A low-altitude traffic flow prediction and diversion method based on a space-time diagram network

Abstract

The application relates to the technical field of low-altitude traffic, in particular to a low-altitude traffic flow prediction and dredging method based on a space-time graph network, which comprises the following steps: multi-source data acquisition and fusion; take-off and landing field demand anomaly monitoring; real-time state monitoring of air route nodes; joint diagnosis of node flow health; airspace resource constraint condition quantification; multi-dimensional evaluation of congestion risk level; dynamic evaluation of conflict risk; dynamic dredging strategy generation; potential risk evolution prediction and plan preparation. The application deeply fuses multi-source heterogeneous real-time and historical data through a space-time graph network, intelligently perceives and predicts the multi-dimensional and progressive low-altitude corridor traffic situation, diagnoses different levels of congestion risks from micro node overload to macro regional paralysis, and automatically generates a dredging strategy, so that the problems of decision lag, rough analysis, inefficient dredging and incapability of coping with dynamic congestion caused by the adoption of an isolated static threshold and single-dimensional analysis are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of low-altitude transportation technology, and in particular to a method for predicting and managing low-altitude traffic flow based on spatiotemporal graph networks. Background Technology

[0002] With the maturation of technologies such as electric vertical takeoff and landing (EVTOL) aircraft and the development of urban three-dimensional transportation, low-altitude operations, characterized by high-frequency, tidal travel, logistics, and emergency support, are driving the rapid evolution of airspace management from traditional models to a high-density, networked urban air traffic system. Against this backdrop, low-altitude traffic exhibits high dynamism, network connectivity, and spatiotemporal coupling, posing unprecedented core demands for real-time perception of operational status, accurate traffic flow prediction, and proactive, intelligent mitigation of congestion risks. These are key issues that must be addressed to ensure the safe and efficient large-scale operation of low-altitude traffic in the future.

[0003] Chinese Patent Application Publication No. CN120748257A discloses a method and system for optimizing traffic control in low-altitude economic zones. The method includes: dividing the airspace in the low-altitude economic zone to generate a multi-layered air corridor topology, and mapping traffic flow data to each layer of air corridors; formulating first flight control conditions based on the capacity constraints of each layer of air corridors in the multi-layered air corridor topology; formulating second flight control conditions based on the safety distance constraints of air corridors at the same layer in the multi-layered air corridor topology; and using the first and second flight control conditions to perform traffic scheduling and control in the low-altitude economic zone.

[0004] Therefore, the proposed method and system for optimizing traffic control in low-altitude economic zones have the following problems: the method relies on a pre-defined fixed air corridor structure, which makes it difficult to adapt to the drastic dynamic changes in urban low-altitude traffic and the real-time reconstruction needs of airspace use; flight control conditions are based on static capacity and safety thresholds, which are passive control measures that are triggered and responded to, lacking the ability to predict congestion trends and proactively alleviate congestion; the method introduces blockchain technology, which focuses on data storage and is prone to increasing system latency. Summary of the Invention

[0005] To address this, the present invention provides a method for predicting and managing low-altitude traffic flow based on spatiotemporal graph networks. This method overcomes the problems of decision lag, rough judgment, inefficient management, and inability to cope with dynamic congestion caused by the use of isolated static thresholds and single-dimensional analysis in existing technologies by constructing a spatiotemporal graph network model and a progressive risk diagnosis and hierarchical management decision chain.

[0006] To achieve the above objectives, this invention provides a method for low-altitude traffic flow prediction and mitigation based on spatiotemporal graph networks, comprising:

[0007] The number of flight plans and the statistics of historical actual flight missions for each take-off and landing field pair in the target corridor of the spatiotemporal map network of urban low-altitude operations within the preset observation period are obtained.

[0008] The degree of difference between the number of flight plans and the flight mission statistics is used to determine whether the traffic flow of the take-off and landing field pairs is abnormal;

[0009] Based on the traffic anomaly results, the instantaneous traffic and average ground speed of each preset route node connected by route segments in the spatiotemporal map network within the previous preset historical observation period are obtained.

[0010] Based on the distribution characteristics and time series characteristics of the instantaneous flow rate, combined with the preset intensity threshold and the threshold of the average flight ground speed, several abnormal flow rate nodes are identified.

[0011] Obtain the planned capacity utilization rate of the airspace sector corresponding to the abnormal traffic node;

[0012] The congestion level is determined based on the average ground speed of the abnormal flow nodes, the spatial characteristics of the abnormal flow nodes, and the planned capacity utilization rate, and several first-level risk nodes and second-level risk nodes are determined in combination with the congestion level.

[0013] Obtain the conflict frequency of the first-level risk nodes within the same preset historical observation period;

[0014] Real-time evacuation suggestions are generated based on the spatial correlation characteristics of the primary risk nodes and the frequency of conflicts.

[0015] Based on the threshold screening results of the instantaneous traffic flow and the number of flight plans, it is determined whether the secondary risk node will be upgraded to the primary risk node within the next preset risk period, and prediction and diversion suggestions are generated in combination with the determination results.

[0016] Furthermore, the process of determining whether the traffic flow of the takeoff and landing field pair is abnormal based on the degree of difference between the number of flight plans and the flight mission statistics includes:

[0017] Calculate the difference between the number of flight plans and the flight mission statistics to obtain the absolute demand increment;

[0018] When the absolute demand increment is greater than a preset increment threshold, the flow of the take-off and landing field pair is determined to be abnormal, so as to obtain the flow abnormality result.

[0019] Furthermore, the process of determining several abnormal traffic flow nodes based on the distribution characteristics and time series characteristics of the instantaneous traffic flow, combined with the preset intensity threshold and the threshold of the average flight ground speed, includes:

[0020] The flow growth intensity of each preset airway node is determined based on the time series of the instantaneous flow within the preset historical observation period;

[0021] Obtain the route segments that are associated with each of the preset route nodes to obtain a number of associated routes for each of the preset route nodes;

[0022] Calculate the average of the ratios of the average flight ground speed and the preset ground speed threshold for all the associated routes to obtain the speed drag of the preset route node;

[0023] When the flow rate increase intensity is greater than the preset intensity threshold and the speed stagnation is less than the preset ratio threshold, the flow rate of the preset route node is determined to be abnormal, so as to obtain a number of flow rate abnormal nodes.

[0024] Furthermore, the process of determining the congestion level based on the average ground speed of the traffic anomaly node, the spatial characteristics of the traffic anomaly node, and the planned capacity utilization rate, and then determining several primary and secondary risk nodes in conjunction with the congestion level, includes:

[0025] Centered on the abnormal traffic node, determine whether there is an abnormal traffic node in the spatiotemporal graph network that is connected to it through a segment of the flight path, so as to obtain a connectivity determination result;

[0026] Based on the connectivity determination result, the number of traffic abnormal nodes in the primary abnormal cluster formed by all the connected traffic abnormal nodes is recorded to obtain the number of abnormal nodes, and the minimum number of airway segments traversed between two connected traffic abnormal nodes in each primary abnormal cluster is calculated to obtain several shortest path segments.

[0027] The average number of shortest path segments within the primary abnormal cluster is calculated based on the number of abnormal nodes and the number of shortest path segments to obtain several cluster densities.

[0028] The resource pressure of a sector is determined by the coupling relationship between the number of abnormal nodes, the cluster density, and the planned capacity utilization rate of the airspace sector to which the primary abnormal cluster belongs, so as to obtain the resource pressure index.

[0029] The congestion level is determined based on the average flight ground speed, the resource pressure index, and the time dimension characteristics of the abnormal flow nodes, so as to obtain a number of first-level risk nodes and second-level risk nodes.

[0030] Furthermore, the process of determining the congestion level based on the average flight ground speed, the resource pressure index, and the time dimension characteristics of the traffic anomaly nodes to obtain a number of primary risk nodes and secondary risk nodes includes:

[0031] Based on the preset historical observation duration, the frequency at which the preset route node is identified as the traffic anomaly node is statistically analyzed to obtain several anomaly identification frequencies, and the standard deviation of the average flight ground speed of the traffic anomaly node is calculated to obtain several speed fluctuation values.

[0032] When the resource pressure index is greater than a preset pressure threshold, the anomaly determination frequency is greater than a preset frequency threshold, and the speed fluctuation value is less than a preset fluctuation threshold, the traffic anomaly node is determined to be the first-level risk node.

[0033] The abnormal traffic nodes that do not belong to the first-level risk nodes are identified as the second-level risk nodes.

[0034] Furthermore, the process of generating real-time mitigation suggestions based on the spatial correlation characteristics of the primary risk nodes and the conflict frequency includes:

[0035] Obtain the preset route nodes within the preset range of the first-level risk node to obtain several range node sets, and record the number of abnormal traffic nodes within the range node sets to obtain several range abnormality numbers.

[0036] The number of associated routes for each of the primary risk nodes is obtained to obtain a number of associated nodes, and the risk type of each primary risk node is determined based on the number of range anomalies and the number of associated nodes.

[0037] The spatiotemporal coupling degree of risk is determined based on the coupling relationship between the frequency of conflicts, the number of anomalies in the range, and the number of associated nodes.

[0038] Real-time traffic diversion instructions are generated based on the spatiotemporal coupling degree of the risk and the risk type to obtain the real-time diversion suggestions.

[0039] Furthermore, the process of determining the risk type of each primary risk node based on the number of anomalies in the range and the number of node associations includes:

[0040] When the number of associated nodes is greater than a preset association threshold and the number of abnormal ranges is less than a preset range threshold, the risk type is determined to be node overload.

[0041] When the number of associated nodes is less than a preset association threshold and the number of abnormal ranges is greater than a preset range threshold, the risk type is determined to be local network saturation.

[0042] When the number of associated nodes exceeds a preset association threshold and the number of abnormal ranges exceeds a preset range threshold, the risk type is determined to be hub overload that has caused regional paralysis.

[0043] Furthermore, the process of generating the real-time traffic diversion instruction based on the risk spatiotemporal coupling degree and the risk type to obtain the real-time diversion suggestion includes:

[0044] When the risk type is node overload, the real-time traffic diversion instruction is generated as dynamic airspace reorganization;

[0045] When the risk type is local network saturation, the real-time traffic diversion instruction is generated as path specification and collaborative sorting;

[0046] When the risk type is that the hub overload has caused regional paralysis, the real-time traffic diversion instruction is generated as a temporary corridor and traffic regulation;

[0047] The primary risk nodes are sorted in ascending order according to their risk spatiotemporal coupling degree to obtain a priority sequence;

[0048] The real-time traffic diversion suggestion is formed based on the priority sequence and the corresponding real-time traffic diversion instruction.

[0049] Furthermore, the process of determining whether the secondary risk node will escalate to the primary risk node within the next preset risk period based on the threshold screening results of the instantaneous traffic flow and the number of flight plans, and generating predictive guidance suggestions based on the determination results, includes:

[0050] Calculate the growth rate of the instantaneous flow rate at any two adjacent moments within the previous preset prediction time to obtain several instantaneous flow rate increases, and calculate the average value of all instantaneous flow rate increases to obtain the average flow rate increase.

[0051] Calculate the relative deviation between the number of flight plans and the preset historical number threshold within the same preset prediction period, and record the relative deviation as the growth rate of the number of flight plans;

[0052] When the average growth rate of traffic is greater than a preset growth rate threshold and the growth rate of quantity is greater than a preset growth rate threshold, it is determined that the secondary risk node will be upgraded to the primary risk node, and the secondary risk node is identified as a high-risk warning node.

[0053] The secondary risk nodes that do not belong to the high-risk early warning nodes are identified as steady-state monitoring nodes;

[0054] Based on the high-risk early warning node and the steady-state monitoring node, a predicted flow diversion instruction is generated to obtain the predicted diversion suggestion.

[0055] Furthermore, the process of generating predicted traffic diversion instructions based on the high-risk early warning nodes and the steady-state monitoring nodes to obtain the predicted diversion suggestions includes:

[0056] When the secondary risk node is the high-risk early warning node, the predicted flow diversion instruction is determined to be a pre-intervention instruction;

[0057] When the secondary risk node is the steady-state monitoring node, the predicted flow diversion instruction is determined to be a continuous monitoring instruction;

[0058] The pre-intervention instructions, the continuous monitoring instructions, and all the secondary risk nodes are integrated and output to obtain the predictive guidance suggestions.

[0059] Compared with existing technologies, the beneficial effects of this invention lie in its ability to achieve multi-dimensional, progressive intelligent perception and prediction of traffic conditions in low-altitude travel and operational corridors by constructing and utilizing a spatiotemporal graph network to deeply integrate multi-source heterogeneous real-time and historical data. Specifically, by performing correlation analysis and coupling calculations on discrete parameters such as flow rate, speed, schedule, and airspace capacity within a unified spatiotemporal graph, it can proactively and accurately diagnose different levels of congestion risks, ranging from micro-node overload to macro-regional paralysis. This not only overcomes the lag and limitations of traditional single-threshold alarms but also automatically generates hierarchical and executable diversion strategies based on risk type and spatiotemporal coupling, encompassing dynamic airspace reorganization, path collaborative optimization, and demand-source control. This represents a paradigm shift from passive response to proactive anticipation and collaborative diversion, significantly improving the operational safety, overall efficiency, and management flexibility of transportation systems in complex urban low-altitude environments. It effectively solves the problems of delayed decision-making, crude judgment, inefficient diversion, and inability to cope with dynamic congestion caused by the use of isolated static thresholds and single-dimensional analysis.

[0060] Furthermore, by dynamically comparing the number of currently confirmed flight plans with the historical statistics of actual flight missions during the same period, and using the quantified absolute demand increment as the anomaly criterion, accurate and forward-looking identification of traffic anomalies at takeoff and landing sites is achieved. In particular, by utilizing historical data validated in actual operation as a dynamic benchmark, daily fluctuations and periodic patterns are effectively separated, enabling the system to keenly capture abnormal demand signals exceeding historical norms. This allows for early warning of potential source traffic pressures during the planning stage before flight missions even take off. Overall, this overcomes the problems of poor adaptability and high false alarm rate of traditional static threshold methods, providing reliable and focused input for subsequent refined analysis at the node and network levels, and fundamentally improving the perception sensitivity and decision-making foresight of the entire prediction and diversion system.

[0061] Furthermore, by integrating the dynamic traffic growth trend and static traffic efficiency status of nodes, a dual-indicator joint judgment anomaly identification model was constructed. Specifically, by calculating the intensity of traffic growth, it can proactively perceive the accelerating convergence trend of traffic flow, rather than simply responding to already established high traffic volume; simultaneously, by quantifying the traffic health of associated routes through speed stagnation, it directly reflects the physical impact of congestion; and by combining the two through an AND logic judgment, false alarms caused by transient fluctuations or local interference are effectively filtered out, greatly improving the accuracy, reliability, and timeliness of anomaly node identification. This lays a solid foundation for subsequent precise risk classification and management, achieving an intelligent leap from judging quantity to diagnosing status.

[0062] Furthermore, by introducing network topology analysis and system resource coupling assessment, precise classification of traffic risks was achieved. Discrete anomalous nodes were placed within the spatiotemporal network for global review. By identifying the scale and structural density of anomalous node clusters and coupling them with the real-time capacity margin of their respective airspace sectors, the actual pressure exerted by localized congestion on the overall system was quantified. This effectively identified primary risk nodes that, due to their clustering and close correlation with each other, and their location in resource-constrained airspace, were highly susceptible to triggering cascading paralysis. These primary risk nodes were scientifically distinguished from isolated secondary risk nodes with limited impact. Through intelligent classification based on network structure and system load, key decision-making basis was provided for subsequent implementation of differentiated traffic management strategies that prioritized and precisely matched traffic needs, fundamentally improving the utilization efficiency of control resources and system resilience.

[0063] Furthermore, by introducing a triple joint verification of time persistence, state stability, and spatial pressure indicators, risk classification is upgraded from a static snapshot to a dynamic diagnosis. A node must simultaneously meet three stringent conditions—high spatial resource pressure, long duration of abnormal state, and persistently low but stable traffic efficiency—to be identified as a Level 1 risk requiring immediate action. This effectively filters out false high-risk signals caused by transient interference, short-term fluctuations, or recovering efficiency, ensuring that identified Level 1 risk nodes are truly deteriorating and pose a stable threat to the system. It also distinguishes these nodes from those that temporarily exhibit high-risk characteristics due to transient factors or short-term changes but do not actually pose a sustained threat to the system. This improves the accuracy of risk warnings, avoids excessive intervention and waste of control resources due to misjudgment, and allows traffic control instructions to be precisely focused on the most critical bottlenecks.

[0064] Furthermore, by intelligently coupling the network topology characteristics of nodes with dynamic temporal risks, a closed-loop decision-making process from risk diagnosis to precise treatment is achieved. Based on the different risk characteristics of nodes—hub-type, regional-type, or hybrid-type—differentiated strategies such as dynamic airspace reorganization, path coordination and sequencing, or temporary corridors and traffic regulation are automatically matched. Simultaneously, the risk density of each node per unit time is quantitatively assessed through spatiotemporal coupling, and execution priorities are determined accordingly. This ensures that valuable control resources and airspace capacity are prioritized and precisely deployed to the most urgent and critical bottlenecks, thereby maximizing traffic management efficiency and intelligent system intervention, fundamentally solving the problems of one-size-fits-all approaches and coarse responses in traditional methods.

[0065] Furthermore, by introducing two orthogonal dimensions—node centrality and regional risk density—a concise yet powerful classification framework was constructed, enabling accurate diagnosis of congestion pathology. From the perspective of network topology and spatial correlation, it can clearly distinguish three fundamentally different operational risks: key point congestion, localized surface congestion, and systemic paralysis involving both points and surfaces. Through intelligent classification, it provides crucial decision-making basis for the subsequent implementation of highly adaptable differentiated diversion strategies, thereby avoiding resource misallocation due to mismatch between strategies and risk types, and ensuring the accuracy, efficiency, and operational resilience of regulatory interventions.

[0066] Furthermore, the instruction generation and sorting process achieves a complete decision-making loop from risk classification to strategy matching and optimal resource allocation. Specifically, by automatically invoking highly compatible mitigation algorithms based on accurately diagnosed risk types, the targeted nature of intervention measures is ensured. Simultaneously, dynamic sorting of intervention targets based on the spatiotemporal coupling degree of risks establishes the priority of resource allocation, enabling limited airspace and control resources to be prioritized and precisely directed to the most urgent and hazardous bottlenecks. This maximizes intervention efficiency and minimizes system losses at the global level, marking a crucial leap in traffic management from homogeneous response to intelligent, differentiated, and precise treatment.

[0067] Furthermore, by integrating real-time traffic flow trends with the growth of planned demand at the source, a forward-looking risk escalation early warning mechanism was constructed. Specifically, through dual verification of the accelerated instantaneous traffic growth at secondary nodes and the abnormal growth of related flight plans, it can accurately identify nodes that are highly likely to deteriorate to Level 1 risk under the dual drive of dynamic and static demand, and precisely mark them as high-risk early warning nodes. This enables early and accurate prediction of potential congestion, allowing management to calmly pre-allocate resources and prepare contingency plans, generating pre-intervention instructions. For nodes with stable risks, only routine monitoring is implemented. This represents a crucial leap from reactive post-event response to proactive pre-event early warning and tiered preparedness, greatly improving the predictability and operational flexibility of low-altitude traffic management.

[0068] Furthermore, by implementing refined differentiation and targeted responses for high-risk early warning and steady-state monitoring of secondary risk nodes, a complete predictive management closed loop is constructed. Based on rigorous trend analysis, specific intervention plans, such as resource pre-allocation and contingency trigger conditions, are generated and prepared in advance for nodes poised for deterioration, achieving a seamless transition from early warning to readiness. Simultaneously, nodes in stable states are maintained only with efficient monitoring. This hierarchical preparation mechanism based on accurate prediction enables proactive and optimized allocation of control resources, significantly shortening the response time from risk identification to actual intervention. It also significantly enhances resilience and operational proactivity during sudden surges in traffic, truly realizing the intelligent management concept from prevention to decision-making. Attached Figure Description

[0069] Figure 1 This is a flowchart of the low-altitude traffic flow prediction and diversion method based on spatiotemporal graph networks in this embodiment.

[0070] Figure 2 This is a logic diagram for determining abnormal flow at the takeoff and landing site in this embodiment;

[0071] Figure 3 This is a logic diagram for determining abnormal traffic flow at preset route nodes in this embodiment;

[0072] Figure 4 This is a logic diagram for determining whether a node with abnormal traffic is a first-level risk node or a second-level risk node in this embodiment. Detailed Implementation

[0073] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0074] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0075] Please see Figure 1 The diagram shown is a flowchart of the low-altitude traffic flow prediction and diversion method based on a spatiotemporal graph network in this embodiment. This embodiment provides a low-altitude traffic flow prediction and diversion method based on a spatiotemporal graph network, including:

[0076] The number of flight plans and the statistics of historical actual flight missions for each take-off and landing field pair in the target corridor of the spatiotemporal map network of urban low-altitude operations within the preset observation period are obtained.

[0077] The degree of difference between the number of flight plans and the flight mission statistics is used to determine whether the traffic flow of the take-off and landing field pairs is abnormal;

[0078] Based on the traffic anomaly results, the instantaneous traffic and average ground speed of each preset route node connected by route segments in the spatiotemporal map network within the previous preset historical observation period are obtained.

[0079] Based on the distribution characteristics and time series characteristics of the instantaneous flow rate, combined with the preset intensity threshold and the threshold of the average flight ground speed, several abnormal flow rate nodes are identified.

[0080] Obtain the planned capacity utilization rate of the airspace sector corresponding to the abnormal traffic node;

[0081] The congestion level is determined based on the average ground speed of the abnormal flow nodes, the spatial characteristics of the abnormal flow nodes, and the planned capacity utilization rate, and several first-level risk nodes and second-level risk nodes are determined in combination with the congestion level.

[0082] Obtain the conflict frequency of the first-level risk nodes within the same preset historical observation period;

[0083] Real-time evacuation suggestions are generated based on the spatial correlation characteristics of the primary risk nodes and the frequency of conflicts.

[0084] Based on the threshold screening results of the instantaneous traffic flow and the number of flight plans, it is determined whether the secondary risk node will be upgraded to the primary risk node within the next preset risk period, and prediction and diversion suggestions are generated in combination with the determination results.

[0085] In this embodiment, the low-altitude traffic flow prediction and management method based on spatiotemporal graph networks is applied to high-frequency, tidal low-altitude travel and operation scenarios within mega-city clusters. This includes, but is not limited to, the comprehensive management of various low-altitude flight missions such as short-haul intercity manned flights, low-altitude logistics transportation, and emergency support. The target corridor is a pre-defined three-dimensional airspace channel designed to meet the low-altitude travel and operation needs of a specific city. It is an air trunk line defined by integrating factors such as origin-destination flow, geographical environment, and airspace restrictions. It is a subgraph or composite structure in the spatiotemporal graph network that connects multiple nodes and carries directional flow. It represents the overall spatial range in which all analyses and calculations occur, and is set based on urban demand analysis and airspace planning, using official aeronautical data such as aeronautical charts and airspace bulletins. The development process involves several key elements: 1. **Airfield pairings:** Within the target corridor, a fixed combination of the starting and ending vertical take-off and landing fields for a specific flight route. These are formed by combining two take-off and landing fields with stable flight operations, based on the operator's route plan. 2. **Pre-defined route nodes:** Within the three-dimensional airspace of the target corridor, a series of key spatial coordinate points pre-defined for refined traffic monitoring and path calculation. These are the vertices of the spatiotemporal map network and the sensor locations for sensing traffic flow. They are coordinate nodes set based on route structure and monitoring needs, such as river bends or canyon exits, and are entered into the navigation database of air traffic control and flight management systems. 3. **Airspace sectors:** Logical responsibility units that divide a vast airspace according to horizontal and vertical boundaries for efficient management. Each sector is managed by an independent control station and has a legally mandated maximum traffic capacity. They serve as management containers for assessing the tightness of airspace resource supply, set up based on national or regional airspace planning schemes or documents and capacity assessments, and directly connected to the air traffic control unit's traffic management system or airspace dynamic management platform.

[0086] In this embodiment, the spatiotemporal graph network is a core data model and computational framework established to realize intelligent management of urban low-altitude traffic. This network deeply integrates the spatial connectivity structure of the low-altitude corridor with the temporal dynamic evolution information of traffic flow. Its specific construction and operation are as follows: First, the construction of the spatial topology skeleton. Firstly, all pre-defined route nodes (such as navigation points, intersections, and rendezvous points) within the target low-altitude corridor are abstracted as "vertices" in the graph network. Each vertex corresponds to an actual route point with a unique identifier and precise three-dimensional geographic coordinates. Then, the route segments connecting these route nodes, which are available for aircraft flight, are abstracted as "edges" in the graph network. This forms a static network skeleton reflecting the physical connectivity of the airspace. The connectivity of this skeleton is defined by the adjacency matrix, clarifying whether there is a direct route connection between any two nodes. Second, the attachment of multi-source dynamic data. On the above static skeleton, time-series data from multiple sources such as monitoring systems and flight planning systems are fused in real time, assigning dynamic attributes to each vertex and each edge. Specifically: For each vertex (i.e., a route node), the system associates and updates a set of attributes at fixed time intervals (e.g., every minute). These attributes mainly include: the instantaneous number of aircraft (instantaneous flow) within the airspace unit near the node, statistically obtained based on real-time monitoring data (such as ADS-B and radar data); and the number of possible aircraft conflicts at the node, predicted based on future short-term flight plans. For each edge (i.e., a route segment), the system also calculates and updates its attributes at fixed time intervals, mainly including: the average ground speed of all aircraft on that route segment. These dynamic attribute values ​​form a continuous sequence over time, collectively constituting the "data layer" attached to the network skeleton. Third, the spatiotemporally coupled computational model. The spatiotemporal graph network, carrying the static skeleton and dynamic data, is input into a specially designed spatiotemporal graph computational model (e.g., a spatiotemporal graph neural network) for processing. The core task of this model is to simultaneously learn the dependency patterns of traffic flow in both spatial and temporal dimensions. One aspect is spatial dependency learning; the model uses mechanisms such as graph convolution to enable each node to automatically aggregate the state information of its neighboring nodes (i.e., nodes directly connected by edges). For example, the congestion state of a node is affected by the flow of its upstream and downstream nodes; this mechanism simulates this spatial propagation effect. Secondly, there is time-dependent learning. The model analyzes the sequential patterns of each node's own attributes (such as flow and speed) over time through mechanisms such as recurrent neural networks or temporal convolution, identifying its growth trends, periodicity, and other temporal characteristics. Through joint training, the model can effectively extract and fuse deep state features of nodes and the network based on historical and current spatiotemporal graph network data, thus providing a unified and powerful feature foundation for subsequent anomaly diagnosis and risk quantification calculations (such as flow growth intensity, cluster density, and risk spatiotemporal coupling). Fourthly, there is integration with the overall method. In this embodiment, the spatiotemporal graph network is not an isolated module but rather the intelligent hub of the entire method.First, multi-source information such as takeoff and landing site demand, flight mission plans, and real-time flight paths is uniformly encoded and injected into the network. Subsequently, the network, through its inherent spatiotemporal computing capabilities, achieves progressive perception and feature enhancement of the traffic situation from micro to macro levels. Finally, all key indicator calculations and logical judgments in subsequent steps regarding node anomaly detection, risk level assessment, and traffic management strategy generation are directly based on the fused features provided and derived by this network.

[0087] In this embodiment, a spatiotemporal map network integrating multi-source real-time data is constructed to achieve a progressive diagnosis from micro-node anomalies to macro-regional risks. The number of flight plans refers to the number of flight missions planned for a specific takeoff and landing pair within a specific future time period, submitted by the operator and formally accepted by the air traffic control system. This is the most direct input for predicting short-term traffic pressure and can be obtained in real-time from the flight plan processing system or the air traffic control's flight data processing system. Flight mission statistics refer to the historical average number of flight missions actually completed for the same takeoff and landing pair in one or more comparable historical periods, such as the same working day or time period last week. This can be obtained from the historical track database or flight mission execution record database of air traffic control or the takeoff and landing pair, categorized by takeoff and landing pair, date type, time period, etc. Historical data is filtered and queried based on dimensions such as weather conditions to calculate historical averages; instantaneous traffic flow refers to the number of aircraft located within a three-dimensional virtual airspace unit centered on a preset route node at a specific moment. This can be calculated based on the precise real-time positions of all aircraft in the network provided by wide-area surveillance networks such as Automatic Dependent Surveillance-Broadcast (ADS-B) and radar, combined with pre-defined geographical boundaries for each preset route node, such as a spherical airspace with a radius of 500 meters, to automatically count the number of aircraft within the airspace of each node; conflict frequency refers to the prediction of future flight schedules based on currently effective flight plans. Within the context of pre-defined route nodes that serve as key convergence points, such as route intersections and mergings, the predicted number of potential conflicts between aircraft that fall below the safe separation standard can be determined using the 4D trajectory-based conflict detection function in the air traffic control automation system. This function converts flight plans into four-dimensional tracks containing precise time and space information. At designated convergence points, all track pairs are compared pairwise. If the predicted distance between them at the same time and space location is lower than the minimum safe separation, it is counted as a conflict. All such events in future time periods are then statistically analyzed to determine the conflict frequency at that convergence point. Average ground speed refers to the average ground speed of all aircraft on a specific route segment. The real-time ground speed of an aircraft, i.e., the arithmetic mean of its horizontal speed relative to the ground, can be obtained by using the same Automatic Dependent Surveillance-Supervisory System (ADS-A) or Radar Surveillance-Supervisory System (RSSS) to determine the aircraft's route segment based on its position, then acquiring its real-time reported or calculated ground speed, and averaging the ground speeds of all aircraft passing through that route segment within a specified time period. Planned capacity utilization refers to the ratio of the number of planned flight missions in a given airspace sector to the sector's rated safe capacity for the corresponding time period. This can be achieved by using the air traffic control system to pre-determine and dynamically adjust sector capacity for each time period based on factors such as sector complexity, controller workload, and weather. This method obtains the planned flow and rated capacity of the target sector in real time through a system interface and calculates their ratio.

[0088] The preset observation period is a fixed time window set for acquiring and analyzing real-time traffic flow data. It depends on the evolution speed of the traffic phenomenon of interest and the reaction time required for control decisions, and is typically set between 15 minutes and 2 hours. In this embodiment, it is set to 1 hour, which effectively covers a complete cycle of traffic accumulation or dissipation for urban low-altitude travel and operational traffic, thus balancing the real-time nature of the data with statistical stability. The preset intensity threshold is a key benchmark value used to determine whether the growth trend of node traffic constitutes an anomaly. It depends on the normal fluctuation range of historical traffic data and the requirements for risk warning sensitivity, and is typically set between 0.2 flights / minute and 0.5 flights / minute. In this embodiment, it is set to increase by 0.3 flights / minute per minute, which can effectively distinguish between a gradual increase caused by random fluctuations and an increase indicating potential congestion. Accelerating growth; the preset historical observation duration is a continuous retrospective time window set for analyzing the historical status of airway nodes. It depends on the typical cycle length of traffic flow patterns and the statistical stability required for analysis, and is usually set between 30 minutes and 2 hours. In this embodiment, it is set to 60 minutes, which can cover several complete short-term traffic fluctuation cycles in urban low-altitude travel and operation scenarios; the preset risk duration is a future time window set for predicting whether a secondary risk node will escalate into a primary risk node. It depends on the decision-making and response time required from risk warning to implementation of diversion, and is usually set between 15 minutes and 1 hour. In this embodiment, it is set to 30 minutes, which can provide a sufficiently forward-looking but relatively reliable short-term prediction period, thereby achieving early warning of risk evolution and leaving the necessary operational window for generating preliminary diversion suggestions.

[0089] By constructing and utilizing a spatiotemporal graph network to deeply integrate multi-source heterogeneous real-time and historical data, multi-dimensional and progressive intelligent perception and prediction of traffic conditions in low-altitude travel and operational corridors have been achieved. Specifically, by performing correlation analysis and coupled calculations on discrete parameters such as flow rate, speed, schedule, and airspace capacity within a unified spatiotemporal graph, it is possible to proactively and accurately diagnose different levels of congestion risk, ranging from micro-node overload to macro-regional paralysis. This not only overcomes the lag and limitations of traditional single-threshold alarms but also automatically generates hierarchical and executable diversion strategies based on risk type and spatiotemporal coupling, encompassing dynamic airspace reorganization, path collaborative optimization, and demand-source control. This represents a paradigm shift from passive response to proactive anticipation and collaborative diversion, significantly improving the operational safety, overall efficiency, and management flexibility of transportation systems in complex urban low-altitude environments. It effectively solves the problems of delayed decision-making, crude judgment, inefficient diversion, and inability to cope with dynamic congestion caused by the use of isolated static thresholds and single-dimensional analysis.

[0090] Please see Figure 2As shown, this is the logic diagram for determining abnormal traffic flow between takeoff and landing fields in this embodiment. In this embodiment, the process of determining whether the traffic flow between the takeoff and landing fields is abnormal based on the degree of difference between the number of flight plans and the flight mission statistics includes:

[0091] Calculate the difference between the number of flight plans and the flight mission statistics to obtain the absolute demand increment;

[0092] When the absolute demand increment is greater than a preset increment threshold, the flow of the take-off and landing field pair is determined to be abnormal, so as to obtain the flow abnormality result.

[0093] The preset incremental threshold is a benchmark value used to determine whether the traffic flow at the take-off and landing field is abnormal. It depends on the normal fluctuation range of historical demand and the sensitivity requirements of the operation strategy for risk warning. It is usually set between 5 and 15 flights. In this embodiment, it is set to 8 flights, which can effectively distinguish between random demand changes caused by plan fluctuations and abnormal demand growth that indicates potential congestion risks. This enables accurate early warning at the source of traffic demand, reduces misjudgments, and improves the pertinence of subsequent node-level analysis.

[0094] By dynamically comparing the number of currently confirmed flight plans with the historical statistics of actual flight missions during the same period, and using quantified absolute demand increments as anomaly criteria, this system achieves accurate and forward-looking identification of traffic anomalies at takeoff and landing sites. Specifically, by utilizing historically validated data as a dynamic benchmark, it effectively separates daily fluctuations and periodic patterns, enabling it to keenly capture abnormal demand signals exceeding historical norms. This allows for early warning of potential source traffic pressures before flight missions even take off. Overall, this system overcomes the problems of poor adaptability and high false alarm rates associated with traditional static threshold methods, providing reliable and focused input for subsequent refined analysis at the node and network levels. It fundamentally improves the perception sensitivity and decision-making foresight of the entire prediction and mitigation system.

[0095] Please see Figure 3 As shown, this is the logic diagram for determining abnormal traffic flow at preset route nodes in this embodiment. In this embodiment, the process of determining several abnormal traffic flow nodes based on the distribution characteristics and time series characteristics of the instantaneous traffic flow, combined with the preset intensity threshold and the threshold of the average flight ground speed, includes:

[0096] The slope of the time-flow curve is obtained by fitting the instantaneous flow rate corresponding to each moment within the preset historical observation period, so as to obtain the flow rate growth intensity of each preset route node.

[0097] Obtain the route segments with each of the preset route nodes as the starting point or the ending point, so as to obtain several associated routes of each of the preset route nodes;

[0098] Calculate the average of the ratios of the average flight ground speed and the preset ground speed threshold for all the associated routes to obtain the speed drag of the preset route node;

[0099] When the flow rate increase intensity is greater than the preset intensity threshold and the speed stagnation is less than the preset ratio threshold, the flow rate of the preset route node is determined to be abnormal, so as to obtain a number of flow rate abnormal nodes.

[0100] The preset ratio threshold is a critical ratio used to determine whether the traffic efficiency of the associated airways has significantly decreased. It depends on the acceptable range of efficiency loss in historical operations and the sensitivity of congestion warnings, and is usually set between 0.7 and 0.85. In this embodiment, it is set to 0.75, which can effectively identify a substantial decrease in the traffic efficiency of associated airways caused by traffic accumulation while ensuring insensitivity to slight speed fluctuations. The preset ground speed threshold is a benchmark speed value for measuring the theoretical traffic capacity of associated airways. It depends on the airway design standards, aircraft type combination, and meteorological condition correction coefficients, and is usually set between 180 km / h and 250 km / h. In this embodiment, it is set to 220 km / h, which can accurately reflect the expected flight speed of associated airways under ideal conditions and provide a reasonable reference benchmark for calculating speed drag.

[0101] By integrating the dynamic traffic growth trend and static traffic efficiency status of nodes, a dual-indicator joint judgment anomaly identification model was constructed. Specifically, by calculating the intensity of traffic growth, it can proactively perceive the accelerating convergence trend of traffic flow, rather than simply responding to existing high traffic volume. Simultaneously, by quantifying the traffic health of associated routes through speed stagnation, it directly reflects the physical impact of congestion. Combining these two metrics through an AND operation effectively filters out false alarms caused by transient fluctuations or localized interference, significantly improving the accuracy, reliability, and timeliness of anomaly node identification. This lays a solid foundation for subsequent precise risk classification and management, achieving an intelligent leap from judging quantity to diagnosing status.

[0102] Specifically, the process of determining the congestion level based on the average ground speed of the traffic anomaly nodes, the spatial characteristics of the traffic anomaly nodes, and the planned capacity utilization rate, and then determining several primary and secondary risk nodes based on the congestion level, includes:

[0103] Centered on the abnormal traffic node, determine whether there is an abnormal traffic node in the spatiotemporal graph network that is connected to it through a segment of the flight path, so as to obtain a connectivity determination result;

[0104] Based on the connectivity determination result, the number of traffic abnormal nodes in the primary abnormal cluster formed by all the connected traffic abnormal nodes is recorded to obtain the number of abnormal nodes, and the minimum number of airway segments traversed between two connected traffic abnormal nodes in each primary abnormal cluster is calculated to obtain several shortest path segments.

[0105] The average number of shortest path segments within the primary abnormal cluster is calculated based on the number of abnormal nodes and the number of shortest path segments to obtain several cluster densities, where M=D / [S×(S-1) / 2], where M is the cluster density of each traffic abnormal node, D is the total number of shortest path segments of each traffic abnormal node, and S is the number of abnormal nodes.

[0106] Based on the number of abnormal nodes, the cluster density, and the planned capacity utilization rate of the airspace sector to which the primary abnormal cluster belongs, the sector resource pressure is determined to obtain the resource pressure index, where Z=[Y×(1 / C)] / (1-R), where Z is the resource pressure index, Y is the number of abnormal nodes, C is the cluster density, and R is the planned capacity utilization rate.

[0107] The congestion level is determined based on the average flight ground speed, the resource pressure index, and the time dimension characteristics of the abnormal flow nodes, so as to obtain a number of first-level risk nodes and second-level risk nodes.

[0108] By introducing network topology analysis and system resource coupling assessment, precise classification of traffic risks is achieved. Discrete anomalous nodes are placed within a spatiotemporal network for global review. By identifying the scale and structural density of anomalous node clusters and coupling them with the real-time capacity margin of their respective airspace sectors, the actual pressure exerted by localized congestion on the overall system can be quantified. This effectively identifies primary risk nodes that are highly susceptible to cascading paralysis due to clustering and close correlation of anomalous nodes within resource-constrained airspace. These primary risk nodes are scientifically distinguished from isolated secondary risk nodes with limited impact. Intelligent classification based on network structure and system load provides crucial decision-making support for subsequent implementation of differentiated traffic management strategies that prioritize and precisely match risks, fundamentally improving the utilization efficiency of control resources and system resilience.

[0109] Please see Figure 4 As shown, this is a logic diagram for determining whether a traffic anomaly node is a first-level risk node or a second-level risk node in this embodiment. In this embodiment, the process of determining the congestion level based on the average flight ground speed, the resource pressure index, and the time dimension characteristics of the traffic anomaly node to obtain several first-level risk nodes and second-level risk nodes includes:

[0110] Based on the preset historical observation duration, the frequency at which the preset route node in the primary anomaly cluster is identified as the traffic anomaly node is statistically analyzed to obtain several anomaly identification frequencies, and the standard deviation of the average flight ground speed of the traffic anomaly node is calculated to obtain several speed fluctuation values.

[0111] When the resource pressure index is greater than a preset pressure threshold, the anomaly determination frequency is greater than a preset frequency threshold, and the speed fluctuation value is less than a preset fluctuation threshold, the traffic anomaly node is determined to be the first-level risk node.

[0112] The abnormal traffic nodes that do not belong to the first-level risk nodes are identified as the second-level risk nodes.

[0113] The preset frequency threshold is a benchmark value used to determine whether a traffic anomaly node is frequently identified as a traffic anomaly node. It depends on the stringency requirements for the persistence of risk and is usually set between 5 and 15 times. In this embodiment, it is set to 10 times, which can effectively distinguish between nodes with more frequent abnormal fluctuations and nodes that occasionally show anomalies. The preset fluctuation threshold is a critical value used to determine whether the speed fluctuation of a traffic anomaly node is too large. It depends on the definition of speed fluctuation in a stable congestion state and is usually set between 0.7 and 0.85. In this embodiment, it is set to 0.75, which can effectively balance the tolerance for normal slight fluctuations and the vigilance for congestion risks. The preset pressure threshold is a reference value used to measure whether a preset airway node is under high load in terms of resource carrying capacity. It depends on the planned upper limit of airway node resources and the actual operational safety margin of the low-altitude traffic management system. It is usually set between 0.6 and 0.8. In this embodiment, it is set to 0.7, which can better reflect the critical point of transition from normal operation to resource shortage state of airway node under typical low-altitude traffic density.

[0114] By introducing a triple verification of time persistence, state stability, and spatial pressure indicators, risk classification is upgraded from a static snapshot to a dynamic diagnosis. A node must simultaneously meet three stringent conditions—high spatial resource pressure, long duration of abnormal state, and persistently low but stable traffic efficiency—to be identified as a Level 1 risk requiring immediate action. This effectively filters out false high-risk signals caused by transient interference, short-term fluctuations, or recovering efficiency, ensuring that identified Level 1 risk nodes are truly deteriorating and pose a stable threat to the system. It also distinguishes these nodes from those that temporarily exhibit high-risk characteristics due to transient factors or short-term changes but do not actually pose a sustained threat to the system. This improves the accuracy of risk warnings, avoids excessive intervention and wasted control resources due to misjudgments, and allows traffic control instructions to be precisely focused on the most critical bottlenecks.

[0115] Specifically, the process of generating real-time mitigation suggestions based on the spatial correlation characteristics of the primary risk nodes and the conflict frequency includes:

[0116] Obtain the preset route nodes within a preset number of flight segments centered on each of the first-level risk nodes to obtain several range node sets, and record the number of traffic abnormal nodes within the range node sets to obtain several range abnormality numbers.

[0117] The number of associated routes for each of the primary risk nodes is obtained to obtain a number of associated nodes, and the risk type of each primary risk node is determined based on the number of range anomalies and the number of associated nodes.

[0118] The risk spatiotemporal coupling degree is determined based on the coupling relationship between the conflict frequency, the number of range anomalies, and the number of node associations, where F = J × (α × L + β × G), where F is the risk spatiotemporal coupling degree, J is the conflict frequency, L is the number of range anomalies, α is a preset quantity weight, G is the number of node associations, and β is a preset association weight.

[0119] Real-time traffic diversion instructions are generated based on the spatiotemporal coupling degree of the risk and the risk type to obtain the real-time diversion suggestions.

[0120] The preset range of flight segments is used to define the topological distance of the local impact area around the primary risk node. It depends on the typical range of traffic flow interaction and risk diffusion in the urban low-altitude network and is usually set between 1 and 3 flight segments. In this embodiment, it is set to 2 flight segments, which can effectively capture the neighboring route nodes that may be directly affected or associated by congestion. The preset quantity weight is a coefficient used to measure the relative importance of the "range anomaly quantity" indicator when calculating the spatiotemporal coupling degree of risk. It depends on the assessment emphasis on regional risk aggregation and the hub importance of the node itself. It is usually set between 0.5 and 0.8. In this embodiment, it is set to 0.7, which can give a higher weight to the regional anomaly density in the comprehensive risk assessment and emphasize the spatial aggregation effect of risk. The preset association weight is a coefficient used to measure the relative importance of the number of node associations when calculating the spatiotemporal coupling degree of risk. It is usually set between 0.2 and 0.5. In this embodiment, it is set to 0.3, which can appropriately consider the structural importance and connection vulnerability of the node itself in the network while emphasizing regional risk.

[0121] By intelligently coupling the network topology characteristics of nodes with dynamic temporal risks, a closed-loop decision-making process from risk diagnosis to precise treatment is achieved. Based on the different risk characteristics of nodes—hub-type, regional-type, or hybrid-type—differentiated strategies such as dynamic airspace reorganization, path coordination and sequencing, or temporary corridors and traffic regulation are automatically matched. Simultaneously, the risk density of each node per unit time is quantitatively assessed through spatiotemporal coupling, and execution priorities are determined accordingly. This ensures that valuable control resources and airspace capacity are prioritized and precisely deployed to the most urgent and critical bottlenecks, thereby maximizing traffic management efficiency and intelligent system intervention, fundamentally solving the problems of one-size-fits-all approaches and coarse responses in traditional methods.

[0122] Specifically, the process of determining the risk type of each primary risk node based on the number of anomalies in the range and the number of associated nodes includes:

[0123] When the number of associated nodes is greater than a preset association threshold and the number of abnormal ranges is less than a preset range threshold, the risk type is determined to be node overload.

[0124] When the number of associated nodes is less than a preset association threshold and the number of abnormal ranges is greater than a preset range threshold, the risk type is determined to be local network saturation.

[0125] When the number of associated nodes exceeds a preset association threshold and the number of abnormal ranges exceeds a preset range threshold, the risk type is determined to be hub overload that has caused regional paralysis.

[0126] In this embodiment, the three risk types defined are essentially precise depictions of different stages of traffic congestion evolution from point to area, based on the importance of a node's own connectivity and the state of its surrounding network. Among them, node overload refers to the risk being highly concentrated in a single critical hub, where the node itself has many connected routes and bears enormous pressure, but the anomaly has not yet spread to its surrounding route network; local network saturation refers to the risk exhibiting distributed characteristics, where multiple interconnected nodes in a certain local geographical area simultaneously experience anomalies, indicating that congestion has formed a networked spread, but the core hub may not have collapsed; hub overload has caused regional paralysis, which means that the above two extreme situations occur simultaneously, that is, the overload of a core hub node and the saturation of its local network superimpose each other, creating a vicious cycle, marking that congestion has escalated from a single point or local problem to a systemic crisis that may cause the entire region's traffic capacity to fail.

[0127] The preset association threshold is a baseline number used to determine whether a node belongs to a high-connectivity critical hub. It depends on the average connectivity of each node in the target corridor network topology and is usually set between 4 and 6. In this embodiment, it is set to 5. That is, when the number of directly associated routes of a node is greater than 5, it is considered to have significant hub attributes and can effectively identify route nodes that play a core transit role in the network. The preset range threshold is a baseline number used to determine the degree of abnormal spread in the local network around a node. The value depends on the definition of the scale of the local area and the sensitivity of abnormal association. It is usually set between 2 and 4. In this embodiment, it is set to 3. That is, when the number of abnormal nodes within a two-hop range around a node is greater than 3, it is considered that the local network where it is located has shown obvious saturation characteristics and can accurately capture the spatial aggregation and spread of abnormalities.

[0128] By introducing two orthogonal dimensions—node centrality and regional risk density—a concise yet powerful classification framework is constructed, enabling accurate diagnosis of congestion pathology. From the perspective of network topology and spatial correlation, it can clearly distinguish three fundamentally different operational risks: key point congestion, localized area congestion, and systemic paralysis involving both points and areas. Through intelligent classification, it provides crucial decision-making basis for the subsequent implementation of highly adaptable differentiated diversion strategies, thereby avoiding resource misallocation due to mismatch between strategies and risk types, and ensuring the accuracy, efficiency, and operational resilience of regulatory interventions.

[0129] Specifically, the process of generating the real-time traffic diversion instruction based on the risk spatiotemporal coupling degree and the risk type to obtain the real-time diversion suggestion includes:

[0130] When the risk type is node overload, the real-time traffic diversion instruction is generated as dynamic airspace reorganization;

[0131] When the risk type is local network saturation, the real-time traffic diversion instruction is generated as path specification and collaborative sorting;

[0132] When the risk type is that the hub overload has caused regional paralysis, the real-time traffic diversion instruction is generated as a temporary corridor and traffic regulation;

[0133] The primary risk nodes are sorted in ascending order according to their risk spatiotemporal coupling degree to obtain a priority sequence;

[0134] The real-time traffic diversion suggestion is formed based on the priority sequence and the corresponding real-time traffic diversion instruction.

[0135] In this embodiment, dynamic airspace reorganization is a system-level intervention strategy for critical hub-type congestion. It temporarily alters local airspace usage rules and traffic allocation logic centered on risk nodes, intelligently and proportionally guiding overloaded traffic converging there to upstream alternative nodes with stronger carrying capacity, thereby achieving rapid and adaptive optimization of the airspace network structure. Path assignment and collaborative sequencing is a refined scheduling strategy for regionally diffused congestion. It does not change the network topology but, within a saturated local network, calculates and assigns the optimal path with the lowest conflict cost for each aircraft and allocates safe and efficient sequential arrival times to multiple aircraft arriving at the same convergence point, thereby tapping and enhancing the capacity of the existing network at a micro-level. Temporary corridors and traffic regulation is a high-level combined strategy for mixed and complex congestion. It simultaneously implements supply-side improvements and demand-side management: on the one hand, it plans and activates emergency diversion corridors to provide new physical channels; on the other hand, it adjusts takeoff times based on traffic gaps at the takeoff and landing fields at the congestion source, addressing regional paralysis caused by hub collapse through a two-pronged approach.

[0136] The instruction generation and sorting process achieves a complete decision-making loop, from risk classification to strategy matching and optimal resource allocation. Specifically, by automatically invoking highly compatible mitigation algorithms based on accurately diagnosed risk types, the targeted nature of intervention measures is ensured. Simultaneously, dynamic sorting of intervention targets based on the spatiotemporal coupling degree of risks establishes the priority of resource allocation, enabling limited airspace and control resources to be prioritized and precisely directed to the most urgent and threatening bottlenecks. This maximizes intervention efficiency and minimizes system losses at the global level, marking a crucial leap in traffic management from homogeneous response to intelligent, differentiated, and precise treatment.

[0137] Specifically, the process of determining whether the secondary risk node will escalate to the primary risk node within the next preset risk period based on the threshold screening results of the instantaneous traffic flow and the number of flight plans, and generating predictive guidance suggestions based on the determination results, includes:

[0138] Calculate the growth rate of the instantaneous flow rate at any two adjacent moments within the previous preset prediction time to obtain several instantaneous flow rate increases, and calculate the average value of all instantaneous flow rate increases to obtain the average flow rate increase.

[0139] Calculate the relative deviation between the number of flight plans and the preset historical number threshold within the same preset prediction period, and record the relative deviation as the growth rate of the number of flight plans;

[0140] When the average growth rate of traffic is greater than a preset growth rate threshold and the growth rate of quantity is greater than a preset growth rate threshold, it is determined that the secondary risk node will be upgraded to the primary risk node, and the secondary risk node is identified as a high-risk warning node.

[0141] The secondary risk nodes that do not belong to the high-risk early warning nodes are identified as steady-state monitoring nodes;

[0142] Based on the high-risk early warning node and the steady-state monitoring node, a predicted flow diversion instruction is generated to obtain the predicted diversion suggestion.

[0143] The preset forecast duration is a continuous observation time window used to analyze short-term traffic flow trends. It depends on the forecasted demand and the statistical stability of short-term traffic flow fluctuations, and is usually set between 20 and 60 minutes. In this embodiment, it is set to 30 minutes, which can provide sufficient data points for calculating the short-term acceleration of traffic flow, thereby reliably determining whether the node traffic has entered a dangerous channel of accelerated growth. The preset historical quantity threshold is a benchmark used to measure whether the number of flight plans has increased abnormally. It depends on the statistical analysis of the operational data of a specific take-off and landing pair in the historical comparable period, and is usually set between 1.2 and 1.5 times the historical baseline value. In this embodiment, it is set to 1.3 times the historical baseline value, which can serve as an objective quantitative benchmark for judging whether the demand for flight plans significantly exceeds the historical normal range. The preset growth threshold is a critical acceleration value used to determine whether the instantaneous traffic growth at a node is too rapid. It depends on the risk tolerance for the accelerated influx of traffic and is usually set between 1 flight / minute and 3 flights / minute. In this embodiment, it is set to 2 flights / minute, which can keenly capture the inflection point where traffic changes from steady or linear growth to accelerated accumulation, and provide timely warnings that the node's load capacity is rising sharply. The preset growth threshold is a critical percentage used to determine whether the growth rate of the number of flight plans is abnormal. It depends on the warning sensitivity to the surge in source demand and is usually set between 15% and 40%. In this embodiment, it is set to 25%, that is, when the number of future plans increases by more than a quarter compared to the historical baseline, an early warning is triggered, which can effectively screen out those source demand surge signals that may cause a chain reaction of overload in downstream nodes.

[0144] By integrating real-time traffic flow trends with planned demand growth at the source, a forward-looking risk escalation early warning mechanism has been constructed. Specifically, through dual verification of the accelerated instantaneous traffic growth at secondary nodes and the abnormal growth of related flight plans, it can accurately identify nodes that are highly likely to deteriorate to Level 1 risk under the dual drive of dynamic and static demand, and precisely mark them as high-risk early warning nodes. This enables early and accurate prediction of potential congestion, allowing management to calmly pre-allocate resources and prepare contingency plans, generating pre-intervention instructions. Nodes with stable risks are only subject to routine monitoring. This represents a crucial leap from reactive post-event response to proactive early warning and tiered preparedness, greatly improving the predictability and operational flexibility of low-altitude traffic management.

[0145] Specifically, the process of generating predicted traffic diversion instructions based on the high-risk early warning nodes and the steady-state monitoring nodes to obtain the predicted diversion suggestions includes:

[0146] When the secondary risk node is the high-risk early warning node, the predicted flow diversion instruction is determined to be a pre-intervention instruction;

[0147] When the secondary risk node is the steady-state monitoring node, the predicted flow diversion instruction is determined to be a continuous monitoring instruction;

[0148] The pre-intervention instructions, the continuous monitoring instructions, and all the secondary risk nodes are integrated and output to obtain the predictive guidance suggestions.

[0149] By implementing refined differentiation and targeted responses for high-risk early warning and steady-state monitoring of secondary risk nodes, a complete predictive management closed loop has been constructed. Based on rigorous trend analysis, specific intervention plans, such as resource pre-allocation and contingency trigger conditions, are generated and prepared in advance for nodes poised for deterioration, achieving a seamless transition from early warning to readiness. Simultaneously, nodes in stable states are maintained only with efficient monitoring. This hierarchical preparation mechanism based on accurate prediction enables proactive and optimized allocation of control resources, significantly shortening the response time from risk identification to actual intervention. It also significantly enhances resilience and operational proactivity during sudden surges in traffic, truly realizing the intelligent management concept from prevention to decision-making.

[0150] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting and managing low-altitude traffic flow based on spatiotemporal graph networks, characterized in that, include: The number of flight plans and the statistics of historical actual flight missions for each take-off and landing field pair in the target corridor of the spatiotemporal map network of urban low-altitude operations within the preset observation period are obtained. The degree of difference between the number of flight plans and the flight mission statistics is used to determine whether the traffic flow of the take-off and landing field pairs is abnormal; Based on the traffic anomaly results, the instantaneous traffic and average ground speed of each preset route node connected by route segments in the spatiotemporal map network within the previous preset historical observation period are obtained. Based on the distribution characteristics and time series characteristics of the instantaneous flow rate, combined with the preset intensity threshold and the threshold of the average flight ground speed, several abnormal flow rate nodes are identified. Obtain the planned capacity utilization rate of the airspace sector corresponding to the abnormal traffic node; The congestion level is determined based on the average ground speed of the abnormal flow nodes, the spatial characteristics of the abnormal flow nodes, and the planned capacity utilization rate, and several first-level risk nodes and second-level risk nodes are determined in combination with the congestion level. Obtain the conflict frequency of the first-level risk nodes within the same preset historical observation period; Real-time evacuation suggestions are generated based on the spatial correlation characteristics of the primary risk nodes and the frequency of conflicts. Based on the threshold screening results of the instantaneous traffic flow and the number of flight plans, it is determined whether the secondary risk node will be upgraded to the primary risk node within the next preset risk period, and prediction and diversion suggestions are generated in combination with the determination results.

2. The method for low-altitude traffic flow prediction and diversion based on spatiotemporal graph networks according to claim 1, characterized in that, The process of determining whether the traffic flow of the takeoff and landing field pairs is abnormal based on the degree of difference between the number of flight plans and the flight mission statistics includes: Calculate the difference between the number of flight plans and the flight mission statistics to obtain the absolute demand increment; When the absolute demand increment is greater than a preset increment threshold, the flow of the take-off and landing field pair is determined to be abnormal, so as to obtain the flow abnormality result.

3. The method for low-altitude traffic flow prediction and diversion based on spatiotemporal graph networks according to claim 2, characterized in that, The process of determining several abnormal flow nodes based on the distribution characteristics and time series characteristics of the instantaneous flow, combined with the preset intensity threshold and the threshold of the average flight ground speed, includes: The flow growth intensity of each preset airway node is determined based on the time series of the instantaneous flow within the preset historical observation period; Obtain the route segments that are associated with each of the preset route nodes to obtain a number of associated routes for each of the preset route nodes; Calculate the average of the ratios of the average flight ground speed and the preset ground speed threshold for all the associated routes to obtain the speed drag of the preset route node; When the flow rate increase intensity is greater than the preset intensity threshold and the speed stagnation is less than the preset ratio threshold, the flow rate of the preset route node is determined to be abnormal, so as to obtain a number of flow rate abnormality nodes.

4. The method for low-altitude traffic flow prediction and diversion based on spatiotemporal graph networks according to claim 3, characterized in that, The process of determining the congestion level based on the average ground speed of the traffic anomaly nodes, the spatial characteristics of the traffic anomaly nodes, and the planned capacity utilization rate, and then determining several primary and secondary risk nodes based on the congestion level, includes: Centered on the abnormal traffic node, determine whether there is an abnormal traffic node in the spatiotemporal graph network that is connected to it through a segment of the flight path, so as to obtain a connectivity determination result; Based on the connectivity determination result, the number of traffic abnormal nodes in the primary abnormal cluster formed by all the connected traffic abnormal nodes is recorded to obtain the number of abnormal nodes, and the minimum number of airway segments traversed between two connected traffic abnormal nodes in each primary abnormal cluster is calculated to obtain several shortest path segments. The average number of shortest path segments within the primary abnormal cluster is calculated based on the number of abnormal nodes and the number of shortest path segments to obtain several cluster densities. The resource pressure of a sector is determined by the coupling relationship between the number of abnormal nodes, the cluster density, and the planned capacity utilization rate of the airspace sector to which the primary abnormal cluster belongs, so as to obtain the resource pressure index. The congestion level is determined based on the average flight ground speed, the resource pressure index, and the time dimension characteristics of the abnormal flow nodes, so as to obtain a number of first-level risk nodes and second-level risk nodes.

5. The method for low-altitude traffic flow prediction and diversion based on spatiotemporal graph networks according to claim 4, characterized in that, The process of determining the congestion level based on the average flight ground speed, the resource pressure index, and the time-dimensional characteristics of the traffic anomaly nodes, to obtain a number of primary risk nodes and secondary risk nodes, includes: Based on the preset historical observation duration, the frequency at which the preset route node is identified as the traffic anomaly node is statistically analyzed to obtain several anomaly identification frequencies, and the standard deviation of the average flight ground speed of the traffic anomaly node is calculated to obtain several speed fluctuation values. When the resource pressure index is greater than a preset pressure threshold, the anomaly determination frequency is greater than a preset frequency threshold, and the speed fluctuation value is less than a preset fluctuation threshold, the traffic anomaly node is determined to be the first-level risk node. The abnormal traffic nodes that do not belong to the first-level risk nodes are identified as the second-level risk nodes.

6. The method for low-altitude traffic flow prediction and diversion based on spatiotemporal graph networks according to claim 5, characterized in that, The process of generating real-time mitigation suggestions based on the spatial correlation characteristics of the primary risk nodes and the conflict frequency includes: Obtain the preset route nodes within the preset range of the first-level risk node to obtain several range node sets, and record the number of abnormal traffic nodes within the range node sets to obtain several range abnormality numbers. The number of associated routes for each of the primary risk nodes is obtained to obtain a number of associated nodes, and the risk type of each primary risk node is determined based on the number of range anomalies and the number of associated nodes. The spatiotemporal coupling degree of risk is determined based on the coupling relationship between the frequency of conflicts, the number of anomalies in the range, and the number of associated nodes. Real-time traffic diversion instructions are generated based on the spatiotemporal coupling degree of the risk and the risk type to obtain the real-time diversion suggestions.

7. The method for low-altitude traffic flow prediction and mitigation based on spatiotemporal graph networks according to claim 6, characterized in that, The process of determining the risk type of each primary risk node based on the number of anomalies in the range and the number of node associations includes: When the number of associated nodes is greater than a preset association threshold and the number of abnormal ranges is less than a preset range threshold, the risk type is determined to be node overload. When the number of associated nodes is less than a preset association threshold and the number of abnormal ranges is greater than a preset range threshold, the risk type is determined to be local network saturation. When the number of associated nodes exceeds a preset association threshold and the number of abnormal ranges exceeds a preset range threshold, the risk type is determined to be hub overload that has caused regional paralysis.

8. The method for low-altitude traffic flow prediction and diversion based on spatiotemporal graph networks according to claim 7, characterized in that, The process of generating the real-time traffic diversion instruction based on the risk spatiotemporal coupling degree and the risk type to obtain the real-time diversion suggestion includes: When the risk type is node overload, the real-time traffic diversion instruction is generated as dynamic airspace reorganization; When the risk type is local network saturation, the real-time traffic diversion instruction is generated as path specification and collaborative sorting; When the risk type is that the hub overload has caused regional paralysis, the real-time traffic diversion instruction is generated as a temporary corridor and traffic regulation; The primary risk nodes are sorted in ascending order according to their risk spatiotemporal coupling degree to obtain a priority sequence; The real-time traffic diversion suggestion is formed based on the priority sequence and the corresponding real-time traffic diversion instruction.

9. The method for low-altitude traffic flow prediction and diversion based on spatiotemporal graph networks according to claim 8, characterized in that, The process of determining whether a secondary risk node will escalate to a primary risk node within the next preset risk period based on the threshold screening results of the instantaneous traffic flow and the number of flight plans, and generating predictive guidance suggestions based on the determination results, includes: Calculate the growth rate of the instantaneous flow rate at any two adjacent moments within the previous preset prediction time to obtain several instantaneous flow rate increases, and calculate the average value of all instantaneous flow rate increases to obtain the average flow rate increase. Calculate the relative deviation between the number of flight plans and the preset historical number threshold within the same preset prediction period, and record the relative deviation as the growth rate of the number of flight plans; When the average growth rate of traffic is greater than a preset growth rate threshold and the growth rate of quantity is greater than a preset growth rate threshold, it is determined that the secondary risk node will be upgraded to the primary risk node, and the secondary risk node is identified as a high-risk warning node. The secondary risk nodes that do not belong to the high-risk early warning nodes are identified as steady-state monitoring nodes; Based on the high-risk early warning node and the steady-state monitoring node, a predicted flow diversion instruction is generated to obtain the predicted diversion suggestion.

10. The method for low-altitude traffic flow prediction and diversion based on spatiotemporal graph networks according to claim 9, characterized in that, The process of generating predicted traffic diversion instructions based on the high-risk early warning nodes and the steady-state monitoring nodes to obtain the predicted diversion suggestions includes: When the secondary risk node is the high-risk early warning node, the predicted flow diversion instruction is determined to be a pre-intervention instruction; When the secondary risk node is the steady-state monitoring node, the predicted flow diversion instruction is determined to be a continuous monitoring instruction; The pre-intervention instructions, the continuous monitoring instructions, and all the secondary risk nodes are integrated and output to obtain the predictive guidance suggestions.

Images