detours, and temporary controls) in the digital twin in advance, its quantitative impact on the risk can be assessed before actual implementation, and the most cost-effective solution can be selected, avoiding secondary risks or waste of resources caused by blind intervention, thereby achieving scientific decision-making and minimizing social costs. As a further implementation of the present invention, the automatically generated cross-departmental work order includes a work order number, responsible unit, risk point ID, evidence package link (including videos / images with timestamps), suggested handling measures, and expected response time limit. It also supports joint approval by multiple departments and closed-loop acceptance records. The standardized work order and evidence package can realize the automatic triggering and traceable handling process of cross-departmental collaboration, shorten the problem response cycle, and ensure the quality of handling. The closed-loop acceptance record provides an objective basis for detecting the rectification effect, promotes the institutionalized flow of problems from discovery to resolution, and improves governance efficiency and public satisfaction. As a further implementation of the present invention, the federated learning adopts a framework of local model updates and central aggregation, and performs differential privacy processing or encryption processing on the updates before aggregation to protect the data privacy of the participants; the system records tamper-proof audit information for key operations such as model updates, risk point changes and work order handling for traceability. The federated learning mechanism achieves model capability sharing without exchanging original data, which breaks down data silos and protects the privacy of the operating entity; combined with differential privacy / encrypted aggregation and tamper-proof auditing, cross-entity collaboration is both secure, controllable and traceable, improving the feasibility and regulatory compliance of industry collaborative modeling. As a further embodiment of the present invention, the technical solution includes: The data acquisition module is used to collect and standardize multi-source data from vehicle, city, and personnel sources. The spatiotemporal storage module is used to store time-series data and spatial elements separately and provide a spatiotemporal retrieval interface; The real-time stream processing module is used to detect the spatiotemporal relationship between vehicles and risk points and trigger real-time alarms; The risk assessment module is used to calculate risk values ​​and generate risk map layers based on LEC rules; The prediction module is used to predict short-term risks based on a spatiotemporal learning model. The digital twin simulation module is used to simulate response strategies and assess their impact on risks; The work order management module is used to automatically generate, issue, and track cross-departmental work orders and record acceptance. The federated learning module is used to collaboratively train predictive models among multiple operating entities without sharing raw data; The auditing and permissions module is used to record immutable logs of critical operations and control data and function access permissions; By integrating capabilities such as data collection, storage, real-time processing, risk assessment, predictive simulation, work order closure, federated collaboration, and auditing permissions, a complete platform system from perception to decision-making to handling has been constructed, which is convenient for operation, maintenance, promotion, and large-scale replication. The system can support daily supervision and provide rapid response and evidence chain support in emergencies, thereby improving the overall safety management level of the urban public transportation industry.

[0019] If certain terms are used in the specification and claims to refer to specific components, those skilled in the art will understand that hardware manufacturers may use different names to refer to the same component. This specification and claims do not distinguish components based on differences in name, but rather on differences in function. The term "comprising" as used throughout the specification and claims is an open-ended term and should be interpreted as "comprising but not limited to." "Approximately" means that within an acceptable margin of error, those skilled in the art can solve the technical problem and substantially achieve the technical effect within a certain margin of error.

[0020] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or system comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the product or system that includes that element.

[0021] The foregoing description illustrates and describes several preferred embodiments of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A risk map-based urban public transportation industry safety supervision method, characterized in that, The method comprises the following steps: S1, collecting multi-source data from the vehicle side, the city side and the personnel side; S2, writing the multi-source data into a time series database and a spatial database respectively and establishing a time-space retrieval index; S3, calculating the risk value D=D(L,E,C) of the candidate risk point based on the LEC method, and classifying the risk points according to a preset threshold to generate a risk map layer; S4, comparing the real-time driving vehicle position with the risk map in time and space, triggering real-time warning and issuing disposal suggestions when the vehicle enters the risk point radius or triggers driving behavior anomaly; S5, predicting the future short-term road risk rise probability by using a time-space learning model based on historical time-space accident data, real-time traffic congestion, weather, construction / activity calendar and near-time driving behavior anomaly density; S6, simulating multiple disposal strategies in the digital twin environment for the predicted or identified high-risk situation and evaluating the influence of each disposal strategy on the risk value; S7, when the risk point involves municipal or facility problems that need external disposal, a cross-department work order containing an evidence package is automatically generated, the response SLA is allocated according to the risk level, and the disposal is tracked until acceptance; S8, without sharing the original sensitive trajectory data, the model collaborative training and updating among multiple operation subjects are carried out by federated learning or parameter aggregation, and the key operations are retained with non-tamperable audit logs for post-tracing.

2. The risk map-based urban public transport industry safety supervision method according to claim 1, characterized in that: In S1, the data collected by the vehicle side includes but is not limited to OBD data, GPS positioning data, vehicle-mounted camera video / picture, ADAS event, CAN bus data, door opening / closing state, passenger flow sensor data and vehicle-mounted voice reporting data.

3. The risk map-based urban public transport industry safety supervision method according to claim 1, characterized in that: In S2, the time series database is used to store high-frequency sensing time series data, the spatial database is used to store risk points, trajectories and geographic elements, and the joint time-space retrieval capability is established through spatial index and time series index.

4. The risk map-based urban public transport industry safety supervision method according to claim 1, characterized in that: In S3, the risk classification adopts the following threshold: D≥320 is a major risk, 160≤D<320 is a larger risk, 70≤D<160 is a general risk, and D<70 is a low risk; and the above threshold can be dynamically adjusted according to the region and time period.

5. The risk map-based urban public transport industry safety supervision method according to claim 1, characterized in that: In S4, the real-time warning is issued to the vehicle terminal in the form of voice prompt and interface pop-up window, and is pushed to the dispatch center and supervision platform at the same time; for major risks, the system is also configured to issue mandatory disposal suggestions and record the driver's response behavior for post-examination. 6.The risk map based urban public transport industry safety supervision method according to claim 1, characterized in that: In S5, the time-space learning model is a model structure combining graph neural network and time series network, which is used to model the road network topology and time-space risk propagation, and the model output is the risk rise probability and confidence in the future short term.

7. The risk map-based urban public transport industry safety supervision method according to claim 1, characterized in that: In S6, the digital twin environment evaluates the risk mitigation effect of the disposal strategy based on the current road network topology, vehicle / passenger flow simulation and weather conditions by using Monte Carlo simulation or simulation method based on strategy optimization, and outputs priority suggestions. 8.The risk map based urban public transport industry safety supervision method according to claim 1, characterized in that: In S7, the automatically generated cross-department work order contains work order number, responsible unit, risk point ID, evidence package link, suggested disposal measures and expected response time limit, and supports joint approval and closed-loop acceptance record of multiple departments. 9.The risk map based urban public transport industry safety supervision method according to claim 1, characterized in that: In S8, federated learning adopts the framework of local model updating and central aggregation, and differentially private processing or encryption processing is performed on the updates before aggregation to protect the privacy of the participants' data; and the system records the audit information of the key operations such as model updating, risk point changing and work order handling, which is tamper-proof for tracing.

10. The risk map based urban public transport industry safety supervision system according to any one of claims 1-9, characterized in that: Comprise: a data acquisition module for acquiring and standardizing multi-source data of vehicle end, city end and personnel end; a space-time storage module for saving time series data and spatial elements respectively and providing a space-time retrieval interface; a real-time stream processing module for detecting the space-time relationship between vehicles and risk points and triggering real-time alarms; a risk assessment module for calculating risk values based on LEC rules and generating risk map layers; a prediction module for predicting short-term risks based on space-time learning models; a digital twin simulation module for simulating disposal strategies and evaluating their impact on risks; a work order management module for automatically generating, issuing and tracking cross-department disposal work orders and recording acceptance; a federated learning module for training prediction models in collaboration among multiple operating subjects without sharing original data; an audit and permission module for recording tamper-proof logs of key operations and controlling data and function access permissions.

Images