Risk prediction and strategy generation method and system

Abstract

The invention provides a risk prediction and strategy generation method and system, and the method comprises the following steps: constructing a scene evolution model of rail transit operation, and obtaining an FAO scene evolution model; based on the FAO scene evolution model, quantifying a scene evolution safety boundary to obtain a scene boundary and a corresponding scene parameter set; based on the scene boundary and the corresponding scene parameter set, identifying FAO scene evolution key features; according to the technical scheme, full-scene dynamic evolution and prediction of the rail transit FAO are realized, accurate space-time quantitative risk assessment is provided, and explainable and intervening active decision support is generated.

Description

Technical Field

[0001] This application relates to the field of autonomous operation technology for rail transit, and in particular to a risk prediction and strategy generation method and system. Background Technology

[0002] With the development of information technology, agent technology has become one of the hot topics in current artificial intelligence research. A single agent often cannot describe and solve complex, large-scale problems. Therefore, an application system often contains multiple agents. These individual agents possess self-organization, perception, reasoning, action, and communication capabilities, and can collaborate in a loosely coupled manner to achieve a common overall goal, resulting in high efficiency and strong robustness. Such a system is called a Multi-Agent System (MAS). Specifically, it has three advantages: First, through communication between agents, common methods and plans can be used to solve problems in parallel, improving the efficiency of solving complex situations. Second, through cooperation between agents, individual agents with different sub-goals can rationally allocate resources to coordinate their respective behaviors and maximize the achievement of their goals. Third, MAS constructs multi-layered and diversified agents based on object-oriented methods, reducing system complexity while supporting distributed applications. It can be organized in a modular style, offering advantages such as easy expansion, simple and flexible design, and overcoming the difficulties of managing large-scale knowledge bases in complex systems.

[0003] As rail transit evolves towards fully automated operation (FAO) and autonomous train operation systems, the coupling relationship between the operating environment and the system becomes increasingly complex, placing higher demands on the advanced perception and precise control of safety risks. Existing technical solutions mainly include:

[0004] Static risk assessment based on historical data and rule bases includes methods such as Fault Tree Analysis (FTA) and Failure Mode and Effects Analysis (FMEA). These methods rely on historical failure statistics and expert experience to construct static risk logic models and identify weak points and high-risk nodes in the system. Their advantages are transparent architecture and high computational efficiency; however, they cannot reflect the dynamic changes in real-time operating status and are ill-suited to handling sudden disturbances and complex coupling risks.

[0005] Dynamic risk assessment based on real-time monitoring data: This approach collects data such as train speed, location, and signal status in real time using sensors, and combines this data with threshold alarms or statistical models (such as regression analysis and simple neural networks) to determine risk. Its advantages include a certain degree of timeliness; however, its significant disadvantages include: poor interpretability, as the methods are mostly "black box" models that fail to clearly reveal the internal logical chain of risk generation and evolution; insufficient systematic modeling, as most analyses focus on specific scenarios or fault conditions, making it difficult to accurately depict the multi-dimensional and non-linear interactions and risk chain propagation effects between trains, between trains and the ground, and between equipment and the environment throughout the entire train operation process; and weak predictive ability, primarily diagnosing current or near-term conditions and lacking the ability to proactively predict the evolution of future risk scenarios.

[0006] The disadvantages of the above-mentioned existing technical solutions include:

[0007] Lack of high-fidelity fault modeling for signal systems across all scenarios: Existing fault modeling methods for signal systems are limited to specific scenarios and single fault types, resulting in insufficient model universality and realism.

[0008] Insufficient consideration of the spatiotemporal factors of complex coupled risks: Existing risk assessment models are mostly based on abstract network analysis, which fails to effectively integrate the spatiotemporal physical attributes of risks. They lack quantitative modeling of the spatial location, temporal evolution and dynamic coupling relationship between risk events and train operation status, resulting in a mismatch between assessment results and real operation scenarios, and insufficient spatiotemporal relevance of early warning information.

[0009] Lack of interpretable train status simulation and scheduling functions: The existing system lacks interpretable train status simulation and scheduling simulation functions, and cannot predict multiple probability evolution paths in the short time domain, resulting in passive safety decision-making and difficulty in actively intervening in the risk evolution process. Summary of the Invention

[0010] This application provides a risk prediction and strategy generation method and system to realize the dynamic evolution and prediction of FAO in all scenarios of rail transit, provide accurate spatiotemporal quantitative risk assessment, and generate interpretable and interventionable proactive decision support.

[0011] Firstly, a risk prediction and strategy generation method is provided, including the following steps:

[0012] A scenario evolution model for rail transit operation is constructed to obtain the FAO scenario evolution model;

[0013] Based on the FAO scenario evolution model, the scenario evolution safety boundary is quantified to obtain the scenario boundary and the corresponding scenario parameter set;

[0014] Based on the scene boundary and the corresponding scene parameter set, the key features of FAO scene evolution are identified.

[0015] In the above technical solution, a scenario evolution model for rail transit operation is constructed to obtain a scenario evolution model for FAO (Fast Occupational Occupation); based on the FAO scenario evolution model, the scenario evolution safety boundary is quantified to obtain the scenario boundary and the corresponding scenario parameter set; based on the scenario boundary and the corresponding scenario parameter set, the key features of FAO scenario evolution are identified; thus, the full-scenario dynamic evolution and prediction of rail transit FAO is realized, accurate spatiotemporal quantitative risk assessment is provided, and interpretable and interventionist proactive decision support is generated.

[0016] In a specific feasible implementation plan, the steps for constructing a scenario evolution model for rail transit operation to obtain the FAO scenario evolution model include:

[0017] Construct a scenario knowledge base for rail transit, and obtain the scenario knowledge base;

[0018] Construct a scenario-based multi-agent evolution model for the FAO system to obtain the multi-agent model;

[0019] Based on the scenario knowledge base and the multi-agent model, the dynamic evolution and prediction of rail transit across all scenarios are completed.

[0020] In a specific feasible implementation, the steps for constructing a scenario-based multi-agent evolution model of the FAO system to obtain the multi-agent model specifically include:

[0021] Construct a single agent to obtain a single agent model;

[0022] Based on the single-agent model, a scenario-multi-agent evolution model of the FAO system is constructed to obtain the multi-agent model.

[0023] In one specific implementation scheme, based on the scenario knowledge base and the multi-agent model, the steps for completing the dynamic evolution and prediction of rail transit across all scenarios specifically include:

[0024] Real-time key feature vectors generated using the multi-agent model;

[0025] Based on the real-time key feature vectors, the matching degree is calculated with the scene knowledge base to complete the dynamic evolution and prediction of the entire scene of rail transit.

[0026] In one specific implementation scheme, the step of quantifying the scene evolution safety boundary based on the FAO scene evolution model to obtain the scene boundary and the corresponding scene parameter set specifically includes:

[0027] The static risk field is quantified to obtain a static risk field model;

[0028] The dynamic risk field is quantified to obtain a dynamic risk field model;

[0029] Based on the static risk field model and the dynamic risk field model, a safety scenario boundary is constructed; and the safety scenario boundary is used to assess the safety status of rail transit operation and provide risk warnings.

[0030] In one specific implementation scheme, the static risk field model includes the following formula:

[0031] ;

[0032] in, for Static risks at the location, For static devices The intensity of the risk, , Represents the location of the equipment. , This represents the radius of influence in the x and y directions. To control the steepness of risk decay;

[0033] ;

[0034] in, Mean time between failures (MTBF) for device i; Risk of equipment malfunction; This poses a risk of fault propagation.

[0035] In one specific implementation scheme, the dynamic risk field model includes the following formula:

[0036] ;

[0037] ;

[0038] ;

[0039] ;

[0040] in, For dynamic vehicles The intensity of the risk; , They represent the first The extent to which vehicle speed and relative distance affect lateral risk; , They represent the first The extent to which vehicle speed and relative distance affect longitudinal risk; , Representing the first The length and width of the vehicle; , These represent the train's lateral speed and longitudinal speed, respectively. For dynamic vehicles At position ( Dynamic risks at this location; For dynamic vehicles At position ( The attenuation coefficient at point ) This is the transverse attenuation coefficient. This is the longitudinal attenuation coefficient.

[0041] In one specific implementation scheme, the step of identifying key features of FAO scene evolution based on the scene boundary and the corresponding scene parameter set specifically includes:

[0042] Based on the risk model, the safety margin of the evolution path of the orbit operation scenario is quantified, and high-risk paths are screened and located.

[0043] The state differences between high-risk paths and baseline safe paths throughout the entire scenario simulation are quantitatively assessed, and key anomaly features are located.

[0044] The key anomaly features are used for structured early warning and strategy generation.

[0045] In one specific implementation, the risk model includes a dynamic / static risk field coupling model.

[0046] Secondly, a risk prediction and strategy generation system is provided, including:

[0047] The scenario simulation and prediction module is used to construct a scenario evolution model for rail transit operation, resulting in the FAO scenario evolution model.

[0048] The risk dynamic quantification module is used to quantify the safety boundary of the scenario evolution based on the FAO scenario evolution model, and obtain the scenario boundary and the corresponding scenario parameter set.

[0049] The key feature identification module is used to identify key features of FAO scene evolution based on the scene boundary and the corresponding scene parameter set.

[0050] In the above technical solution, a scenario inference and prediction module is set up to construct a scenario evolution model for rail transit operation, thereby obtaining a FAO scenario evolution model; a risk dynamic quantification module is used to quantify the scenario evolution safety boundary based on the FAO scenario evolution model, thereby obtaining the scenario boundary and the corresponding scenario parameter set; a key feature identification module is used to identify the key features of FAO scenario evolution based on the scenario boundary and the corresponding scenario parameter set; thus, the full-scenario dynamic evolution and prediction of rail transit FAO is realized, accurate spatiotemporal quantitative risk assessment is provided, and interpretable and interventionist proactive decision support is generated. Attached Figure Description

[0051] Figure 1 A flowchart illustrating the risk prediction and strategy generation method provided in this application embodiment;

[0052] Figure 2 This is a structural block diagram of the risk prediction and strategy generation system provided in the embodiments of this application;

[0053] Figure 3 This is a schematic diagram of the structure of a partial scenario library provided in the embodiments of this application;

[0054] Figure 4 This is a schematic diagram of the MAS-based autonomous train system modeling architecture provided in an embodiment of this application;

[0055] Figure 5 A flowchart of the scene matching algorithm provided in the embodiments of this application;

[0056] Figure 6 A schematic diagram of the parameter boundaries for a positive line operation scenario provided in an embodiment of this application;

[0057] Figure 7 A GIS schematic diagram of the area near Yancun East Station provided in this application embodiment. Detailed Implementation

[0058] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. Through these descriptions, the features and advantages of the present application will become clearer and more apparent.

[0059] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments. Although various aspects of embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless specifically indicated otherwise.

[0060] Furthermore, the technical features involved in the different embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0061] To facilitate understanding of the risk prediction and strategy generation method and system provided in this application embodiment, its application scenario is first explained. The risk prediction and strategy generation method and system provided in this application embodiment are used to realize the dynamic evolution and prediction of rail transit FAO across all scenarios, provide accurate spatiotemporal quantitative risk assessment, and generate interpretable and interventionist proactive decision support. The shortcomings of existing technical solutions include: lack of high-fidelity signal system fault modeling across all scenarios: existing signal system fault modeling methods are limited to specific scenarios and single fault types, resulting in insufficient model universality and realism. Insufficient consideration of spatiotemporal factors of complex coupled risks: existing risk assessment models are mostly based on abstract network analysis, failing to effectively integrate the spatiotemporal physical attributes of risks, lacking quantitative modeling of the spatial location, temporal evolution, and dynamic coupling relationship between risk events and train operation status, leading to a mismatch between assessment results and real operation scenarios, and insufficient spatiotemporal targeting of early warning information. Lack of interpretable train state deduction and scheduling functions: existing systems lack interpretable train state deduction and scheduling simulation functions, unable to predict multi-probability evolution paths in the short time domain, resulting in passive safety decisions and difficulty in proactively intervening in the risk evolution process. Therefore, this application provides a risk prediction and strategy generation method and system to realize the dynamic evolution and prediction of rail transit FAO in all scenarios, provide accurate spatiotemporal quantitative risk assessment, and generate interpretable and interventionist proactive decision support. The following detailed description, in conjunction with specific accompanying drawings, illustrates the embodiments.

[0062] exist Figure 1 In this application, an embodiment provides a risk prediction and strategy generation method, including the following steps:

[0063] A scenario evolution model for rail transit operation is constructed to obtain the FAO scenario evolution model;

[0064] Based on the FAO scenario evolution model, the scenario evolution safety boundary is quantified to obtain the scenario boundary and the corresponding scenario parameter set;

[0065] Based on the scene boundary and the corresponding scene parameter set, the key features of FAO scene evolution are identified.

[0066] In the above technical solution, a scenario evolution model for rail transit operation is constructed to obtain a scenario evolution model for FAO (Fast Occupational Occupation); based on the FAO scenario evolution model, the scenario evolution safety boundary is quantified to obtain the scenario boundary and the corresponding scenario parameter set; based on the scenario boundary and the corresponding scenario parameter set, the key features of FAO scenario evolution are identified; thus, the full-scenario dynamic evolution and prediction of rail transit FAO is realized, accurate spatiotemporal quantitative risk assessment is provided, and interpretable and interventionist proactive decision support is generated.

[0067] Specifically, the beneficial effects include:

[0068] This application achieves forward-looking dynamic evolution and prediction across all scenarios: By using multi-agent modeling to perform integrated simulation and parallel interactive deduction of the system, it overcomes the limitations of traditional methods that rely on fixed rules and static scenarios. It can simulate the system state under the coupling of multiple factors in the short time domain in the future through multi-path and probabilistic scenario evolution, thereby supporting forward-looking security early warning and decision-making.

[0069] It provides accurate spatiotemporal quantitative risk assessment: This application deeply couples multi-agent inference with static and dynamic spatiotemporal risk field models, so that risk assessment is transformed from abstract network analysis into a quantitative description closely related to specific physical locations, time nodes and train operation status, realizing accurate positioning and visualization of risks in real operation scenarios.

[0070] This application generates interpretable and interventionable proactive decision support: By defining safety boundaries and comparing path characteristics, it establishes a transmission identification model from micro-characteristics such as equipment health to macro-scenario risks. This model can automatically trace the root causes of key risks and quantify their contribution, thereby generating structured decision information that integrates root cause diagnosis and control plans, realizing the transformation from passive alarm to proactive and precise intervention.

[0071] In a specific feasible implementation plan, the steps for constructing a scenario evolution model for rail transit operation to obtain the FAO scenario evolution model include:

[0072] Construct a scenario knowledge base for rail transit, and obtain the scenario knowledge base;

[0073] Construct a scenario-based multi-agent evolution model for the FAO system to obtain the multi-agent model;

[0074] Based on the scenario knowledge base and the multi-agent model, the dynamic evolution and prediction of rail transit across all scenarios are completed.

[0075] In a specific feasible implementation, the steps for constructing a scenario-based multi-agent evolution model of the FAO system to obtain the multi-agent model specifically include:

[0076] Construct a single agent to obtain a single agent model;

[0077] Based on the single-agent model, a scenario-multi-agent evolution model of the FAO system is constructed to obtain the multi-agent model.

[0078] In one specific implementation scheme, based on the scenario knowledge base and the multi-agent model, the steps for completing the dynamic evolution and prediction of rail transit across all scenarios specifically include:

[0079] Real-time key feature vectors generated using the multi-agent model;

[0080] Based on the real-time key feature vectors, the matching degree is calculated with the scene knowledge base to complete the dynamic evolution and prediction of the entire scene of rail transit.

[0081] In a specific feasible implementation plan, the steps for constructing a scenario evolution model for rail transit operation to obtain the FAO scenario evolution model include:

[0082] 1. Construction of Scenario Knowledge Base

[0083] Based on the "Scenario Specifications for Fully Automated Driving Systems," this paper outlines the components involved, information transmission paths, and functional dependencies in each scenario. It also analyzes the chain reaction effects triggered by fault injection, ultimately forming a standardized scenario library covering 134 core scenarios. These scenarios are categorized according to fault severity: basic scenarios, accident scenarios, operational adjustment scenarios, emergency scenarios, and fault scenarios. Furthermore, the scenarios are structured using real-time train location, train operation mode, track environmental parameters, and component fault status as scenario elements.

[0084] For each scenario in the scenario library, we conduct in-depth analysis of the functional execution status of each component, accurately extract the component failure characteristics, and match them with the corresponding scenarios in the scenario library. We clarify the component characteristics covered by each scenario, thereby establishing the correlation mapping relationship between component failure characteristics and scenarios.

[0085] Some scene libraries, such as Figure 3 As shown, different fault characteristics within a scenario influence the potential evolution direction. Each scenario can be reached through multiple branching paths, collectively constructing a scenario evolution network. Taking "route arrangement failure" as an example, although "turnout switching failure" is one of the main characteristics leading to this scenario, when accompanied by "turnout status loss," the two form a coupled influence, with the latter having a more prominent decisive role in the scenario's consequences. Fault characteristics not only independently affect the scenario's evolution direction but also interact and compete, ultimately determining the dominant characteristics and consequences of the scenario. By extracting and analyzing this coupling relationship, potential major influencing factors can be identified more accurately.

[0086] 2. FAO System Scenario Evolution Model Based on MAS Modeling

[0087] This section forms the basis for subsequent analysis. Its focus is on constructing a high-fidelity simulation environment across the entire scenario, incorporating GIS data to simulate the complex interactions between various components. It also considers spatiotemporal influencing factors, fully exploring the future evolutionary path of the vehicle under specific conditions. This includes:

[0088] 2.1 Single Agent Modeling

[0089] The GAMA platform, as an open-source multi-agent simulation environment, supports complex spatial modeling, hierarchical agent architecture design, and large-scale concurrent interactive simulation. To formally describe the behavioral logic of each subsystem of the urban rail transit system, this work abstracts it into a quintuple structure:

[0090]

[0091] in:

[0092] ID: Agent ID;

[0093] Attribute: Basic attributes of an agent;

[0094] Perception: A collection of perceptions, including key functional parameters, interaction information from other intelligent agents, and changes in the operating environment;

[0095] Decision: A set of decisions that determine the next course of action based on current perception;

[0096] Action: The specific behavior.

[0097] For example, the ATP subsystem is shown in Table 1.

[0098] tuple Specific content Id ATP(0) Attribute current_v, MA, precise_p, etc. Perception isolateDoorID, subsystem interaction information, control commands, etc. Decision Fault isolation ID translation, overspeed protection, autonomous positioning, etc. Action Output EB, share information with ATO, and control PSD switch, etc.

[0099] Table 1 Behavioral Logic Table of ATP Subsystem

[0100] 2.2 Multi-agent modeling

[0101] Based on single-agent modeling, a multi-level architecture is introduced:

[0102] The train is abstracted as an intelligent agent, and its onboard equipment (ATP / ATO, etc.) works collaboratively as internal micro-agents; trackside equipment (such as signals and axle counters) are deployed as independent macro agents, without hierarchical inclusion relationships.

[0103] Inter-agent interaction employs the FIPA-ACL standard communication protocol. FIPA-ACL stands for "Agent Communication Language," a standard developed by the Foundation for Intelligent Physical Agents (FIPA). It is primarily used for coordination and communication in distributed environments, and its design aims to provide a universal, standardized approach to support cross-platform agent interaction. In this work, structured message passing is used to achieve cross-subsystem collaboration, ensuring the standardization and scalability of information exchange.

[0104] This design supports dynamic interaction in multi-vehicle tracking scenarios and clearly distinguishes the logical boundaries between on-board and trackside equipment. The overall system modeling architecture is as follows: Figure 4As shown, it includes 8 types of macro-intelligent agents and 14 types of micro-intelligent agents.

[0105] In addition, traditional signal system modeling lacks spatial dimension correlation, but assessing the threat level of a fault to trains in a specific section is related to the actual line topology, and risk propagation also depends on physical distance. Therefore, introducing GIS is an important part of realizing high-fidelity train autonomous operation system scenario evolution modeling.

[0106] This work utilizes the QGIS tool to extract track data for a specific line. Simultaneously, it combines this data with a station signal layout map to approximate the locations of trackside equipment. Some of the extracted GIS track data is shown below. Figure 7 As shown, brown represents stations, green represents the actual track locations, and pink dots represent various signaling devices. The advantages of the model integrated with GIS are reflected in:

[0107] (1) Spatial source of fault: Based on geographic coordinate mapping, the precise location of the fault (e.g., “the signal cannot be displayed at {(12924381.40069999918341637 4826428.8110999995842576)}”) is used to distinguish key scenarios such as tunnels / platforms;

[0108] (2) Risk propagation quantification: Combine the dynamic calculation of the fault impact radius with the line topology to predict the spatial threat level to trains operating in the neighboring area.

[0109] 3. Construction of Scene Matching Algorithm

[0110] Building upon the completion of the scenario library and MAS model construction, this section further develops a scenario matching algorithm. This algorithm collects real-time key feature vectors generated by the MAS evolution model, including ATP status, ATO mode, TGMS signal, ZC equipment status, and turnout position. These vectors are then compared with the "scenario library" from step one to calculate the matching degree. The flowchart is shown below. Figure 5 As shown.

[0111] The specific matching formula is as follows:

[0112]

[0113]

[0114] in, This represents the first scene in the scene library. A scenario and Moment Scene Features The matching degree, with a value ranging from 0 to 1; Representing a scene The Middle Each feature weight. It is a piecewise function, if the feature to be matched Appearing in a certain scene, The scene weight value corresponding to this feature is taken, otherwise it is -1. Scene matching degree is determined by... Sum each item and normalize it to the highest matching score in scenario l.

[0115] In this embodiment, the autonomous evolution process of MAS is used to generate complex and comprehensive scene features across multiple time steps. Then, the scene is matched with a pre-prepared knowledge base through a real-time scene matching algorithm. The combination of these two methods realizes the scene evolution process without the need to artificially construct a massive scene evolution network.

[0116] In one specific implementation scheme, the step of quantifying the scene evolution safety boundary based on the FAO scene evolution model to obtain the scene boundary and the corresponding scene parameter set specifically includes:

[0117] The static risk field is quantified to obtain a static risk field model;

[0118] The dynamic risk field is quantified to obtain a dynamic risk field model;

[0119] Based on the static risk field model and the dynamic risk field model, a safety scenario boundary is constructed; and the safety scenario boundary is used to assess the safety status of rail transit operation and provide risk warnings.

[0120] In one specific implementation scheme, the static risk field model includes the following formula:

[0121] ;

[0122] in, for Static risks at the location, For static devices The intensity of the risk, , Represents the location of the equipment. , This represents the radius of influence in the x and y directions. To control the steepness of risk decay;

[0123] ;

[0124] in, Mean time between failures (MTBF) for device i; Risk of equipment malfunction; This poses a risk of fault propagation.

[0125] In one specific implementation scheme, the dynamic risk field model includes the following formula:

[0126] ;

[0127] ;

[0128] ;

[0129] ;

[0130] in, For dynamic vehicles The intensity of the risk; , They represent the first The extent to which vehicle speed and relative distance affect lateral risk; , They represent the first The extent to which vehicle speed and relative distance affect longitudinal risk; , Representing the first The length and width of the vehicle; , These represent the train's lateral speed and longitudinal speed, respectively.

[0131] In a specific feasible implementation, the steps of quantifying the safety boundary of the scenario evolution based on the FAO scenario evolution model to obtain the scenario boundary and the corresponding scenario parameter set are designed based on the core understanding of "gradual accumulation of risk, and sudden occurrence of danger": risk itself is a physical quantity that continuously and gradually changes and accumulates in a spatiotemporal field, while the occurrence of a dangerous state is a discrete jump event when the risk value exceeds the safety threshold or the spatial margin is exhausted. Therefore, a risk field model is used to describe and calculate the risk state of each future scenario node in the simulation in a refined and spatial manner. This model can simultaneously characterize the continuous impact of trackside and onboard risk sources due to their own gradual decay, as well as the instantaneous changes caused by sudden disturbances, thereby accurately judging the comprehensive threat level faced by the train at the present and future moments, and clarifying its distance from dangerous jumps. Specifically, this includes:

[0132] 1. Static risk field quantification

[0133] Static risk fields are generated by trackside / ground fixed facilities (such as signals, axle counters, turnouts, etc.), and their characteristics are that the spatial location of the risk source is constant and the range of influence is unchanging.

[0134] ;

[0135] for Static risks at the location, For static devices The intensity of the risk, , Represents the location of the equipment. , This represents the radius of influence in the x and y directions. Control the steepness of risk decay.

[0136] ;

[0137] Risk intensity This includes the risk of random equipment failure, the risk of equipment malfunction, and the risk of fault propagation.

[0138] in, : Mean time between failures for device i, representing random disturbances / failures;

[0139] Equipment malfunction risk Measured by the functional health of the equipment;

[0140] Risk of fault propagation This refers to a situation where the device itself functions normally, but due to other malfunctions, it cannot perform its function properly, resulting in relatively serious consequences. This is measured by the device's health status.

[0141] 2. Dynamic Risk Field Quantification

[0142] The dynamic risk field is generated by the moving train. It includes the following formulas:

[0143] ;

[0144] ;

[0145] ;

[0146] ;

[0147] For dynamic vehicles The intensity of the risk. , They represent the first The degree to which vehicle speed and relative distance affect lateral risk. , Indicates the first The degree to which vehicle speed and relative distance affect longitudinal risk. , Representing the first The length and width of the vehicle. , Represents the train's lateral and longitudinal speeds.

[0148] The health of each component included in the Train This is referred to as a semi-static risk factor, meaning it is a static factor in itself, but acquires dynamic characteristics due to its movement along with the train. Therefore, by combining the risk values ​​of each onboard device included in the train with different AHP weights under different basic scenarios, we obtain... , which characterizes the strength of the dynamic risk field.

[0149] Specifically, in a feasible implementation plan, combined with the existing MAS urban rail model, firstly, the risk field quantification model can accurately characterize the propagation range of different types of risks; secondly, risk field superposition calculations are performed with the operating train as the center, and irrelevant risk sources are filtered out through spatial topology relationships to focus on assessing the comprehensive risks currently faced by the train. Finally, the risk field can also output the environmental risk level of each section based on the line link segmentation, providing spatial decision support for the dispatching system.

[0150] 3. Security Scenario Boundary Construction

[0151] After quantifying risks and describing scenarios for autonomous trains, there is still a lack of understanding of the train's operational safety status—it is impossible to confirm whether the current operating state conforms to the verified safety model, nor is it easy to predict the development trend of state deviation. However, by establishing mature safety scenario boundaries, the system can compare real-time operating parameters with fully verified safety scenario benchmarks, thereby achieving continuous assessment of the safety status and early warning of risk deviations.

[0152] First, we need to define a security function.

[0153]

[0154] in, Risks to be implemented for trains, For the set risk threshold, The location where the risk source of the train decays by 90% along the track is defined as the train risk boundary. Similarly, The risk sources representing trackside equipment can be viewed as the risk boundaries of virtual obstacles. This refers to the emergency braking distance of the train. In other words, this work considers a train to be safe when the risk is within a threshold and the distance between the train and the risk source is greater than the emergency braking distance.

[0155] Tabu Search is used to explore the parameter space and determine the parameter boundaries under different basic scenarios. The step size for each parameter is 0.05. The objective function is as follows:

[0156]

[0157]

[0158] For the same operational scenario, the earlier the safety boundary is triggered, the higher the potential risk to the system. Based on this, we focus on exploring the set of unsafe states. Within the generated unsafe set, we use a minimization method to accurately extract the health value boundary of each parameter and establish it as the safety boundary for that scenario. See [link to details]. Figure 6 Once the boundary is obtained, the quantified safety margin, such as "time margin (seconds)" or "space margin (meters)," is output by calculating the "distance" between the current parameter and the boundary value.

[0159] In this embodiment, the output of the scenario simulation of rail transit operation (including MAS scenario simulation) is combined with a "spatiotemporal risk field" model that integrates static basic risk and dynamic propagation risk. This model aims to quantify the risk level and safety margin assessment of different paths and different spatiotemporal nodes in the simulation, providing a risk assessment perspective that is more in line with the actual spatiotemporal characteristics of operation.

[0160] In one specific implementation scheme, the step of identifying key features of FAO scene evolution based on the scene boundary and the corresponding scene parameter set specifically includes:

[0161] Based on the risk model, the safety margin of the evolution path of the orbit operation scenario is quantified, and high-risk paths are screened and located.

[0162] The state differences between high-risk paths and baseline safe paths throughout the entire scenario simulation are quantitatively assessed, and key anomaly features are located.

[0163] The key anomaly features are used for structured early warning and strategy generation.

[0164] In one specific implementation, the risk model includes a dynamic / static risk field coupling model.

[0165] The construction process of the dynamic / static risk field coupling model is as follows:

[0166] The static risk field is quantified to obtain a static risk field model;

[0167] The dynamic risk field is quantified to obtain a dynamic risk field model.

[0168] Furthermore, the static risk field model includes the following formula:

[0169] ;

[0170] in, for Static risks at the location, For static devices The intensity of the risk, , Represents the location of the equipment. , This represents the radius of influence in the x and y directions. To control the steepness of risk decay;

[0171] ;

[0172] in, Mean time between failures (MTBF) for device i; Risk of equipment malfunction; This poses a risk of fault propagation.

[0173] Furthermore, the dynamic risk field model includes the following formula:

[0174] ;

[0175] ;

[0176] ;

[0177] ;

[0178] in, For dynamic vehicles The intensity of the risk; , They represent the first The extent to which vehicle speed and relative distance affect lateral risk; , They represent the first The extent to which vehicle speed and relative distance affect longitudinal risk; , Representing the first The length and width of the vehicle; , These represent the train's lateral speed and longitudinal speed, respectively.

[0179] Furthermore, combining the existing MAS urban rail model, firstly, the risk field quantification model can accurately characterize the propagation range of different types of risks; secondly, by performing risk field superposition calculations centered on operating trains, irrelevant risk sources are filtered out through spatial topological relationships, focusing on assessing the comprehensive risks currently faced by the trains. Finally, the risk field can also output the environmental risk level of each section based on the line link, providing spatial decision support for the dispatching system.

[0180] In the aforementioned technical solutions, the static risk field model can accurately characterize relatively stable risk factors in rail transit systems, such as risks arising from line layout and inherent equipment characteristics; while the dynamic risk field model can capture dynamic risk factors such as fluctuations in equipment operating status. The combination of these two models enables a comprehensive consideration of different risk factors, accurately quantifying the risk levels of different paths and spatiotemporal nodes in rail transit operation scenario simulations, thus providing a solid data foundation for risk assessment.

[0181] In a specific feasible implementation, the step of identifying key features of FAO scenario evolution based on the scenario boundary and the corresponding scenario parameter set, through in-depth analysis of the scenario evolution network generated by multi-agent inference, achieves a closed loop from risk prediction to decision support. The occurrence of risk is essentially an evolutionary process from gradual to abrupt change: gradual factors such as equipment performance degradation and accumulated scheduling deviations continuously reduce the system's safety margin, and when these risks accumulate to a critical point or encounter sudden disturbances, the system may jump from a safe state to a dangerous state. The core design of this method is to simultaneously capture these two types of risk characteristics—identifying both the gradual factors that lead to the slow accumulation of risk and warning of abrupt signals that may trigger state jumps, thereby achieving full-process monitoring and prediction of risk evolution. Specifically, this includes:

[0182] 1. Screening and Locating High-Risk Paths

[0183] Based on a dynamic / static risk field coupling model, the safety margin of each evolution path is quantified. By comparing the path risk curves with preset safety boundaries, high-risk paths with insufficient safety margins are automatically identified and screened, thus completing the initial risk classification.

[0184] 2. Key Feature Origin Tracing and Contribution Analysis

[0185] To accurately identify the root causes driving risk evolution, this step provides a quantitative analysis method based on multi-path comparison and log tracing. This method clearly distinguishes risk characteristics into two categories: gradual characteristics and abrupt changes. The former continuously deviates from normal values ​​in dangerous paths, leading to slow risk accumulation; the latter undergoes drastic changes during evolution, directly triggering or accelerating risk jumps. Through a full-scenario evolution network generated in parallel simulation (containing all safe and risky paths), the system systematically locates key anomaly characteristics from two dimensions.

[0186] (1) Path set construction and initial screening of abnormal features

[0187] To overcome the limitations of traditional analyses based on single paths or limited scenario comparisons, this method utilizes a full-scenario evolution network generated through parallel inference for global statistical analysis. Starting from the same initial state, the system generates a set containing multiple evolution paths through parallel inference, and explicitly divides each path into a set of safe paths based on its final risk state. With dangerous path set By aligning the time series of all paths, the system uses the following two core statistical indicators to initially screen out a set of anomalous features that play a key role in the evolution of danger from a global perspective.

[0188] Feature Discrimination Index: This index measures the degree of difference in the overall distribution of a feature across safe and hazardous path sets. It is mainly used to identify "gradual" anomalous features that persist in hazardous paths. Although these features do not cause drastic changes, they provide the basic conditions for risk jumps.

[0189]

[0190] in, and Features In the safe path set With dangerous path set The mean of the above, and This represents the corresponding variance. The higher the value, the more significant the feature is as a "background risk" in dangerous paths.

[0191] Evolutionary Path Contribution: This indicator quantifies the real-time contribution of changes in characteristics to the evolution of a path towards danger over time, and is used to capture anomalous "mutation" features that play a key driving role in the specific evolution process. Changes in these features are directly related to jumps in the safety state.

[0192]

[0193] in, Representation of features At time step The range of change. For path difference weights, and The sets of safe and dangerous paths are respectively located at time... Average safety margin, This is the sensitivity parameter. The higher the value, the more critical the dynamic role of this feature in driving risk evolution.

[0194] To ensure comprehensive coverage of potential risk factors, including systemic background anomalies and sudden critical disturbances, the system employs a lenient initial screening strategy: [The system will then consider factors that meet the following criteria]. or Features of any one of these conditions are initially included in the candidate set of key anomaly features. and This is a threshold set based on experience or distribution. This candidate set will serve as input for subsequent refined contribution analysis.

[0195] (2) Anomaly feature extraction and quantitative evaluation

[0196] Based on the key anomaly feature candidate set obtained from the above steps, this step performs quantitative evaluation at a finer temporal resolution. For each feature in the candidate set... In dangerous paths With reference safety path A comparative analysis was conducted to calculate three core quantitative indicators:

[0197] Timing Deviation Index Characterizes the timing and severity of abnormal occurrences. Its calculation formula is:

[0198]

[0199] in, for exist First deviation in the middle The time within the normal fluctuation range; This represents the absolute value of the instantaneous rate of change at that moment; , These are the weighting coefficients.

[0200] Amplitude Deviation Index Quantify the anomalous magnitude of the feature within the divergence window. Its calculation formula is:

[0201]

[0202] Influence Transmission Index : Assess the breadth of the impact of feature anomalies on subsequent system interactions and decision-making.

[0203] analyze back MAS event logs, statistics due to Number of key decision events directly triggered or significantly altered by state anomalies Based on this calculation:

[0204]

[0205] in This represents the total number of critical events within a certain period after the window opens.

[0206] (3) Comprehensive contribution analysis and key feature determination

[0207] After normalizing the above indicators, a weighted composite contribution score for each feature is calculated. :

[0208]

[0209] in, , , These are the normalized index values, , , For configurable weights, satisfy Then, the overall contribution score will be calculated. The most significant anomaly is defined as the key anomaly.

[0210] 3. Structured Early Warning and Strategy Generation

[0211] Based on the above analysis, actionable decision-making information is generated:

[0212] Prediction and profiling: Provides probability, risk profiles, and safety margin time series diagrams for core evolutionary paths.

[0213] Root cause diagnosis: Exceeding the set threshold Furthermore, the top-ranked features (such as the top 3) are identified as key features, and a structured root cause list is output to accurately locate the risk source. For example: "The main reason for the decrease in safety margin of path P1 (contribution of 80%) is that the ATP communication health of train T203 dropped from 92 to 58 within 180 seconds."

[0214] Generate semantic strategies: Transform the results into natural language warnings, clarify the risk patterns and risk evolution, and push targeted prevention and control plans and key scenario parameters based on scenario knowledge base matching, forming a decision-making closed loop.

[0215] In this embodiment, passive alarms are transformed into proactive decision support with foresight and interpretability: by defining safety boundaries and comparing path characteristics, a transmission identification model is established from micro-characteristics such as equipment health to macro-scenario risks. This model can automatically trace the root causes of key risks and quantify their contribution, thereby generating structured decision information that integrates root cause diagnosis and control plans. This generates interpretable and interventionable proactive decision support, realizing the transformation from passive alarms to proactive and precise intervention.

[0216] exist Figure 2 In this application, an embodiment provides a risk prediction and strategy generation system, including:

[0217] The scenario simulation and prediction module is used to construct a scenario evolution model for rail transit operation, resulting in the FAO scenario evolution model.

[0218] The risk dynamic quantification module is used to quantify the safety boundary of the scenario evolution based on the FAO scenario evolution model, and obtain the scenario boundary and the corresponding scenario parameter set.

[0219] The key feature identification module is used to identify key features of FAO scene evolution based on the scene boundary and the corresponding scene parameter set.

[0220] In the above technical solution, a scenario inference and prediction module is set up to construct a scenario evolution model for rail transit operation, thereby obtaining a FAO scenario evolution model; a risk dynamic quantification module is used to quantify the scenario evolution safety boundary based on the FAO scenario evolution model, thereby obtaining the scenario boundary and the corresponding scenario parameter set; a key feature identification module is used to identify the key features of FAO scenario evolution based on the scenario boundary and the corresponding scenario parameter set; thus, the full-scenario dynamic evolution and prediction of rail transit FAO is realized, accurate spatiotemporal quantitative risk assessment is provided, and interpretable and interventionist proactive decision support is generated.

[0221] Those skilled in the art will know that this application can be implemented as a system, method, or computer program product.

[0222] Therefore, this disclosure can be implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this application can also be implemented as a computer program product in one or more computer-readable media, the computer-readable media containing computer-readable program code.

[0223] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0224] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application. Based on this, various substitutions and improvements can be made to this application, all of which fall within the protection scope of this application.

Claims

1. A method for risk prediction and strategy generation, characterized in that, Includes the following steps: A scenario evolution model for rail transit operation is constructed to obtain the FAO scenario evolution model; Based on the FAO scenario evolution model, the scenario evolution safety boundary is quantified to obtain the scenario boundary and the corresponding scenario parameter set; Based on the scene boundary and the corresponding scene parameter set, the key features of FAO scene evolution are identified.

2. The risk prediction and strategy generation method according to claim 1, characterized in that, The steps to construct a scenario evolution model for rail transit operation and obtain the FAO scenario evolution model specifically include: Construct a scenario knowledge base for rail transit, and obtain the scenario knowledge base; Construct a scenario-based multi-agent evolution model for the FAO system to obtain the multi-agent model; Based on the scenario knowledge base and the multi-agent model, the dynamic evolution and prediction of rail transit across all scenarios are completed.

3. The risk prediction and strategy generation method according to claim 2, characterized in that, The steps for constructing a scenario-based multi-agent evolution model of the FAO system to obtain the multi-agent model specifically include: Construct a single agent to obtain a single agent model; Based on the single-agent model, a scenario-multi-agent evolution model of the FAO system is constructed to obtain the multi-agent model.

4. The risk prediction and strategy generation method according to claim 3, characterized in that, Based on the aforementioned scenario knowledge base and the aforementioned multi-agent model, the steps for completing the dynamic evolution and prediction of rail transit across all scenarios specifically include: Real-time key feature vectors generated using the multi-agent model; Based on the real-time key feature vectors, the matching degree is calculated with the scene knowledge base to complete the dynamic evolution and prediction of the entire scene of rail transit.

5. The risk prediction and strategy generation method according to claim 4, characterized in that, Based on the aforementioned FAO scenario evolution model, the steps of quantifying the scenario evolution safety boundary to obtain the scenario boundary and the corresponding scenario parameter set specifically include: The static risk field is quantified to obtain a static risk field model; The dynamic risk field is quantified to obtain a dynamic risk field model; Based on the static risk field model and the dynamic risk field model, a safety scenario boundary is constructed; and the safety scenario boundary is used to assess the safety status of rail transit operation and provide risk warnings.

6. The risk prediction and strategy generation method according to claim 5, characterized in that, The static risk field model includes the following formula: ； in, for Static risks at the location, For static devices The intensity of the risk, , Represents the location of the equipment. , This represents the radius of influence in the x and y directions. To control the steepness of risk decay; ； in, Mean time between failures (MTBF) for device i; Risk of equipment malfunction; This poses a risk of fault propagation.

7. The risk prediction and strategy generation method according to claim 6, characterized in that, The dynamic risk field model includes the following formula: ； ； ； ； in, For dynamic vehicles The intensity of the risk; , They represent the first The extent to which vehicle speed and relative distance affect lateral risk; , They represent the first The extent to which vehicle speed and relative distance affect longitudinal risk; , Representing the first The length and width of the vehicle; , These represent the train's lateral speed and longitudinal speed, respectively.

8. The risk prediction and strategy generation method according to claim 7, characterized in that, The steps for identifying key features of FAO scene evolution based on the scene boundary and the corresponding scene parameter set specifically include: Based on the risk model, the safety margin of the evolution path of the orbit operation scenario is quantified, and high-risk paths are screened and located. The state differences between high-risk paths and baseline safe paths throughout the entire scenario simulation are quantitatively assessed, and key anomaly features are located. The key anomaly features are used for structured early warning and strategy generation.

9. The risk prediction and strategy generation method according to claim 8, characterized in that, The risk model includes a dynamic / static risk field coupling model.

10. A risk prediction and strategy generation system, characterized in that, include: The scenario simulation and prediction module is used to construct a scenario evolution model for rail transit operation, resulting in the FAO scenario evolution model. The risk dynamic quantification module is used to quantify the safety boundary of the scenario evolution based on the FAO scenario evolution model, and obtain the scenario boundary and the corresponding scenario parameter set. The key feature identification module is used to identify key features of FAO scene evolution based on the scene boundary and the corresponding scene parameter set.

Images