A method for dynamic assessment and risk warning of railway coupling network resilience under multiple disturbance scenarios

Abstract

The application discloses a kind of railway coupling network's tenacity dynamic evaluation and risk early warning method under multiple disturbance scene, comprising the following steps: S1: multi-modal data acquisition and pre-processing;S2: physical-service dual-network coupling modeling;S3: multiple disturbance identification and quantification;S4: network tenacity evaluation;S5: cascading failure evolution simulation;S6: risk early warning and hierarchical determination;S7: emergency decision support.The application is based on physical-service dual-network coupling model, establishes two-way closed-loop coupling feedback mechanism, truly reflects the interactive process of physical anomaly-service adjustment-load change, realizes the dynamic coordination of physical network state and service network scheduling, solves the problem of physical and service split in traditional modeling method, the application can predict risk 4-72 hours in advance by fusing GCN-GRU multi-spatiotemporal prediction model and cascading failure evolution simulation technology, accurately identifies risk transmission path, and the early warning accuracy is improved compared with traditional method.

Description

Technical Field

[0001] This invention belongs to the fields of railway transportation organization optimization, complex network modeling, transportation system resilience assessment and intelligent early warning technology, specifically involving a method for dynamic resilience assessment and risk early warning of railway coupled networks under multiple disturbance scenarios. Background Technology

[0002] As the core backbone of the land transportation corridor and the northward extension link of the Western Land-Sea New Corridor, the western railway undertakes national strategic transportation tasks such as China-Europe freight trains, Xinjiang coal transportation, and cross-border production capacity cooperation. The railway network in this region faces unique challenges such as complex natural environment (high altitude, cold climate, high plateau, sandstorms, permafrost), mixed passenger and freight traffic, and variable operating conditions. Frequent natural disasters along the line, surging transportation demand, and frequent external economic fluctuations bring triple challenges to the railway network, including structural fragility, functional blockage, and systemic risk transmission.

[0003] With the improvement of railway informatization, massive amounts of multidimensional and multimodal railway spatiotemporal data are constantly being generated, but existing technologies have the following problems: Insufficient data fusion and perception capabilities: Multi-source heterogeneous data (train operation, passenger and freight demand, environmental disasters, infrastructure) suffer from inconsistencies in spatial accuracy, sampling frequency, and semantic expression, resulting in prominent data silos and making it difficult to support real-time dynamic analysis; Limitations of network modeling methods: Traditional railway network modeling often focuses on physical topology or service scheduling alone, lacking a dynamic coupling mechanism between physical networks and service networks, and cannot truly reflect the interaction process of physical anomalies and service adjustments; Static resilience assessment methods: Existing resilience assessments are mostly based on static topology analysis, which ignores the dynamic characteristics of transportation service networks and makes it difficult to capture the nonlinear evolution and cascading failure process under multiple disturbances. Poor timeliness of risk warnings: The lack of accurate modeling of complex disturbance causal chains makes it difficult to effectively identify the cross-level propagation path of risks, resulting in warnings lagging behind the development of risks and passive emergency responses; Poor adaptability to the west: Existing technologies are mostly developed based on the eastern plains, which are not adapted to the current situation of sparse sensors and many communication blind spots in western railways. They also do not take into account the unique environmental factors such as high altitude, sandstorms, and permafrost, and are not suitable for the operation characteristics of low redundancy and high-capacity strategic transportation in the west, resulting in poor actual application effects. Summary of the Invention

[0004] (1) Technical problems to be solved To address the shortcomings of existing technologies, the present invention aims to provide a method for dynamic assessment and risk warning of the resilience of railway coupled networks under multi-disturbance scenarios, which can be used to achieve real-time identification, dynamic assessment and intelligent early warning of risks in railway transportation systems.

[0005] (2) Technical solution To address the aforementioned technical problems, this invention provides a method for dynamic resilience assessment and risk warning of railway coupled networks under multiple disturbance scenarios, comprising the following steps: S1. Multimodal data acquisition and preprocessing: Collect multi-source data such as railway infrastructure status, train operation, environmental weather, and passenger and freight demand. Clean, denoise, interpolate, and lightweight process the collected multi-source data to construct a unified spatiotemporal benchmark for railway data spatial foundation. S2. Physical-Service Dual-Network Coupling Modeling: Construct the railway physical network layer and service network layer respectively, establish cross-layer mapping relationship and state synchronization mechanism between nodes and edges, and form a dynamically evolving physical-service dual-network coupling model; S3. Multiple Disturbance Identification and Quantification: Construct a three-dimensional disturbance index system of natural environment, security risk and demand fluctuation to identify different types of disturbances and quantify their intensity and impact range; S4. Network resilience assessment: Based on the coupled network model, a dynamic assessment index system is constructed from three dimensions: structural resilience, functional resilience, and recovery resilience. A method for quantifying resilience decay curves is proposed to assess the network resilience status in real time. S5. Cascade Failure Evolution Simulation: Based on cellular automata and graph neural networks, simulate the cascade propagation process of node failure - line blockage - regional paralysis caused by disturbance, and identify key vulnerabilities and risk transmission paths. S6. Risk warning and classification: Integrate historical data and real-time monitoring information to construct the GCN-GRU risk prediction model, and conduct risk classification and warning based on the three-dimensional indicators of minimum required performance, recovery timeliness and function retention rate. S7. Emergency Decision Support: Based on reinforcement learning, the optimal emergency dispatch and recovery strategies are generated, forming a closed-loop mechanism of risk warning, dynamic adjustment and efficient recovery.

[0006] Preferably, step S2 includes the following steps: S201. Dual-network infrastructure construction and cross-layer mapping: First, taking infrastructure such as stations, sections, maintenance depots, and locomotive depots as nodes and line connections as edges, real-time monitoring data is accessed to construct a dynamically evolving infrastructure operation map and complete the construction of the railway physical network layer. Then, using train number, operation plan, and transportation service as nodes and train operation path as edges, a task-driven virtual network is constructed. The service network topology is dynamically adjusted according to fluctuations in transportation demand and event response. Hypergraph theory is introduced to construct reconfigurable service units, and the service network layer is constructed. Secondly, establish a many-to-many cross-layer mapping relationship between physical nodes, physical edges and service nodes, and service edges to form a cross-layer association matrix or mapping function: Among them, the railway physical network layer transforms the infrastructure state into the service network constraint conditions through state-service mapping, and the service network layer generates virtual topology through optimization engine to drive the resource scheduling and channel allocation of the physical network in reverse. Finally, a two-way closed loop is achieved through a feedback mechanism: changes in the state of the railway physical network layer trigger adjustments in the service network layer, and the adjustment results of the service network layer are fed back to the railway physical network layer to update the resource occupancy status. At this point, the construction of the physical-service dual-network coupling model is completed. S202, Dynamic Coupling and Optimization Loop, specifically includes the collaborative work of the following modules: Data-driven module: Reads real-time data from step S1, performs lightweight data processing, and then inputs the data into the dynamic coupling module; Dynamic Coupling Module: Integrates the real-time status of the railway physical network layer and service network layer, calculates the system performance indicators under the current dual-network coupling, and identifies key aspects that need optimization; Blocking edge removal module: When a physical interval is detected to be completely unavailable due to disaster, failure or construction, the edge is temporarily removed from the service network virtual topology; for edges with reduced capacity, the weight is adjusted to reflect the increase in their passage cost. Node Priority Adjustment Module: Based on the node importance output by the dynamic coupling module, dynamically adjust the service priority of nodes in the service network to ensure that critical node resources are given priority. Intelligent dispatching module: Based on reinforcement learning or genetic algorithms, it generates the optimal train route reallocation and capacity resource rescheduling scheme and executes the adjustment; Infrastructure upgrade module: For physical sections that repeatedly encounter bottlenecks or high risks, output infrastructure upgrade suggestions and provide long-term feedback to the planning department; The feedback mechanism operates continuously: After each operation such as removing blocking edges, adjusting node priorities, intelligent allocation, and upgrading infrastructure is completed, the new network state is transmitted to the data-driven module again through the feedback mechanism, forming a closed-loop iteration. S203, Continuous Optimization Module: The continuous optimization module records the response process and effect of the physical-service dual-network coupling model under each disturbance event, forming a case library. Historical data is used periodically to retrain and fine-tune the parameters of the dynamic coupling module. The optimized parameters are pushed to the dynamic coupling module and the node priority adjustment module to realize the continuous evolution of the physical-service dual-network coupling model. S204, Lightweight and Timeliness Judgment: After completing the optimization loop described in S202, enter the judgment module: determine whether the data meets the requirements of lightweight and timeliness; If the following conditions are met, then output the current physical-service dual-network coupling model and proceed to step S3; If the requirements are not met, the process returns to the data-driven module to re-filter, compress, or prioritize data transmission until the requirements are met.

[0007] Furthermore, step S3 includes the following steps: S301. Construct a multi-dimensional disturbance index system: First, to comprehensively cover the complex risk sources faced by railways, a three-dimensional disturbance index system of natural environment, safety risk, and demand fluctuation is constructed to provide a classification framework for subsequent disturbance identification and quantification, as detailed below: Natural environment disturbance module construction: Integrating climate change factors, natural disaster factors, and environmental change factors, specifically including extreme temperatures, strong winds, rainstorms, blizzards, earthquakes, landslides, debris flows, and permafrost thawing, and establishing the mapping relationship between each factor and railway infrastructure and operating environment; Safety risk disturbance module construction: integrates equipment failure factors, operation delay factors, and resource capacity gap factors, specifically including signal system failures, abnormal contact network / track geometry parameters, train accidents, line interruptions, and capacity blockages caused by maintenance work. By accessing fault alarm information and operation monitoring data in real time, the risk status is dynamically updated. Demand fluctuation disturbance module construction: integrate freight demand elasticity factors, passenger travel elasticity factors, and regional economic linkage factors, specifically including surges in passenger flow during holidays, sudden changes in the volume of bulk commodities such as Xinjiang coal transportation, and abnormal fluctuations in transportation demand caused by regional economic activities. Using the passenger ticketing system, freight plans, and historical data, a short-term demand forecasting model is established to identify abnormal fluctuation patterns. S302, Multi-source disturbance identification, quantification, and coupling effect analysis, as detailed below: Real-time disturbance identification: Based on the data of natural environmental disturbance, safety risk disturbance and demand fluctuation disturbance in step S301, the real-time monitoring data stream is analyzed using machine learning anomaly detection algorithms to automatically identify the occurrence of the above three-dimensional disturbances. Natural language processing technology is used to parse the text of the emergency and the operation log to automatically extract the type, time, location and description information of the disturbance. Disturbance intensity quantification and coupling: A combined weighting method combining the analytic hierarchy process (AHP) and the entropy weighting method is used to determine the weight of each disturbance factor, construct a comprehensive disturbance intensity index, and perform real-time quantitative assessment of the severity of a single disturbance or multiple disturbances superimposed to obtain the quantified multidimensional disturbance intensity. Establish a perturbation-performance degradation mapping: The quantized multidimensional perturbation intensity is used as input and passed to the physical-service dual-network coupling model. Through a pre-trained machine learning model, the initial performance degradation trajectory prediction of the network performance under the perturbation scenario is initially mapped, providing initial conditions for subsequent accurate resilience assessment.

[0008] Furthermore, step S4 includes the following steps: S401. Initiate the static and dynamic resilience assessment of the coupled network, as detailed below: Update the coupled network state: Based on the identified disturbance type, strength and coupling effect, dynamically update the physical-service dual-network coupling model; specific operations include: adjusting the weights of nodes and edges affected by disturbances in the physical network layer, updating train operation plan constraints in the service network layer, and activating the feedback mechanism between the two networks; Perform static and dynamic resilience assessments: First, by calling the static assessment module, based on the updated physical-service dual-network coupling model, using a hierarchical Bayesian network and an improved PageRank+Shapley value model, the static resilience index in the three-dimensional index system of structure-function-recovery is quickly calculated to assess the theoretical pressure boundary of the network under the current disturbance. Then, by executing the dynamic evaluation module: introducing the time dimension, utilizing the resilience evolution and dynamic evaluation module, combined with the LSTM-Attention temporal prediction model and cellular automata, the dynamic decay process of network performance over time is simulated, generating a preliminary resilience decay curve, and calculating the resilience index: in, This represents the system performance under normal conditions. The time of the disturbance Recovery time; Identify key nodes and their contributions: The key node identification and contribution assessment module analyzes which key nodes or links, under the current disturbance scenario, contribute the most to the overall network resilience degradation due to their failure. Preliminary results of the resilience assessment can be obtained by performing static and dynamic resilience assessment steps: The toughness threshold is determined as follows: The initial performance degradation trajectory prediction obtained in step S302 and the preliminary resilience assessment results obtained in step S401 are input into the judgment module, which determines whether cascading failure has occurred or whether the performance is below the threshold. If so: first trigger step S5 to obtain the cascading failure impact range, critical failure path and risk propagation coefficient; then input the output of step S5 into step S6 to complete risk probability prediction and graded early warning; If "No": This means that the current disturbance is still within the network's resilience range. The process then returns to steps S1 and S2 to continue monitoring the network status and performing routine weight updates, operational status corrections, or lightweight scheduling optimizations without immediately triggering a deep alert.

[0009] Furthermore, step S5 specifically includes the following steps: S501, cascading failure evolution simulation, specifically includes the coordinated operation of the following modules: Cellular Automata and Rule Base: Define the state and evolution rules of nodes / edges, consider the load-capacity model, trigger failure when the node load exceeds the capacity threshold, and redistribute the load to adjacent nodes; LSTM-Attention temporal prediction model: Combines historical cascading failure data to predict the failure propagation range and performance degradation trajectory at multiple future time steps; Graph Neural Networks and Percolation Theory: Extracting the critical diffusion path of the network, identifying the main propagation direction of cascade failure, calculating the risk propagation coefficient of each node / link, and generating a risk propagation heatmap; The above modules work together to output the risk causality of cascading failures, including the scope of impact, critical failure paths, and estimated time of network-wide paralysis. S502. Risk Causal Graph Construction: By integrating historical emergency texts, train operation data, and resilience assessment results, a risk causal graph is constructed using a hierarchical Bayesian network. The graph nodes include various perturbation factors, intermediate events, and final consequences, and the edges represent causal probability relationships. Through Bayesian inference, the probability of occurrence of different risk evolution paths is quantified.

[0010] Furthermore, step S6 specifically includes: GCN+GRU Multi-Spatiotemporal Risk Prediction Model: It integrates graph convolutional networks to capture network topology spatial features and gated cyclic units to capture temporal evolution features. Using cascaded failure simulation results and real-time monitoring data as input, it predicts the probability and risk level of risk occurrence for each node / interval in the next 4-72 hours. Railway Operation Management Risk Layering and Early Warning System: Based on a three-dimensional index of minimum required performance, recovery time, and functional retention rate, risks are divided into four levels: Level I is an exceptionally significant risk: system performance drops below the minimum required performance, and recovery time exceeds 72 hours; Level II is a major risk: system performance drops to less than 30% of normal levels, with a recovery time of 24-72 hours; Level III represents a significant risk: system performance drops to less than 50% of normal levels, with a recovery time of 12-24 hours. Level IV is a general risk: system performance drops to below 80% of normal levels, with a recovery time of less than 12 hours; Step S6 is used to trigger the corresponding level of early warning information, that is, to output the risk stratification early warning result and push it to the dispatch and command center. The early warning content includes: risk type, scope of impact, expected duration, and recommended handling measures.

[0011] Furthermore, step S7 receives the risk stratification warning result output in step S6, and activates the network resilience enhancement mechanism based on the risk level and main failure modes identified by the judgment module, as follows: Risk level and failure mode determination: Based on the risk stratification warning results generated in step S6, the results are input into the judgment module. The judgment module combines the risk causal graph and the output of cascaded failure simulation to determine the risk level and main failure modes under the current disturbance scenario. The judgment results flow to two parallel modules: the resilience enhancement module and the transportation organization optimization module, which perform adaptive adjustments respectively. Network topology optimization specifically involves the coordinated work of the following modules: Load balancing optimization module: Monitors the real-time throughput and saturation of each section and station, dynamically adjusts train flow distribution, and avoids cascading failures caused by local overload; Shortest path optimization module: When the main line is damaged, it quickly calculates the backup path based on real-time topology weights and uses dynamic programming algorithm to achieve smooth path switching; Network redundancy optimization module: Configures physical and service redundancy for critical nodes and vulnerable areas to ensure that local failures do not spread to the entire network; Dynamic adjustment mechanism: Establish a closed loop of perception-decision-execution, update the network status every five minutes, dynamically adjust edge weights and node priorities, and adapt to the characteristics of limited communication and rapidly changing environment in western railways; The study on the cascading propagation mechanism of node failures and strategies to prevent cascading failures specifically includes the collaborative work of the following modules: Node importance analysis module: Utilizes an improved PageRank+Shapley value method, combined with real-time running data, to dynamically sort node importance; Cascade Effect Spatiotemporal Evolution Module: Based on cellular automata and graph neural networks, it simulates the load redistribution process after node failure and predicts the fault propagation range and time window. Anti-cascading failure strategy module: Outputs three types of intervention measures: Active isolation: Implement rate limiting or tripping operations on nodes that are about to be overloaded; Resource pre-positioning: Deploy repair teams and emergency supplies in advance in predicted high-risk areas; Topology reconstruction: Temporarily disconnect high-risk edges and activate redundant paths; Emergency response mechanism: Based on the risk level determined by the judgment module, the system automatically matches the handling procedures in the emergency plan library, sends the optimization results of the resilience enhancement module to the transportation organization optimization module, and initiates coordinated adjustments.

[0012] Furthermore, step S7 also includes transportation organization optimization and collaborative decision-making: Based on enhanced resilience, the transportation organization system is optimized through multiple objectives, and a comprehensive balance among resilience, efficiency, cost, and environment is achieved through a collaborative mechanism. The specific process is as follows: The first step is the restructuring of the transportation organization system, which specifically includes the collaborative work of the following modules: Hub Node Layout and Function Allocation Module: Adjusts the functional positioning of hub stations based on the dynamic rating of node importance output in step S6; Transportation efficiency improvement module: Based on reinforcement learning, temporary operation maps are generated to optimize train stopping schemes and tracking intervals, maximizing throughput under constrained conditions; High-efficiency recovery module: For the recovery phase after disturbance, a genetic algorithm is used to optimize the sorting of emergency repair tasks and the order of resource allocation, shortening the time for system performance to recover to 90%; Cost-effectiveness improvement module: Embed economic constraints in the adjustment plan to avoid resource waste caused by excessive redundancy; Environmental sustainability module: Prioritizes low-carbon scheduling strategies, calculates the changes in carbon emissions resulting from adjustment plans, and supports green operation goals; Secondly, there is a collaborative optimization mechanism, as detailed below: Dual-network collaboration: The repair progress of the physical network is synchronized with the adjustment plan of the service network in real time to avoid the situation where the road is open but the car has not arrived, or the car has arrived but the road is not open. Evaluation-Optimization Closed Loop: The optimization results of the resilience enhancement module and the transportation organization plan are fed back to the resilience evaluation system to recalculate the resilience index R and the resilience decay curve DRC. If the performance does not reach the threshold, the loop is repeated to the judgment module for re-judgment. Continuous learning: The processing of each perturbation event is stored in the case library and used periodically to fine-tune the policy network of the reinforcement learning agent, so as to achieve long-term evolution of the model; Finally, the output and execution are detailed below: Generate an executable instruction set for the dispatch and command center, including train timetable adjustment tables, route switching commands, emergency repair resource dispatch orders, and passenger / freight announcement suggestions; push these instructions to the railway dispatching system (CTC / TDCS) and emergency command platform via interfaces to complete the entire closed loop of assessment → early warning → resilience enhancement → transportation organization optimization → execution feedback.

[0013] Beneficial effects Compared with the prior art, the beneficial effects of the present invention are as follows: This invention is based on a physical-service dual-network coupling model and establishes a two-way closed-loop coupling feedback mechanism to realistically reflect the interaction process of physical anomalies, service adjustments, and load changes. It realizes dynamic coordination between physical network status and service network scheduling, and solves the problem of physical and service separation in traditional modeling methods. As a result, this application has the advantage of modeling that is more in line with reality.

[0014] This invention constructs a three-dimensional evaluation index system of structural resilience, functional resilience, and recovery resilience, and proposes a method for quantifying resilience decay curves to achieve real-time dynamic evaluation of network resilience. This overcomes the limitations of static snapshot-based evaluation and can evaluate the resilience status of railway networks under multiple disturbances in real time, comprehensively, and accurately, thus giving this application the advantage of more comprehensive evaluation.

[0015] This invention integrates the GCN-GRU multi-temporal prediction model with cascaded failure evolution simulation technology, which can predict risks 4-72 hours in advance, accurately identify risk transmission paths, and improve the early warning accuracy compared with traditional methods, thus making this application have the advantage of more accurate and timely early warning.

[0016] This invention forms a closed-loop mechanism covering the entire chain of assessment, early warning, decision-making, and feedback. Based on reinforcement learning, it automatically generates emergency dispatch plans, reducing the response time for handling emergencies by more than 50% and significantly improving the efficiency of emergency decision-making. Thus, this application has the advantage of making decisions more intelligent and efficient. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall workflow of the present invention.

[0018] Figure 2 This is a schematic diagram of the railway data space base of the present invention.

[0019] Figure 3 This is a schematic diagram of the physical-service dual-network coupling modeling and optimization process of the present invention.

[0020] Figure 4 This is a schematic diagram illustrating the resilience assessment and risk warning process of railway coupled networks under multiple disturbance scenarios according to the present invention.

[0021] Figure 5 This is a schematic diagram of the network resilience enhancement and transportation organization optimization process under adaptive conditions according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] It should be noted that the use of terms such as "an embodiment," "embodiment," and "exemplary embodiment" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when describing a specific feature, structure, or characteristic in conjunction with embodiments, implementing such a feature, structure, or characteristic in conjunction with other embodiments should be within the knowledge of those skilled in the art.

[0024] This invention provides a method for dynamic resilience assessment and risk early warning of railway coupled networks under multiple disturbance scenarios, comprising the following steps: S1. Multimodal data acquisition and preprocessing: Collect multi-source data such as railway infrastructure status, train operation, environmental weather, and passenger and freight demand. Clean, denoise, interpolate, and lightweight process the collected multi-source data to construct a unified spatiotemporal benchmark for railway data spatial foundation.

[0025] Step S1 aims to construct a multimodal data intelligent sensing and data space foundation for western railways. The specific process is as follows: Figure 2 As shown, it mainly includes four sub-steps: multi-source data acquisition, data preprocessing, multi-modal data fusion and foundation construction, dynamic perception and feature extraction.

[0026] The multi-source data collection is as follows: The following multimodal data were collected by multi-source sensor terminals deployed along the railway line and incorporated into a dynamic physical database: Infrastructure data: track geometry parameters, bridge and tunnel status, power supply equipment status, signaling equipment status, etc. Train operation data: train number, location, speed, delay status, train formation information, etc.; Environmental meteorological data: temperature, humidity, wind speed, rainfall, snow depth, geological disaster early warning, etc. Passenger and freight demand data: passenger flow, freight volume, transportation plans, holiday information, etc.

[0027] Among them, data preprocessing is completed by cleaning, denoising, and lightweighting the collected data before it enters the base. The collected multi-source data underwent the following preprocessing to ensure data quality and computational efficiency: Infrastructure data preprocessing: For time-series data of track geometry parameters (such as gauge, level, elevation, etc.), wavelet denoising and moving average methods are used to eliminate high-frequency noise and instantaneous interference; for discrete state quantities of bridges, tunnels, and power supply equipment, autoencoders and generative adversarial networks (GANs) are used to identify and remove outliers caused by sensor failures or communication errors; ARIMA models are used to perform trend prediction and anomaly identification on infrastructure monitoring data, and interpolation is used to repair missing or abrupt data.

[0028] Train operation data preprocessing: Wavelet denoising and moving average methods are used to smooth time series data such as speed and delay time, and LSTM-Autoencoder reconstruction error detection mechanism is used to correct abnormal operation records.

[0029] Environmental meteorological data preprocessing: Outliers are removed by autoencoders and GANs, and missing data are repaired by spatial interpolation (such as Kriging) or temporal interpolation (linear interpolation).

[0030] Preprocessing of passenger and freight demand data: Use interpolation algorithms (such as cubic spline interpolation) to repair missing data, and introduce LSTM-Autoencoder to dynamically correct abnormal values ​​of transportation volume caused by extreme weather or sudden events.

[0031] Lightweight processing: In view of the current situation of sparse sensors and limited communication bandwidth in the western region, techniques such as model compression, quantization pruning, and low-rank decomposition are applied to perform lightweight processing on high-frequency dynamic monitoring data and high-dimensional static infrastructure data, thereby reducing transmission and storage costs.

[0032] Among them, multimodal data fusion and foundation construction are composed of Figure 2 The data fusion and foundation construction layer is now complete. This layer is responsible for fusing preprocessed multi-source heterogeneous data into a unified spatiotemporal data foundation, providing support for subsequent dynamic sensing, as detailed below: A framework based on the fusion of complex network theory and hypergraph model is constructed: Using stations as nodes and sections as edges, a hypergraph representation of the railway physical topology is constructed. The hyperedges can connect multiple nodes simultaneously (such as multiple stations on a line) or describe the relationship between multimodal data (such as track status + weather conditions + train operation parameters of a certain section). Spatiotemporal network modeling: Through the spatiotemporal network modeling module, a cross-modal spatiotemporal alignment framework of track geometry, environmental monitoring and train operation is constructed. Combining Diffusion Models and Spatiotemporal Graph Neural Network (STGNN), heterogeneous fusion of vector data (train position), raster imagery (satellite cloud image), 3D model (terrain) and streaming data (real-time sensor) is achieved to generate a spatiotemporal topology model. This spatiotemporal topology model can capture the deep coupling relationship between stations and sections, as well as the evolution law of operating status in time and space.

[0033] Elastic Virtual Topology Network Generation: Based on the fused data foundation, an elastic virtual topology network is constructed. This elastic virtual topology network can dynamically update the weights of nodes and edges according to real-time data (such as increasing edge impedance when the track condition deteriorates), providing a flexible and evolvable basic topology for dual-network coupling.

[0034] Semantic information fusion: Using the BERT model to perform semantic mining on operational scheduling text data (such as scheduling commands and fault reports), key information such as event type and scope of impact is extracted and associated with spatiotemporal topology models to enhance the semantic expressive power of the data base.

[0035] Dynamic Sensing and Feedback Layer: This layer, based on the established data foundation, senses changes in the environment and operational status in real time, and feeds back to the elastic virtual topology network through weight updates, as detailed below: Multi-dimensional dynamic perception module: Weather change dynamic perception module: Real-time access to meteorological monitoring and forecast data, identification of extreme weather events such as strong winds, rainstorms, blizzards, and permafrost thawing, and quantification of their impact on the line's traffic capacity; Equipment Fault Dynamic Sensing Module: Monitors the status of equipment such as signals, overhead contact lines, and tracks, identifies fault types and levels, and predicts equipment degradation trends; Passenger demand fluctuation response module: Analyzes ticketing system and passenger flow forecast data to identify sudden changes in demand such as the Spring Festival travel rush and freight peak season, and predicts transportation pressure; Emergency Response Module: Parses emergency texts (such as accident reports and social events) through natural language processing to extract disturbance features and affected areas.

[0036] Virtual arc weight update and feedback mechanism: Virtual arc weight update module: Based on the output of the above four perception modules, dynamically adjust the weight parameters of the edges in the elastic virtual topology network (such as running speed, throughput, and delay probability). For example: the weight of strong wind sections is increased (reducing the permissible speed), the weight of equipment failure sections is set to infinity (temporary blockage), and the weight of adjacent edges of stations is reduced when passenger flow surges (reflecting congestion).

[0037] The updated virtual topology network serves as the input for step S2 (physical-service dual-network coupling modeling), realizing a closed loop of data acquisition → fusion modeling → dynamic perception → weight feedback.

[0038] S2. Physical-Service Dual-Network Coupling Modeling: Construct railway physical network layer and service network layer respectively, establish cross-layer mapping relationship and state synchronization mechanism between nodes and edges, and form a dynamically evolving physical-service dual-network coupling model.

[0039] Step S2 aims to establish a two-layer coupling model of the railway physical network and service network (physical-service dual-network coupling model) to achieve full-process collaboration of perception-mapping-feedback-optimization. The establishment process of the physical-service dual-network coupling model mainly includes: dual-network infrastructure construction and mapping, dynamic coupling optimization, lightweight and timeliness judgment, and continuous evolution mechanism, specifically including the following steps: S201. Dual-network infrastructure construction and cross-layer mapping: First, with infrastructure such as stations, sections, maintenance depots, and locomotive depots as nodes and line connections as edges, real-time monitoring data (including but not limited to track status, power supply load, environmental conditions, etc.) are accessed to construct a dynamically evolving infrastructure operation map and complete the construction of the railway physical network layer. Then, using train number, operation plan, and transportation service as nodes and train operation path as edges, a task-driven virtual network is constructed. The service network topology is dynamically adjusted according to fluctuations in transportation demand and event response. Hypergraph theory is introduced to construct reconfigurable service units, and the service network layer is constructed. Secondly, establish a many-to-many cross-layer mapping relationship between physical nodes, physical edges and service nodes, and service edges to form a cross-layer association matrix or mapping function (e.g., a physical node of a station corresponds to several service nodes for train services).

[0040] Among them, the railway physical network layer transforms infrastructure status (such as track speed limits and power outages) into service network constraints (such as train path unavailability and reduced throughput) through state-service mapping. The service network layer generates virtual topology (such as detour paths and train cancellations / mergers) through an optimization engine, which in turn drives the resource scheduling and channel allocation of the physical network; thus completing the establishment of the mapping and feedback mechanism. Finally, a two-way closed loop is achieved through a feedback mechanism: changes in the state of the railway physical network layer trigger adjustments in the service network layer, and the adjustment results of the service network layer are fed back to the railway physical network layer to update the resource occupancy status. At this point, the construction of the physical-service dual-network coupling model is completed.

[0041] S202. After the mapping and feedback mechanism is established, the dynamic coupling optimization loop begins, which specifically includes the collaborative work of the following modules: Data-driven module: Reads real-time data (including but not limited to train location, equipment status, weather warnings, passenger and freight demand, etc.) from step S1, performs lightweight data processing, and then inputs the data into the dynamic coupling module; Dynamic Coupling Module: Integrates the real-time status of the railway physical network layer and service network layer, calculates the system performance indicators under the current dual-network coupling (such as section throughput capacity, train punctuality rate, hub transfer efficiency), and identifies key aspects that need optimization (such as excessive load in a certain section or track occupancy conflicts at a certain station). Blocking edge removal module: When a physical interval is detected to be completely unavailable due to disaster, failure or construction, the edge is temporarily removed from the service network virtual topology; for edges with partially degraded capabilities, the increase in their passage cost is reflected by weight adjustment (rather than complete removal); Node Priority Adjustment Module: Based on the node importance output by the dynamic coupling module (such as hub stations and strategic material transfer stations), dynamically adjust the service priority of nodes in the service network to ensure that key node resources are given priority. Intelligent dispatching module: Based on reinforcement learning or genetic algorithms, it generates the optimal train route redistribution and capacity resource rescheduling scheme, and executes adjustments (such as changing train stops and adjusting train formations). Infrastructure upgrade module: For physical sections that repeatedly encounter bottlenecks or high risks, output infrastructure upgrade suggestions (such as adding refuge lines, reinforcing slopes, and deploying redundant sensors), and provide long-term feedback to the planning department; The feedback mechanism operates continuously: After each operation such as removing blocking edges, adjusting node priorities, intelligent allocation, and upgrading infrastructure is completed, the new network state is transmitted to the data-driven module again through the feedback mechanism, forming a closed-loop iteration. S203, Continuous Optimization Module: The continuous optimization module records the response process and effect of the physical-service dual-network coupling model under each disturbance event, forming a case library. It periodically (e.g., monthly) uses historical data to retrain and fine-tune the parameters (such as load threshold and priority weight) of the dynamic coupling module, and pushes the optimized parameters to the dynamic coupling module and the node priority adjustment module to realize the continuous evolution of the physical-service dual-network coupling model. S204, Lightweight and Timeliness Judgment: After completing the optimization loop described in S203, proceed to the judgment module: determine whether the data meets the requirements of lightweight and timeliness; If the following conditions are met, then output the current physical-service dual-network coupling model and proceed to step S3; If the requirements are not met, the process returns to the data-driven module to re-filter, compress, or prioritize data transmission until the requirements are met.

[0042] This invention is based on a physical-service dual-network coupling model and establishes a two-way closed-loop coupling feedback mechanism to realistically reflect the interaction process of physical anomalies, service adjustments, and load changes. It realizes dynamic coordination between physical network status and service network scheduling, and solves the problem of physical and service separation in traditional modeling methods. As a result, this application has the advantage of modeling that is more in line with reality.

[0043] S3. Multiple Disturbance Identification and Quantification: Construct a three-dimensional disturbance index system of natural environment, security risk and demand fluctuation to identify different types of disturbances and quantify their intensity and impact range.

[0044] Step S3 includes the following steps: S301. Construct a multi-dimensional disturbance index system: First, to comprehensively cover the complex risk sources faced by western railways, a three-dimensional disturbance index system of natural environment, safety risk, and demand fluctuation is constructed to provide a classification framework for subsequent disturbance identification and quantification, as detailed below: Natural environment disturbance module construction: Integrate climate change factors, natural disaster factors, and environmental change factors, specifically including extreme temperatures, strong winds, rainstorms, blizzards, earthquakes, landslides, debris flows, and permafrost thawing and settlement, and establish mapping relationships between each factor and railway infrastructure (tracks, bridges, catenary) and operating environment (such as gale warnings and rainfall warning values); Safety risk disturbance module construction: integrates equipment failure factors, operation delay factors, and resource capacity gap factors, specifically including signal system failures, abnormal contact network / track geometry parameters, train accidents, line interruptions, and capacity blockages caused by maintenance work. By accessing fault alarm information and operation monitoring data in real time, the risk status is dynamically updated. Demand fluctuation disturbance module construction: Integrating freight demand elasticity factors, passenger travel elasticity factors, and regional economic linkage factors, specifically including surges in passenger flow during holidays, sudden changes in the volume of bulk commodities such as Xinjiang coal transportation, and abnormal fluctuations in transportation demand caused by regional economic activities. Using the passenger ticketing system, freight plans, and historical data, a short-term demand forecasting model is established to identify abnormal fluctuation patterns.

[0045] S302, Multi-source disturbance identification, quantification, and coupling effect analysis, as detailed below: Real-time disturbance identification: Based on the data sources of natural environmental disturbances, safety risk disturbances, and demand fluctuation disturbances in step S301 (such as natural environmental data, demand fluctuation data, emergency event text, and train operation data), machine learning anomaly detection algorithms (such as Isolation Forest and LSTM-Autoencoder) are used to analyze the real-time monitoring data stream, automatically identify the occurrence of the above three-dimensional disturbances, and use natural language processing (NLP) technology to parse emergency event text and operation logs to automatically extract the type, time, location, and description information of the disturbance; Disturbance intensity quantification and coupling: A combined weighting method combining the analytic hierarchy process (AHP) and the entropy weighting method is used to determine the weight of each disturbance factor and construct the Disturbance Intensity Index (DSI). The severity of a single disturbance or multiple disturbances is quantitatively assessed in real time to obtain the quantified multidimensional disturbance intensity, including disturbance type, intensity and coupling effect. The above steps enable the analysis of the coupling effects between different disturbances. For example, when strong winds and equipment failures occur simultaneously, or when freight peaks coincide with blizzards, the combined impact on network performance degradation is far greater than the sum of individual disturbances. By establishing a disturbance coupling matrix, multi-dimensional disturbance scenarios can be provided for subsequent input coupling network models.

[0046] Establish a perturbation-performance degradation mapping: The quantized multidimensional perturbation intensity is used as input and passed to the physical-service dual-network coupling model. Through a pre-trained machine learning model (such as XGBoost or neural network), the initial performance degradation trajectory prediction of the network performance (P(t)) under the perturbation scenario is initially mapped, providing initial conditions for subsequent accurate resilience assessment.

[0047] S4. Network Resilience Assessment: Based on the coupled network model, a dynamic assessment index system is constructed from three dimensions: structural resilience, functional resilience, and recovery resilience. A method for quantifying resilience decay curves is proposed to assess the network resilience status in real time.

[0048] Step S4 includes the following steps: S401. Initiate the static and dynamic resilience assessment of the coupled network, as detailed below: Update the coupled network state: Based on the identified disturbance type, intensity, and coupling effect, dynamically update the physical-service dual-network coupling model; specific operations include: adjusting the weights of nodes and edges affected by disturbances in the physical network layer (e.g., reducing the throughput capacity of windy sections), updating train operation plan constraints in the service network layer (e.g., speed limits, service suspensions), and activating the feedback mechanism between the two networks (e.g., initiating detour path search). Perform static and dynamic resilience assessments: First, by calling the static assessment module, based on the updated physical-service dual-network coupling model, using a hierarchical Bayesian network and an improved PageRank+Shapley value model, the static resilience index in the three-dimensional index system of structure-function-recovery is quickly calculated to assess the theoretical pressure boundary of the network under the current disturbance. Then, by executing the dynamic evaluation module: introducing the time dimension, utilizing the resilience evolution and dynamic evaluation module, combined with the LSTM-Attention temporal prediction model and cellular automata, the dynamic decay process of network performance over time is simulated, generating a preliminary resilience decay curve, and calculating the resilience index: in, This represents the system performance under normal conditions. The time of the disturbance Recovery time; Identify key nodes and their contributions: The key node identification and contribution assessment module analyzes which key nodes or links (such as hub stations or single-line sections) have the highest contribution to the overall network resilience decline under the current disturbance scenario. Preliminary results of the resilience assessment can be obtained by performing static and dynamic resilience assessment steps.

[0049] The toughness threshold is determined as follows: The initial performance degradation trajectory prediction obtained in step S302 and the preliminary resilience assessment results obtained in step S401 are input into the judgment module, which determines whether cascading failure has occurred or whether the performance is below the threshold. If so: first trigger step S5 to obtain the cascading failure impact range, critical failure path and risk propagation coefficient; then input the output of step S5 into step S6 to complete risk probability prediction and graded early warning; If "No": This means that the current disturbance is still within the network's resilience range. The process then returns to steps S1 and S2 to continue monitoring the network status and performing routine weight updates, operational status corrections, or lightweight scheduling optimizations without immediately triggering a deep alert.

[0050] This invention constructs a three-dimensional evaluation index system of structural resilience, functional resilience, and recovery resilience, and proposes a method for quantifying resilience decay curves to achieve real-time dynamic evaluation of network resilience. This overcomes the limitations of static snapshot-based evaluation and can evaluate the resilience status of railway networks under multiple disturbances in real time, comprehensively, and accurately, thus giving this application the advantage of more comprehensive evaluation.

[0051] S5. Cascade Failure Evolution Simulation: Based on cellular automata and graph neural networks, this simulation simulates the cascade propagation process of node failure - line blockage - regional paralysis caused by disturbances, and identifies key vulnerabilities and risk transmission paths.

[0052] When the judgment module outputs "Yes" in step S401, the cascaded failure evolution simulation, risk causal graph construction, and multi-temporal risk early warning process are initiated. Step S5 specifically includes the following steps: S501, cascading failure evolution simulation, specifically includes the coordinated operation of the following modules: Cellular Automata and Rule Base: Define the state (normal, overload, failure, recovery) and evolution rules of nodes / edges, consider the load-capacity model, trigger failure when the node load exceeds the capacity threshold, and redistribute the load to adjacent nodes; LSTM-Attention temporal prediction model: Combines historical cascading failure data to predict the failure propagation range and performance degradation trajectory at multiple future time steps; Graph Neural Networks and Percolation Theory: Extracting the critical diffusion path of the network, identifying the main propagation direction of cascade failure, calculating the risk propagation coefficient of each node / link, and generating a risk propagation heatmap; The above modules work together to output the risk causation of cascading failures, which includes the scope of impact of cascading failures, critical failure paths, and the expected time of network-wide paralysis.

[0053] S502. Risk Causal Graph Construction: By integrating historical emergency texts, train operation data, and resilience assessment results, a risk causal graph is constructed using a hierarchical Bayesian network. The graph nodes include various disturbance factors, intermediate events (such as node overload and section congestion), and final consequences (widespread delays and service disruptions). The edges represent causal probability relationships. Through Bayesian inference, the probability of different risk evolution paths is quantified.

[0054] This invention integrates cellular automata and graph neural networks to accurately simulate the cascading failure process caused by disturbances, identify key vulnerabilities and risk transmission paths, and provide targeted objectives for risk prevention and control.

[0055] S6. Risk warning and classification: Integrate historical data and real-time monitoring information to construct the GCN-GRU risk prediction model, and conduct risk classification and warning based on the three-dimensional indicators of minimum required performance, recovery timeliness and function retention rate. Step S6 specifically involves: GCN+GRU Multi-Spatiotemporal Risk Prediction Model: It integrates Graph Convolutional Network (GCN) to capture network topology spatial features and Gated Recurrent Unit (GRU) to capture temporal evolution features. Using cascade failure simulation results and real-time monitoring data as input, it predicts the probability and level of risk occurrence of each node / interval in the next 4-72 hours. Railway Operation Management Risk Layering and Early Warning System: Based on a three-dimensional index of minimum required performance, recovery time, and functional retention rate, risks are divided into four levels: Level I is an exceptionally significant risk: system performance drops below the minimum required performance, and recovery time exceeds 72 hours; Level II is a major risk: system performance drops to less than 30% of normal levels, with a recovery time of 24-72 hours; Level III represents a significant risk: system performance drops to less than 50% of normal levels, with a recovery time of 12-24 hours. Level IV is a general risk: system performance drops to below 80% of normal levels, with a recovery time of less than 12 hours; Step S6 is used to trigger the corresponding level of early warning information, that is, to output the risk stratification early warning result and push it to the dispatch and command center. The early warning content includes: risk type, scope of impact, expected duration, and suggested handling measures (train adjustment, emergency repair resource scheduling, etc.).

[0056] This invention combines the advantages of graph convolutional networks and gated recurrent units, while simultaneously capturing the spatial and temporal evolution characteristics of the network topology, to achieve accurate risk prediction across multiple spatiotemporal scales, significantly improving the accuracy and timeliness of early warnings.

[0057] This invention integrates the GCN-GRU multi-temporal prediction model with cascaded failure evolution simulation technology, which can predict risks 4-72 hours in advance, accurately identify risk transmission paths, and improve the early warning accuracy compared with traditional methods, thus making this application have the advantage of more accurate and timely early warning.

[0058] S7. Emergency Decision Support: Based on reinforcement learning, the optimal emergency dispatch and recovery strategies are generated, forming a closed-loop mechanism of risk warning, dynamic adjustment and efficient recovery.

[0059] Step S7 receives the risk stratification warning result output in step S6, and activates the network resilience enhancement mechanism based on the risk level and main failure modes identified by the judgment module. This mechanism mainly includes: risk level judgment, resilience enhancement module, and internal optimization strategies, as detailed below: Risk level and failure mode judgment: Based on the risk stratification warning results (levels I to IV) generated in step S6, the results are input into the judgment module. The judgment module combines the risk causal graph and the output of cascading failure simulation to determine the risk level (particularly serious / serious / relatively serious / general) and the main failure modes (such as node overload, section blockage, hub paralysis, etc.) under the current disturbance scenario. The judgment results flow to two parallel modules: the resilience enhancement module and the transportation organization optimization module, which perform adaptive adjustments respectively.

[0060] Network topology optimization specifically involves the coordinated work of the following modules: Load balancing optimization module: Monitors the real-time throughput and saturation of each section and station, dynamically adjusts train flow distribution, and avoids cascading failures caused by local overload.

[0061] Shortest path optimization module: When the main line is damaged, it quickly calculates the backup path based on real-time topology weights (such as speed limit, blockage) and uses dynamic programming algorithm to achieve smooth path switching.

[0062] Network redundancy optimization module: Configure physical redundancy (backup switches, emergency power supply) and service redundancy (backup trains, virtual routes) for key nodes (hub stations, strategic material transfer stations) and vulnerable sections to ensure that local failures do not spread to the entire network.

[0063] Dynamic adjustment mechanism: Establish a closed loop of perception-decision-execution, update the network status every five minutes, dynamically adjust edge weights and node priorities, and adapt to the characteristics of limited communication and rapidly changing environment in western railways; The study on the cascading propagation mechanism of node failures and strategies to prevent cascading failures specifically includes the collaborative work of the following modules: Node importance analysis module: Utilizes an improved PageRank+Shapley value method, combined with real-time operational data (late arrival rate, load rate), to dynamically rank node importance; Cascade Effect Spatiotemporal Evolution Module: Based on cellular automata and graph neural networks, it simulates the load redistribution process after node failure and predicts the fault propagation range and time window. Anti-cascading failure strategy module: Outputs three types of intervention measures: Active isolation: Implement rate limiting or tripping operations on nodes that are about to be overloaded; Resource pre-positioning: Deploy repair teams and emergency supplies in advance in predicted high-risk areas; Topology reconstruction: Temporarily disconnect high-risk edges and activate redundant paths; Emergency response mechanism: Based on the risk level determined by the judgment module, the emergency response mechanism automatically matches the handling procedures in the emergency plan library (such as train suspension scope, passenger evacuation and emergency resource scheduling), and sends the optimization results of the resilience enhancement module (such as new topology and load distribution scheme) to the transportation organization optimization module to initiate linkage adjustment.

[0064] Step S7 further includes transportation organization optimization and collaborative decision-making: Based on enhanced resilience, this step optimizes the transportation organization system through multiple objectives and achieves a comprehensive balance between resilience, efficiency, cost, and environment through a collaborative mechanism. The specific process is as follows: The first step is the restructuring of the transportation organization system, which specifically includes the collaborative work of the following modules: Hub Node Layout and Function Allocation Module: Based on the dynamic rating of node importance output in step S6, adjust the functional positioning of the hub station (such as whether to accept diversion trains or whether to open emergency tracks). Transportation efficiency improvement module: Based on reinforcement learning, temporary operation maps are generated to optimize train stopping schemes and tracking intervals, maximizing throughput under constrained conditions; High-efficiency recovery module: For the recovery phase after disturbance, a genetic algorithm is used to optimize the sorting of emergency repair tasks and the order of resource allocation, shortening the time for system performance to recover to 90%; Cost-effectiveness improvement module: embed economic constraints (such as additional fuel costs and delay compensation) into the adjustment plan to avoid resource waste caused by excessive redundancy; Environmental sustainability module: Prioritizes low-carbon scheduling strategies (such as reducing waiting and avoiding, and reducing ineffective start-stop), calculates the carbon emission changes brought about by the adjustment plan, and supports green operation goals; Secondly, there is a collaborative optimization mechanism, as detailed below: Dual-network coordination: The repair progress of the physical network (track, power supply) is synchronized with the adjustment plan of the service network (train route, car body route) in real time to avoid the situation where the road is open but the train has not arrived or the train has arrived but the road is not open; Evaluation-Optimization Closed Loop: The optimization results of the toughness enhancement module in step S704 and the transportation organization scheme are fed back to the toughness evaluation system, and the toughness index R and toughness decay curve DRC are recalculated. If the performance does not reach the threshold (e.g., P(t) < 0.5P(t0)), the loop is repeated to the judgment module for re-judgment. Continuous learning: The process of handling each disturbance event (including judgment by the judgment module, resilience enhancement actions, and organizational adjustment plans) is stored in the case library and used periodically to fine-tune the policy network of the reinforcement learning agent, so as to achieve the long-term evolution of the model. Finally, the output and execution are detailed below: Generate an executable instruction set for the dispatch and command center, including train timetable adjustment tables, route switching commands, emergency repair resource dispatch orders, and passenger / freight announcement suggestions; push these instructions to the railway dispatching system (CTC / TDCS) and emergency command platform via interfaces to complete the entire closed loop of assessment → early warning → resilience enhancement → transportation organization optimization → execution feedback.

[0065] In response to the characteristics of sparse sensors, numerous communication blind spots, harsh environment, and low network redundancy in western railways, this application optimizes lightweight data processing and low-redundancy network resilience enhancement algorithms to ensure the stable operation of the national strategic transportation corridor, thus giving it the advantage of being more adaptable to western regions. This invention organically combines resilience assessment, risk warning, and emergency decision-making, and generates adaptive emergency strategies based on reinforcement learning, forming a complete closed-loop management system to enhance the proactive defense and rapid recovery capabilities of the railway network.

[0066] This invention forms a closed-loop mechanism covering the entire chain of assessment, early warning, decision-making, and feedback. Based on reinforcement learning, it automatically generates emergency dispatch plans, reducing the response time for handling emergencies by more than 50% and significantly improving the efficiency of emergency decision-making. Thus, this application has the advantage of making decisions more intelligent and efficient.

[0067] A specific application embodiment of the present invention is as follows: This embodiment uses a section of the Lanzhou-Xinjiang Railway as an application object to verify the effectiveness of the method of the present invention: Data collection and preprocessing: Data on infrastructure status, train operation, meteorological environment, passenger and freight demand, etc. for this section from 2023 to 2025 were collected, totaling about 5TB; after preprocessing, the data accuracy was improved to 98.5%, and the data volume was reduced to 30% of the original.

[0068] Dual-network coupling modeling: Construct a physical network layer containing 23 stations and 32 sections, and a service network layer containing 128 pairs of trains. Establish a dual-network mapping relationship and coupling mechanism. The model can reflect network status changes in real time, with an update frequency of 5 minutes / time.

[0069] Multiple disturbance identification and quantification: 12 strong winds, 8 rainstorms, 3 equipment failures and 5 passenger flow peaks were successfully identified in this section, and the quantification results of disturbance intensity matched the actual situation by 92%.

[0070] Resilience dynamic assessment: The network resilience status of this segment is assessed in real time. The resilience index is calculated to be 0.87 under normal conditions, 0.63 under strong winds, and 0.51 under equipment failure. Three critical vulnerable nodes and two high-risk links are identified.

[0071] Cascade Failure Simulation: The cascade failure process after the failure of a critical station was simulated. The prediction results show that if no intervention measures are taken, the fault will spread to 5 surrounding stations within 6 hours, resulting in a 70% decrease in the section's transportation capacity.

[0072] Risk warning: Based on the GCN-GRU model, the risk prediction for the next 24 hours is carried out with an accuracy rate of 89% and an average advance warning time of 4.2 hours.

[0073] Emergency decision support: For simulated station failure scenarios, the system generates the optimal emergency dispatch plan, including train detours, capacity adjustments, and allocation of repair resources. After implementing the plan, the recovery time was reduced from the estimated 36 hours to 18 hours, and transportation losses were reduced by 65%.

[0074] As can be seen from this embodiment, the present invention can be widely applied to railway bureaus and groups across the country, and is particularly suitable for railway networks in complex environments in the west. By adopting the method of the present invention, the anti-interference capability and emergency response level of the railway network can be significantly improved, the transportation losses caused by emergencies can be reduced, and the safety and smooth flow of national strategic transportation channels can be guaranteed.

[0075] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Any modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for dynamic assessment of the resilience and risk early warning of railway coupled networks under multiple disturbance scenarios, characterized in that, Includes the following steps: S1. Multimodal data acquisition and preprocessing: Collect multi-source data such as railway infrastructure status, train operation, environmental weather, and passenger and freight demand. Clean, denoise, interpolate, and lightweight process the collected multi-source data to construct a unified spatiotemporal benchmark for railway data spatial foundation. S2. Physical-Service Dual-Network Coupling Modeling: Construct the railway physical network layer and service network layer respectively, establish cross-layer mapping relationship and state synchronization mechanism between nodes and edges, and form a dynamically evolving physical-service dual-network coupling model; S3. Multiple Disturbance Identification and Quantification: Construct a three-dimensional disturbance index system of natural environment, security risk and demand fluctuation to identify different types of disturbances and quantify their intensity and impact range; S4. Network resilience assessment: Based on the coupled network model, a dynamic assessment index system is constructed from three dimensions: structural resilience, functional resilience, and recovery resilience. A method for quantifying resilience decay curves is proposed to assess the network resilience status in real time. S5. Cascade Failure Evolution Simulation: Based on cellular automata and graph neural networks, simulate the cascade propagation process of node failure - line blockage - regional paralysis caused by disturbance, and identify key vulnerabilities and risk transmission paths. S6. Risk warning and classification: Integrate historical data and real-time monitoring information to construct the GCN-GRU risk prediction model, and conduct risk classification and warning based on the three-dimensional indicators of minimum required performance, recovery timeliness and function retention rate. S7. Emergency Decision Support: Based on reinforcement learning, the optimal emergency dispatch and recovery strategies are generated, forming a closed-loop mechanism of risk warning, dynamic adjustment and efficient recovery.

2. The method for dynamic assessment and risk warning of the resilience of railway coupled networks under multiple disturbance scenarios according to claim 1, characterized in that, Step S2 includes the following steps: S201. Dual-network infrastructure construction and cross-layer mapping: First, taking infrastructure such as stations, sections, maintenance depots, and locomotive depots as nodes and line connections as edges, real-time monitoring data is accessed to construct a dynamically evolving infrastructure operation map and complete the construction of the railway physical network layer. Then, using train number, operation plan, and transportation service as nodes and train operation path as edges, a task-driven virtual network is constructed. The service network topology is dynamically adjusted according to fluctuations in transportation demand and event response. Hypergraph theory is introduced to construct reconfigurable service units, and the service network layer is constructed. Secondly, establish a many-to-many cross-layer mapping relationship between physical nodes, physical edges and service nodes, and service edges to form a cross-layer association matrix or mapping function: Among them, the railway physical network layer transforms the infrastructure state into the service network constraint conditions through state-service mapping, and the service network layer generates virtual topology through optimization engine to drive the resource scheduling and channel allocation of the physical network in reverse. Finally, a two-way closed loop is achieved through a feedback mechanism: changes in the state of the railway physical network layer trigger adjustments in the service network layer, and the adjustment results of the service network layer are fed back to the railway physical network layer to update the resource occupancy status. At this point, the construction of the physical-service dual-network coupling model is completed. S202, Dynamic Coupling and Optimization Loop, specifically includes the collaborative work of the following modules: Data-driven module: Reads real-time data from step S1, performs lightweight data processing, and then inputs the data into the dynamic coupling module; Dynamic Coupling Module: Integrates the real-time status of the railway physical network layer and service network layer, calculates the system performance indicators under the current dual-network coupling, and identifies key aspects that need optimization; Blocking edge removal module: When a physical interval is detected to be completely unavailable due to disaster, failure or construction, the edge is temporarily removed from the service network virtual topology; for edges with reduced capacity, the weight is adjusted to reflect the increase in their passage cost. Node Priority Adjustment Module: Based on the node importance output by the dynamic coupling module, dynamically adjust the service priority of nodes in the service network to ensure that critical node resources are given priority. Intelligent dispatching module: Based on reinforcement learning or genetic algorithms, it generates the optimal train route reallocation and capacity resource rescheduling scheme and executes the adjustment; Infrastructure upgrade module: For physical sections that repeatedly encounter bottlenecks or high risks, output infrastructure upgrade suggestions and provide long-term feedback to the planning department; The feedback mechanism operates continuously: After each operation such as removing blocking edges, adjusting node priorities, intelligent allocation, and upgrading infrastructure is completed, the new network state is transmitted to the data-driven module again through the feedback mechanism, forming a closed-loop iteration. S203, Continuous Optimization Module: The continuous optimization module records the response process and effect of the physical-service dual-network coupling model under each disturbance event, forming a case library. Historical data is used periodically to retrain and fine-tune the parameters of the dynamic coupling module. The optimized parameters are pushed to the dynamic coupling module and the node priority adjustment module to realize the continuous evolution of the physical-service dual-network coupling model. S204, Lightweight and Timeliness Judgment: After completing the optimization loop described in S202, enter the judgment module: determine whether the data meets the requirements of lightweight and timeliness; If the following conditions are met, then output the current physical-service dual-network coupling model and proceed to step S3; If the requirements are not met, the process returns to the data-driven module to re-filter, compress, or prioritize data transmission until the requirements are met.

3. The method for dynamic assessment and risk warning of the resilience of railway coupled networks under multiple disturbance scenarios according to claim 1, characterized in that, Step S3 includes the following steps: S301. Construct a multi-dimensional disturbance index system: First, to comprehensively cover the complex risk sources faced by railways, a three-dimensional disturbance index system of natural environment, safety risk, and demand fluctuation is constructed to provide a classification framework for subsequent disturbance identification and quantification, as detailed below: Natural environment disturbance module construction: Integrating climate change factors, natural disaster factors, and environmental change factors, specifically including extreme temperatures, strong winds, rainstorms, blizzards, earthquakes, landslides, debris flows, and permafrost thawing, and establishing the mapping relationship between each factor and railway infrastructure and operating environment; Safety risk disturbance module construction: integrates equipment failure factors, operation delay factors, and resource capacity gap factors, specifically including signal system failures, abnormal contact network / track geometry parameters, train accidents, line interruptions, and capacity blockages caused by maintenance work. By accessing fault alarm information and operation monitoring data in real time, the risk status is dynamically updated. Demand fluctuation disturbance module construction: integrate freight demand elasticity factors, passenger travel elasticity factors, and regional economic linkage factors, specifically including surges in passenger flow during holidays, sudden changes in the volume of bulk commodities such as Xinjiang coal transportation, and abnormal fluctuations in transportation demand caused by regional economic activities. Using the passenger ticketing system, freight plans, and historical data, a short-term demand forecasting model is established to identify abnormal fluctuation patterns. S302, Multi-source disturbance identification, quantification, and coupling effect analysis, as detailed below: Real-time disturbance identification: Based on the data of natural environmental disturbance, safety risk disturbance and demand fluctuation disturbance in step S301, the real-time monitoring data stream is analyzed using machine learning anomaly detection algorithms to automatically identify the occurrence of the above three-dimensional disturbances. Natural language processing technology is used to parse the text of the emergency and the operation log to automatically extract the type, time, location and description information of the disturbance. Disturbance intensity quantification and coupling: A combined weighting method combining the analytic hierarchy process (AHP) and the entropy weighting method is used to determine the weight of each disturbance factor, construct a comprehensive disturbance intensity index, and perform real-time quantitative assessment of the severity of a single disturbance or multiple disturbances superimposed to obtain the quantified multidimensional disturbance intensity. Establish a perturbation-performance degradation mapping: The quantized multidimensional perturbation intensity is used as input and passed to the physical-service dual-network coupling model. Through a pre-trained machine learning model, the initial performance degradation trajectory prediction of the network performance under the perturbation scenario is initially mapped, providing initial conditions for subsequent accurate resilience assessment.

4. The method for dynamic assessment and risk warning of the resilience of a railway coupled network under multiple disturbance scenarios according to claim 3, characterized in that, Step S4 includes the following steps: S401. Initiate the static and dynamic resilience assessment of the coupled network, as detailed below: Update the coupled network state: Based on the identified disturbance type, strength and coupling effect, dynamically update the physical-service dual-network coupling model; specific operations include: adjusting the weights of nodes and edges affected by disturbances in the physical network layer, updating train operation plan constraints in the service network layer, and activating the feedback mechanism between the two networks; Perform static and dynamic resilience assessments: First, by calling the static assessment module, based on the updated physical-service dual-network coupling model, using a hierarchical Bayesian network and an improved PageRank+Shapley value model, the static resilience index in the three-dimensional index system of structure-function-recovery is quickly calculated to assess the theoretical pressure boundary of the network under the current disturbance. Then, by executing the dynamic evaluation module: introducing the time dimension, utilizing the resilience evolution and dynamic evaluation module, combined with the LSTM-Attention temporal prediction model and cellular automata, the dynamic decay process of network performance over time is simulated, generating a preliminary resilience decay curve, and calculating the resilience index: in, This represents the system performance under normal conditions. The time of the disturbance Recovery time; Identify key nodes and their contributions: The key node identification and contribution assessment module analyzes which key nodes or links, under the current disturbance scenario, contribute the most to the overall network resilience degradation due to their failure. Preliminary results of the resilience assessment can be obtained by performing static and dynamic resilience assessment steps: The toughness threshold is determined as follows: The initial performance degradation trajectory prediction obtained in step S302 and the preliminary resilience assessment results obtained in step S401 are input into the judgment module, which determines whether cascading failure has occurred or whether the performance is below the threshold. If so: first trigger step S5 to obtain the cascading failure impact range, critical failure path and risk propagation coefficient; then input the output of step S5 into step S6 to complete risk probability prediction and graded early warning; If "No": This indicates that the current disturbance is still within the network's resilience range. The process then returns to steps S1 and S2 to continue monitoring the network status and performing routine weight updates, operational status corrections, or lightweight scheduling optimizations without immediately triggering a deep alert.

5. The method for dynamic assessment and risk warning of the resilience of a railway coupled network under multiple disturbance scenarios according to claim 1, characterized in that, Step S5 specifically includes the following steps: S501, cascading failure evolution simulation, specifically includes the coordinated operation of the following modules: Cellular Automata and Rule Base: Define the state and evolution rules of nodes / edges, consider the load-capacity model, trigger failure when the node load exceeds the capacity threshold, and redistribute the load to adjacent nodes; LSTM-Attention temporal prediction model: Combines historical cascading failure data to predict the failure propagation range and performance degradation trajectory at multiple future time steps; Graph Neural Networks and Percolation Theory: Extracting the critical diffusion path of the network, identifying the main propagation direction of cascade failure, calculating the risk propagation coefficient of each node / link, and generating a risk propagation heatmap; The above modules work together to output the risk causality of cascading failures, including the scope of impact, critical failure paths, and estimated time of network-wide paralysis. S502. Risk Causal Graph Construction: By integrating historical emergency texts, train operation data, and resilience assessment results, a risk causal graph is constructed using a hierarchical Bayesian network. The graph nodes include various perturbation factors, intermediate events, and final consequences, and the edges represent causal probability relationships. Through Bayesian inference, the probability of occurrence of different risk evolution paths is quantified.

6. The method for dynamic assessment and risk warning of the resilience of a railway coupled network under multiple disturbance scenarios according to claim 1, characterized in that, Step S6 specifically involves: GCN+GRU Multi-Spatiotemporal Risk Prediction Model: It integrates graph convolutional networks to capture network topology spatial features and gated cyclic units to capture temporal evolution features. Using cascaded failure simulation results and real-time monitoring data as input, it predicts the probability and risk level of risk occurrence for each node / interval in the next 4-72 hours. Railway Operation Management Risk Layering and Early Warning System: Based on a three-dimensional index of minimum required performance, recovery time, and functional retention rate, risks are divided into four levels: Level I is an exceptionally significant risk: system performance drops below the minimum required performance, and recovery time exceeds 72 hours; Level II is a major risk: system performance drops to less than 30% of normal levels, with a recovery time of 24-72 hours; Level III represents a significant risk: system performance drops to less than 50% of normal levels, with a recovery time of 12-24 hours. Level IV is a general risk: system performance drops to below 80% of normal levels, with a recovery time of less than 12 hours; Step S6 is used to trigger the corresponding level of early warning information, that is, to output the risk stratification early warning result and push it to the dispatch and command center. The early warning content includes: risk type, scope of impact, expected duration, and recommended handling measures.

7. The method for dynamic assessment and risk warning of the resilience of railway coupled networks under multiple disturbance scenarios according to claim 1, characterized in that, Step S7 receives the risk stratification warning result output in step S6, and activates the network resilience enhancement mechanism based on the risk level and main failure modes identified by the judgment module, as follows: Risk level and failure mode determination: Based on the risk stratification warning results generated in step S6, input them into the judgment module. The judgment module combines the risk cause-effect graph and the output of cascaded failure simulation to determine the risk level and main failure modes under the current disturbance scenario. The judgment results flow to two parallel modules: the resilience enhancement module and the transportation organization optimization module, which perform adaptive adjustments respectively; Network topology optimization specifically involves the coordinated work of the following modules: Load balancing optimization module: Monitors the real-time throughput and saturation of each section and station, dynamically adjusts train flow distribution, and avoids cascading failures caused by local overload; Shortest path optimization module: When the main line is damaged, it quickly calculates the backup path based on real-time topology weights and uses dynamic programming algorithm to achieve smooth path switching; Network redundancy optimization module: Configures physical and service redundancy for critical nodes and vulnerable areas to ensure that local failures do not spread to the entire network; Dynamic adjustment mechanism: Establish a closed loop of perception-decision-execution, update the network status every five minutes, dynamically adjust edge weights and node priorities, and adapt to the characteristics of limited communication and rapidly changing environment in western railways; The study on the cascading propagation mechanism of node failures and strategies to prevent cascading failures specifically includes the collaborative work of the following modules: Node importance analysis module: Utilizes an improved PageRank+Shapley value method, combined with real-time running data, to dynamically sort node importance; Cascade Effect Spatiotemporal Evolution Module: Based on cellular automata and graph neural networks, it simulates the load redistribution process after node failure and predicts the fault propagation range and time window. Anti-cascading failure strategy module: Outputs three types of intervention measures: Active isolation: Implement rate limiting or tripping operations on nodes that are about to be overloaded; Resource pre-positioning: Deploy repair teams and emergency supplies in advance in predicted high-risk areas; Topology reconstruction: Temporarily disconnect high-risk edges and activate redundant paths; Emergency response mechanism: Based on the risk level determined by the judgment module, the system automatically matches the handling procedures in the emergency plan library, sends the optimization results of the resilience enhancement module to the transportation organization optimization module, and initiates coordinated adjustments.

8. The method for dynamic assessment and risk warning of the resilience of a railway coupled network under multiple disturbance scenarios according to claim 7, characterized in that, Step S7 further includes transportation organization optimization and collaborative decision-making: Based on enhanced resilience, the transportation organization system is optimized through multiple objectives, and a comprehensive balance among resilience, efficiency, cost, and environment is achieved through a collaborative mechanism. The specific process is as follows: The first step is the restructuring of the transportation organization system, which specifically includes the collaborative work of the following modules: Hub Node Layout and Function Allocation Module: Adjusts the functional positioning of hub stations based on the dynamic rating of node importance output in step S6; Transportation efficiency improvement module: Based on reinforcement learning, temporary operation maps are generated to optimize train stopping schemes and tracking intervals, maximizing throughput under constrained conditions; High-efficiency recovery module: For the recovery phase after disturbance, a genetic algorithm is used to optimize the sorting of emergency repair tasks and the order of resource allocation, shortening the time for system performance to recover to 90%; Cost-effectiveness improvement module: Embed economic constraints in the adjustment plan to avoid resource waste caused by excessive redundancy; Environmental sustainability module: Prioritizes low-carbon scheduling strategies, calculates the changes in carbon emissions resulting from adjustment plans, and supports green operation goals; Secondly, there is a collaborative optimization mechanism, as detailed below: Dual-network collaboration: The repair progress of the physical network is synchronized with the adjustment plan of the service network in real time to avoid the situation where the road is open but the car has not arrived, or the car has arrived but the road is not open. Evaluation-Optimization Closed Loop: The optimization results of the resilience enhancement module and the transportation organization plan are fed back to the resilience evaluation system to recalculate the resilience index R and the resilience decay curve DRC. If the performance does not reach the threshold, the loop is repeated to the judgment module for re-judgment. Continuous learning: The processing of each perturbation event is stored in the case library and used periodically to fine-tune the policy network of the reinforcement learning agent, so as to achieve long-term evolution of the model; Finally, the output and execution are detailed below: Generate an executable instruction set for the dispatch and command center, including train timetable adjustment tables, route switching commands, emergency repair resource dispatch orders, and passenger / freight announcement suggestions; push these instructions to the railway dispatching system (CTC / TDCS) and emergency command platform via interfaces to complete the entire closed loop of assessment → early warning → resilience enhancement → transportation organization optimization → execution feedback.

Images