AI-based large sports event air transportation flow prediction and guidance control method
By constructing a spatiotemporal topology graph and using hierarchical reinforcement learning methods, the challenges of air traffic management during major events were solved, enabling accurate prediction and dynamic diversion, and improving the safety and efficiency of air transport.
Patent Information
- Application Number
- CN202610805276.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-05
- Publication Date
- 2026-08-25
AI Technical Summary
Existing technologies are ill-equipped to handle the sudden surge in air travel demand caused by major sporting events, cross-regional network congestion, and the lag in static contingency plans and experience-based decision-making, leading to flight delays and airport operational paralysis.
By employing multi-source data acquisition, graph neural network prediction, and hierarchical reinforcement learning collaborative optimization methods, a spatiotemporal topology map is constructed to generate dynamic traffic management strategies, which are then combined with airspace operation rules to optimize flight scheduling.
It enables accurate traffic forecasting and control during major sporting events, improves the efficiency of airspace resource utilization, ensures the safety and smooth operation of air transport, and reduces flight delays and airspace congestion.
Smart Images
Figure CN122637644A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent air traffic management technology, specifically to an AI-based method for predicting and controlling air traffic flow for large-scale sporting events. Background Technology
[0002] Current air traffic flow management in air transport systems primarily relies on historical statistical data and the experience of air traffic controllers for prediction and scheduling. Air traffic flow management systems have applied short-term forecasting based on traditional time-series analysis methods, such as using autoregressive models or simple causal models to estimate daily traffic flow. Some advanced systems incorporate diverse information such as meteorological data and flight schedules, employing rule engines or simple machine learning models to assist decision-making. At the infrastructure level, air traffic control departments manage traffic by issuing flow control instructions, adjusting flight schedules, and setting congestion procedures. These operations largely depend on pre-established static plans and the real-time responses of controllers.
[0003] Existing technologies struggle to adapt to the sudden surges in air travel demand triggered by large-scale sporting events. Factors such as event schedules, ticket sales, and group travel lead to non-periodic, sudden peaks in traffic. Traditional models, relying on historical data, cannot effectively depict demand patterns driven by such complex social behaviors. The system cannot provide accurate medium- to long-term forecasts during the event preparation phase, impacting overall capacity planning. During the event, the simultaneous start or end of events in multiple regions can easily trigger cross-regional network congestion. Static contingency plans and experience-based decisions cannot quickly generate coordinated management solutions, leading to management delays and resource misallocation. When sudden weather events or other disruptions occur concurrently, the system lacks adaptive adjustment capabilities, potentially causing widespread flight delays and airport ground operations paralysis, severely impacting event organization and passenger travel efficiency. This deficiency in forecasting and decision-making capabilities is a core problem currently facing the technology. Summary of the Invention
[0004] To address the shortcomings of existing technologies, this invention provides an AI-based method for predicting and managing air traffic flow for large-scale sporting events. The technical problem this invention aims to solve is how to address the difficulties existing technologies face in dealing with sudden peak air traffic demand, cross-regional network congestion, and the lag in static contingency plans and experience-based decision-making caused by large-scale sporting events through multi-source data acquisition, graph neural network prediction, and hierarchical reinforcement learning collaborative optimization methods.
[0005] To achieve the above objectives, the present invention provides the following technical solution: an AI-based method for predicting and managing air traffic flow for large-scale sporting events, comprising: S1. Collect multi-source heterogeneous data of the air transport process, and standardize the multi-source heterogeneous data to form the event aviation dataset. The event aviation dataset includes airport location data, continuous flight trajectory data, airspace sectors, meteorological status data and spatiotemporal influencing factors. S2. Dynamically construct a spatiotemporal topology graph from the aviation dataset of the competition to form a spatiotemporal topology graph. The spatiotemporal topology graph includes graph nodes, graph edges, environmental attributes, and load attributes. The graph nodes include airport nodes, airspace sector nodes, and key waypoint nodes. S3. Perform graph neural network prediction on the spatiotemporal topology graph to form multi-scale traffic prediction results. The graph neural network prediction process uses a spatiotemporal graph attention network. S4. The multi-scale flow prediction results are used to perform collaborative optimization to generate a dynamic diversion strategy. The collaborative optimization is performed using a hierarchical reinforcement learning method. S5. Based on the spatiotemporal topology map and airspace operation rules, the dynamic diversion strategy is deduced and verified to generate a flight scheduling plan. The flight scheduling plan is then published to the air traffic control system for execution. Execution deviation data is collected, and the execution deviation data is used to update the parameters of the spatiotemporal graph attention network and the hierarchical reinforcement learning method to form a closed-loop control.
[0006] Preferably, the standardization process includes flight track data fusion, meteorological data spatial mapping, and event data feature extraction.
[0007] Preferably, the dynamic airspace map construction generates four-dimensional flight trajectory data by performing trajectory association calibration on continuous flight trajectory data, uses the four-dimensional flight trajectory data as graph edges, rasterizes and maps meteorological state data to form meteorological state vectors, uses the meteorological state vectors as environmental attributes, uses spatiotemporal influence factors as load attributes, uses the runway center point of the airport physical location data as airport nodes, uses the centroid coordinates of the airspace sector as airspace sector nodes, and uses the traffic convergence points of the four-dimensional flight trajectory data as key waypoint nodes.
[0008] Preferably, the graph neural network prediction includes an encoding stage and a decoding stage. The spatiotemporal graph attention network includes an encoder, a decoder, and a fully connected layer. The encoding stage includes inputting a spatiotemporal topology graph into the encoder, which extracts spatiotemporal encoded features from the nodes in the spatiotemporal topology graph using the time dimension. The decoding stage includes inputting the spatiotemporal encoded features into the decoder, which maps the spatiotemporal encoded features into a future time series in the fully connected layer. The future time series includes a predicted sequence of airport takeoffs and landings, a predicted sequence of aircraft numbers in a sector, and a predicted sequence of waypoint transit traffic. The future time series is used as the result of multi-scale traffic prediction.
[0009] Preferably, the encoder internally stacks temporal convolutional layers, and the temporal dimension extraction includes capturing the temporal evolution pattern of the spatiotemporal topology. The decoder includes a first prediction head, a second prediction head, and a third prediction head. The first prediction head maps the spatiotemporal coding features of airport nodes to airport takeoff and landing count prediction sequences, with an interval of 1 hour and a time frame within the next 24 hours. The second prediction head maps the spatiotemporal coding features of airspace sector nodes to sector aircraft count prediction sequences, with an interval of 15 minutes and a time frame within the next 6 hours. The third prediction head maps the spatiotemporal coding features of key waypoint nodes to waypoint transit traffic prediction sequences, with an interval of 5 minutes and a time frame within the next 3 hours.
[0010] Preferably, the dynamic guidance strategy includes a macro-level resource pre-allocation strategy and specific instructions, and the hierarchical reinforcement learning method includes an upper-level meta-controller and a lower-level executor.
[0011] Preferably, the upper-level meta-controller outputs a macro-resource pre-allocation strategy with an hourly decision cycle based on the airport takeoff and landing count prediction sequence in the multi-scale traffic prediction results, and the lower-level actuator generates specific instructions with a minute-based decision cycle based on the sector aircraft number prediction sequence and waypoint transit traffic prediction sequence in the multi-scale traffic prediction results, under the constraint of the macro-resource pre-allocation strategy.
[0012] Preferably, the macro-resource pre-allocation strategy includes a flight schedule, flight core capacity control parameters, and a target airspace sector list, and the specific instructions include flight rerouting instructions, flight delay instructions, and flight sequencing instructions.
[0013] Preferably, the airspace operation rules are obtained based on the standard operating procedures and interval standards of air traffic control. The simulation verification includes simulating the implementation adaptability of the dynamic diversion strategy under different complex scenario variables to generate adaptive results, comparing the analysis of the adaptive results with the preset target values to generate strategy simulation results, and calibrating the dynamic diversion strategy based on the strategy simulation results and multi-scale traffic prediction results to generate flight scheduling schemes.
[0014] Preferably, the preset target value is set according to airspace capacity standards and event support requirements, and the complex scenario variables include spectator traffic, airline flight schedules, sudden weather changes, waypoint density, and airspace resource occupancy.
[0015] This invention provides an AI-based method for predicting and managing air traffic flow during large-scale sporting events. It offers the following advantages: This AI-based method for predicting and controlling air traffic flow for large-scale sporting events collects multi-source heterogeneous data during air transport and standardizes the data. The standardization process integrates and extracts features from flight track data, meteorological data, and event data to form an event aviation dataset, enabling the prediction and control of air traffic flow for large-scale sporting events.
[0016] This invention employs spatiotemporal topology graphs and graph neural networks for multi-scale traffic flow prediction, combined with a spatiotemporal graph attention network for encoding and decoding, generating accurate traffic flow prediction results. A hierarchical reinforcement learning method is introduced to collaboratively optimize the prediction results, formulating dynamic traffic management strategies, including macro-level resource pre-allocation and specific operational instructions. The method has been validated under complex scenarios, adjusting flight scheduling schemes according to actual conditions, improving airspace resource utilization efficiency, and ensuring safe and smooth air transport during events. Attached Figure Description
[0017] Figure 1 This is a flowchart illustrating the overall process of implementing an invention. Figure 2 It is a flowchart of data acquisition and standardized processing for implementing an invention; Figure 3 This is a schematic diagram of a graph neural network prediction model for implementing the invention; Figure 4 It is a flowchart of hierarchical reinforcement learning collaborative optimization for realizing an invention; Figure 5 It is a flowchart for the deduction, verification, and closed-loop control of an invention. Detailed Implementation
[0018] 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.
[0019] Example 1 like Figure 1-5 As shown, this embodiment of the invention provides an AI-based method for predicting and managing air traffic flow for large-scale sporting events, including: S1. Collect multi-source heterogeneous data from the air transport process, and standardize the multi-source heterogeneous data to form the event aviation dataset. The event aviation dataset includes airport location data, continuous flight trajectory data, airspace sectors, meteorological status data, and spatiotemporal influencing factors. Standardization processing includes flight trajectory data fusion, spatial mapping of meteorological data, and feature extraction of event data.
[0020] S2. Dynamic airspace graph construction is performed on the event aviation dataset to form a spatiotemporal topology graph. The spatiotemporal topology graph includes graph nodes, graph edges, environmental attributes, and load attributes. Graph nodes include airport nodes, airspace sector nodes, and key waypoint nodes. Dynamic airspace graph construction involves trajectory association calibration of continuous flight trajectory data to generate four-dimensional flight trajectory data. The four-dimensional flight trajectory data is used as graph edges. Meteorological state data is rasterized and mapped to form meteorological state vectors, which are used as environmental attributes. Spatiotemporal influence factors are used as load attributes. The runway center point of the airport physical location data is used as airport nodes. The centroid coordinates of the airspace sector are used as airspace sector nodes. The traffic convergence points of the four-dimensional flight trajectory data are used as key waypoint nodes.
[0021] S3. Graph neural network (GNN) prediction is performed on the spatiotemporal topology map to generate multi-scale traffic prediction results. The GNN prediction process uses a spatiotemporal graph attention network. The GNN prediction includes an encoding stage and a decoding stage. The spatiotemporal graph attention network includes an encoder, a decoder, and a fully connected layer. The encoding stage involves inputting the spatiotemporal topology map into the encoder, which extracts the time dimension of the nodes in the spatiotemporal topology map and outputs spatiotemporal encoded features. The decoding stage involves inputting the spatiotemporal encoded features into the decoder, which maps the spatiotemporal encoded features into future time series in the fully connected layer. The future time series includes airport takeoff and landing prediction sequences, sector aircraft number prediction sequences, and waypoint transit traffic prediction sequences. The future time series are used as the multi-scale traffic prediction results. The encoder contains stacked temporal convolutional layers. Temporal dimension extraction includes capturing the temporal evolution pattern of the spatiotemporal topology. The decoder includes a first prediction head, a second prediction head, and a third prediction head. The first prediction head maps the spatiotemporal coding features of airport nodes to airport takeoff and landing count prediction sequences with an interval of 1 hour and a time frame within the next 24 hours. The second prediction head maps the spatiotemporal coding features of airspace sector nodes to sector aircraft count prediction sequences with an interval of 15 minutes and a time frame within the next 6 hours. The third prediction head maps the spatiotemporal coding features of key waypoint nodes to waypoint transit traffic prediction sequences with an interval of 5 minutes and a time frame within the next 3 hours.
[0022] S4. A dynamic diversion strategy is generated by collaboratively optimizing the multi-scale traffic forecast results. This collaborative optimization employs a hierarchical reinforcement learning method. The dynamic diversion strategy includes a macro-level resource pre-allocation strategy and specific instructions. The hierarchical reinforcement learning method comprises an upper-level meta-controller and a lower-level actuator. The upper-level meta-controller outputs the macro-level resource pre-allocation strategy based on the airport takeoff and landing prediction sequence from the multi-scale traffic forecast results, with a decision cycle of hours. The lower-level actuator generates specific instructions with a decision cycle of minutes, based on the sector aircraft quantity prediction sequence and waypoint transit traffic prediction sequence from the multi-scale traffic forecast results, under the constraints of the macro-level resource pre-allocation strategy. The macro-level resource pre-allocation strategy includes flight schedules, core flight capacity control parameters, and a target airspace sector list. Specific instructions include flight rerouting instructions, flight delay instructions, and flight reordering instructions.
[0023] S5. Based on the spatiotemporal topology map and airspace operation rules, the dynamic traffic management strategy is simulated and verified, a flight scheduling plan is generated, and the flight scheduling plan is published to the air traffic control system for execution. Execution deviation data is collected, and the execution deviation data is used to update the parameters of the spatiotemporal map attention network and hierarchical reinforcement learning method to form a closed-loop control. The airspace operation rules are obtained based on the standard operating procedures and interval standards of air traffic control. The simulation and verification include simulating the implementation adaptability of the dynamic traffic management strategy under different complex scenario variables to generate adaptive results. The adaptive results are compared with the preset target values to generate strategy simulation results. Based on the strategy simulation results and multi-scale traffic prediction results, the dynamic traffic management strategy is calibrated to generate a flight scheduling plan. The preset target values are set according to airspace capacity standards and event support requirements. Complex scenario variables include spectator traffic, airline flight schedules, sudden weather changes, waypoint density, and airspace resource occupancy.
[0024] This invention accurately predicts air traffic flow and dynamically adjusts traffic management strategies to improve air transport efficiency. It optimizes airspace resource allocation using spatiotemporal topology maps and hierarchical reinforcement learning methods to avoid resource waste. Global and local collaborative optimization of air traffic flow ensures flexible and precise flight scheduling. The system adapts to complex scenario changes, ensuring smooth air transport under various variables and enhancing its adaptability and robustness.
[0025] AI-based methods for predicting and managing air traffic flow during large-scale sporting events possess significant emergency response capabilities and are suitable for large-scale air events. In the event of emergencies, the system rapidly adjusts flight schedules and airspace management to ensure the safety and smooth operation of air transport during the event. This reduces flight delays and airspace congestion, lowers fuel consumption and ground waiting times, reduces transportation costs, and improves passenger experience. A reinforcement learning closed-loop control mechanism optimizes the prediction model and decision-making strategies based on execution deviation data, improving prediction accuracy and the intelligence level of scheduling strategies.
[0026] Example 2 This embodiment presents an AI-based method for predicting and managing air traffic flow during major sporting events. It applies a spatiotemporal graph attention network to predict and manage air traffic flow during such events, aiming to optimize air traffic management and reduce flight delays and airspace congestion. The specific implementation method is as follows: 1. Data collection and standardization processing This example uses air traffic volume during a major sporting event held in Shanghai in December 2025 as the forecast target. Data sources include air traffic management systems, meteorological monitoring systems, flight management systems, and airspace management data. Actual data is as follows: Airport location data: Shanghai Pudong Airport is node 1, with location coordinates (31.2304°N, 121.4737°E), and Beijing Capital Airport is node 2, with location coordinates (39.9042°N, 116.4074°E).
[0027] Flight trajectory data: The data comes from the flight management system and provides real-time flight trajectory data. The following are actual flight takeoff and landing data: Flight 1 is Shanghai-Beijing: The flight departs from Shanghai Pudong International Airport at 8:00 AM on December 1, 2025 (31.2304°N, 121.4737°E), passing through (32.1000, 121.6000) and (33.2000, 121.7000), and finally arrives at Beijing Capital International Airport at 10:30 AM on December 1, 2025.
[0028] Flight 2 Beijing-Shanghai: The flight departs from Beijing Capital International Airport (39.9042, 116.4074) at 09:00 on December 1, 2025, passing through (36.8000, 118.5000) and (34.4000, 121.4000), and finally arrives at Shanghai Pudong International Airport (31.2304°N, 121.4737°E), with a landing time of 11:30 on December 1, 2025.
[0029] Flight quantity data: According to data provided by the flight management system, Shanghai Pudong Airport will have an average of about 1,800 takeoffs and landings per day on December 1, 2025, while Beijing Capital Airport will have about 2,000.
[0030] Airspace sector data: The air traffic management system provides the following information: the southeast airspace is node 3, and the northwest airspace is node 4. The number of aircraft is 50 and 30 respectively.
[0031] Meteorological status data: Meteorological data provided by the weather forecast systems of the Shanghai Meteorological Bureau and the Beijing Meteorological Bureau show a wind speed of 10 m / s, a temperature of 15°C, and precipitation of 0 mm.
[0032] Spatiotemporal influencing factors related to the event: According to the traffic forecast provided by the event organizers, the traffic impact factor for the event is 1. Since the air traffic at Shanghai Pudong Airport is expected to increase by 20% during the event, the traffic adjustment coefficient is 1.2.
[0033] 2. Construction of spatiotemporal topology graph Based on the collected data, a spatiotemporal topology graph is constructed. The specific data for nodes and edges are as follows: Table 1: Node and edge data table.
[0034]
[0035] Key Waypoints: Waypoint 1 is a waypoint between Shanghai and Beijing, located at (35.0000°N, 120.0000°E). The flow of traffic passing through this waypoint is 30 flights. Edge definition: Connecting Shanghai Pudong Airport with the southeast airspace indicates that a flight departs from Shanghai Pudong Airport and passes through the southeast airspace.
[0036] Connecting Beijing Capital International Airport with the northwest airspace indicates that a flight departs from Beijing Capital International Airport and passes through the northwest airspace.
[0037] The weight of each edge is calculated based on a combination of factors, including the number of flights, flight time, and weather conditions. The weight of the edge for flights from Shanghai Pudong Airport to the southeast airspace is weighted according to the number of flights and weather conditions.
[0038] 3. Applications of Spatiotemporal Graph Attention Networks Encoding phase: The node and edge information of the spatiotemporal topology graph is input into the encoder of the spatiotemporal graph attention network. Each node contains features such as time, space, traffic conditions, and weather conditions.
[0039] The encoder extracts the spatiotemporal features of each node through graph convolution operations and uses graph attention mechanism to calculate the spatiotemporal relationship between nodes.
[0040] Decoding stage: The encoded spatiotemporal features are input into the decoder to generate future traffic prediction results.
[0041] Table 2: Forecast of takeoff and landing traffic at Shanghai Pudong Airport in the next 24 hours.
[0042]
[0043] Table 3: Forecast of the number of aircraft in the southeast airspace sector in the next 6 hours.
[0044]
[0045] Table 4: Forecast of Shanghai-Beijing waypoint traffic flow in the next 3 hours.
[0046]
[0047] 4. Output and Application Multi-scale traffic prediction results were obtained through prediction using a spatiotemporal graph attention network: Airport takeoff and landing frequency forecast: In the next 24 hours, the number of takeoffs and landings at Shanghai Pudong Airport will gradually increase from 150 to 160, taking into account the increased traffic caused by the event.
[0048] Aircraft number forecast for the airspace sector: In the next 6 hours, the number of aircraft in the southeast airspace will increase from 50 to 58.
[0049] Forecast of transit traffic at waypoints: In the next 3 hours, transit traffic at the Shanghai-Beijing waypoint is expected to increase from 32 flights to 35 flights.
[0050] By applying spatiotemporal graph attention networks, air traffic flow prediction based on real data was achieved, accurately predicting takeoffs and landings, the number of aircraft in airspace sectors, and waypoint traffic at Shanghai Pudong Airport and Beijing Capital Airport. Considering the factors influencing event traffic, the prediction results provide support for dynamic scheduling and traffic management strategies, helping to optimize airspace resource allocation and providing a scientific basis for flight scheduling and air traffic management, thus reducing airspace congestion and flight delays during events.
[0051] Example 3 This embodiment presents an AI-based method for predicting and managing air traffic flow during major sporting events. It optimizes the scheduling of air traffic flow during such events, ensuring efficient utilization of airspace resources and on-time flight takeoffs and landings. The specific implementation method is as follows: 1. Data Collection and Preparation In this embodiment, the data used comes from the following sources: Historical flight data: According to flight scheduling data provided by Y Airport, this data covers flight traffic during a major international sporting event over the past month. Specific information includes flight takeoffs and landings, flight schedules, and airspace usage.
[0052] Weather and meteorological data: Weather forecast data for the event, including wind speed, precipitation, and visibility, was obtained from the China Meteorological Administration.
[0053] Event schedule data: The event organizing committee provides flight schedules related to the event, including departure and arrival times for key event flights and VIP flights. Specific details include the number of event-related flights and the priority arrangement for key flights.
[0054] Airspace resource distribution and operation rules: Information such as airspace distribution, airspace capacity, airspace division, and waypoint density provided by the air traffic control system.
[0055] 2. Input data Based on historical data and event schedules, the following specific prediction data are obtained: Airport takeoff and landing count prediction sequence: Based on historical flight data of Airport Y and flight schedules during the event, predict the number of takeoffs and landings at the airport in the next 24 hours. The predicted data is as follows: Table 5: Airport Takeoff and Landing Number Forecast Sequence Table.
[0056]
[0057] Aircraft Count Prediction Sequence for Airspace Sectors: Based on airspace traffic analysis and airspace capacity data, the number of aircraft in each airspace sector is predicted, as shown in the following data: There were 50 flights in airspace sector 1, 65 flights in airspace sector 2, 55 flights in airspace sector 3, and 45 flights in airspace sector 4.
[0058] Waypoint-to-way traffic forecast sequence: Based on historical flight data and airspace density analysis, the transit traffic at waypoints is predicted, as shown in the following data: There were 18 flights to key waypoint 1, 20 flights to key waypoint 2, 22 flights to key waypoint 3, and 15 flights to key waypoint 4.
[0059] 3. Application of hierarchical reinforcement learning methods Upper-level meta controller: Based on the predicted number of airport takeoffs and landings, the upper-level controller formulates a macro-level resource allocation strategy. Considering the special nature of flights during the event, priority is given to ensuring the smooth operation of flights related to the event.
[0060] During time slot 1, the airport had 160 takeoffs and landings. The controller determined: Increase runway scheduling frequency: Increase takeoff and landing frequency to 8 takeoffs and landings every 15 minutes to avoid congestion.
[0061] Prioritize flights for the event: Give priority to flights related to the event to ensure their timely take-off and landing.
[0062] Lower-level actuator: Based on the predicted airspace sector traffic and waypoint-to-waypoint traffic, the lower-level actuator generates specific diversion instructions.
[0063] The number of aircraft in airspace sector 1 is 50, exceeding the airspace capacity by 40. Therefore, the actuator decides: Flight diversion instruction: Some flights will be diverted to airspace sector 2 to avoid excessive airspace congestion.
[0064] The throughput at critical waypoint 1 was 18 flights, exceeding the set throughput threshold of 15 flights. Therefore, the actuator decided: Flight diversion instruction: Flight A was originally scheduled to pass through key waypoint 1, but will now detour to key waypoint 2. The estimated detour time is 4 minutes.
[0065] Flight B Delay Instruction: Delay for 5 minutes near critical waypoint 1 to ensure traffic balance.
[0066] 4. Generation of dynamic diversion strategies Macro-level resource pre-allocation: Based on the macro-resource pre-allocation strategy formulated by the upper-level meta-controller, the airport has 160 takeoffs and landings, and the decision is made on how to allocate resources reasonably.
[0067] For time period 1, based on 160 takeoffs and landings, the meta controller decides: Increase runway scheduling frequency: schedule 8 flights to take off and land every 15 minutes.
[0068] Priority will be given to flights involved in the event: dedicated channels and priority take-off and landing rights will be provided to ensure the smooth take-off and landing of flights involved in the event.
[0069] Specific instruction generation: Based on the prediction results of the lower-level actuator, specific flight scheduling instructions are generated.
[0070] Flight rerouting instruction: For the transit traffic at critical waypoint 1, the actuator issues the following instruction: Flight A will detour to key waypoint 2, with an estimated detour time of 4 minutes. Flight B will detour to key waypoint 3, with an estimated detour time of 6 minutes.
[0071] Flight Delay Instruction: Due to the flow of traffic in Department 1 exceeding airspace capacity, the actuator issues an instruction: Flight C is postponed to time slot 2 to avoid excessive airspace congestion.
[0072] Flight sequencing instructions: Adjust the takeoff and landing order of flights based on flight priority and airspace resource utilization. Flight D, being an emergency flight, will be given priority for takeoff and landing; Flight E, being a non-emergency flight, will be delayed.
[0073] 5. Implementation and Feedback Implementation and Monitoring: The air traffic control system implements flight scheduling based on dynamic traffic management strategies and monitors flight takeoffs and landings, airspace resource usage, and waypoint traffic in real time.
[0074] During time period 1, the actual number of takeoffs and landings was 165, slightly more than the predicted 160. The system detected that Flight A was delayed by 4 minutes after being rerouted to the critical waypoint 2 due to a rerouting instruction, and Flight B was delayed by 5 minutes.
[0075] Feedback and updates: The collected execution deviation data, showing that Flight A was delayed by 4 minutes due to a detour, was used to update the parameters in the spatiotemporal graph attention network and the hierarchical reinforcement learning model.
[0076] Weights are adjusted for nodes such as waypoints and airspace sectors in the spatiotemporal topology map to reduce future prediction errors.
[0077] Update the policies in the reinforcement learning model to make them more accurate and avoid future resource misallocation and congestion.
[0078] 6. Implementation Results and Effects Airspace and runway resources were effectively allocated: through accurate forecasting and flexible scheduling, airspace and runway resources were rationally distributed, avoiding excessive congestion.
[0079] Flight punctuality improved: Flights related to the event were given priority, ensuring that important flights took off and landed on time, while delays of non-event flights were effectively controlled.
[0080] Deviation data supports optimization: The collected execution deviation data provides a strong basis for the subsequent optimization of artificial intelligence models used to process and optimize air traffic flow forecasting and control, improving the accuracy and efficiency of scheduling strategies.
[0081] In summary, accurate prediction and real-time adjustment of airport takeoffs and landings, airspace traffic, and waypoint traffic ensured efficient flight scheduling and rational allocation of airspace resources. The use of spatiotemporal graph attention networks and hierarchical reinforcement learning techniques addressed sudden traffic changes, optimized flight takeoff and landing sequences and airspace utilization, and improved flight punctuality. Through a feedback mechanism, the system continuously optimized its prediction model, enhanced its adaptive capabilities, and improved the accuracy of scheduling strategies, providing technical support for air transport management in the future for similar large-scale events.
[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. An AI-based method for predicting and managing air traffic flow during large-scale sporting events, characterized in that: include: S1. Collect multi-source heterogeneous data of the air transport process, and standardize the multi-source heterogeneous data to form the event aviation dataset. The event aviation dataset includes airport location data, continuous flight trajectory data, airspace sectors, meteorological status data and spatiotemporal influencing factors. S2. Dynamically construct a spatiotemporal topology graph from the aviation dataset of the competition to form a spatiotemporal topology graph. The spatiotemporal topology graph includes graph nodes, graph edges, environmental attributes, and load attributes. The graph nodes include airport nodes, airspace sector nodes, and key waypoint nodes. S3. Perform graph neural network prediction on the spatiotemporal topology graph to form multi-scale traffic prediction results. The graph neural network prediction process uses a spatiotemporal graph attention network. S4. The multi-scale flow prediction results are used to perform collaborative optimization to generate a dynamic diversion strategy. The collaborative optimization is performed using a hierarchical reinforcement learning method. S5. Based on the spatiotemporal topology map and airspace operation rules, the dynamic diversion strategy is deduced and verified to generate a flight scheduling plan. The flight scheduling plan is then published to the air traffic control system for execution. Execution deviation data is collected, and the execution deviation data is used to update the parameters of the spatiotemporal graph attention network and the hierarchical reinforcement learning method to form a closed-loop control.
2. The AI-based method for predicting and managing air traffic flow for large-scale sporting events according to claim 1, characterized in that: The standardized processing includes flight track data fusion, spatial mapping of meteorological data, and feature extraction of event data.
3. The AI-based method for predicting and managing air traffic flow for large-scale sporting events according to claim 1, characterized in that: The dynamic airspace map construction generates four-dimensional flight trajectory data by performing trajectory association calibration on continuous flight trajectory data. The four-dimensional flight trajectory data is used as graph edges, and the meteorological state data is rasterized and mapped to form a meteorological state vector. The meteorological state vector is used as an environmental attribute, the spatiotemporal influence factor is used as a load attribute, the runway center point of the airport physical location data is used as an airport node, the centroid coordinates of the airspace sector are used as airspace sector nodes, and the traffic convergence points of the four-dimensional flight trajectory data are used as key waypoint nodes.
4. The AI-based method for predicting and managing air traffic flow for large-scale sporting events according to claim 1, characterized in that: The graph neural network prediction includes an encoding stage and a decoding stage. The spatiotemporal graph attention network includes an encoder, a decoder, and a fully connected layer. The encoding stage includes inputting a spatiotemporal topology graph into the encoder, which extracts spatiotemporal encoded features from the nodes in the spatiotemporal topology graph using the time dimension. The decoding stage includes inputting the spatiotemporal encoded features into the decoder, which maps the spatiotemporal encoded features into future time series in the fully connected layer. The future time series includes airport takeoff and landing prediction series, sector aircraft number prediction series, and waypoint transit traffic prediction series. The future time series is used as the multi-scale traffic prediction result.
5. The AI-based method for predicting and managing air traffic flow for large-scale sporting events according to claim 4, characterized in that: The encoder internally stacks temporal convolutional layers. The temporal dimension extraction includes capturing the temporal evolution pattern of the spatiotemporal topology. The decoder includes a first prediction head, a second prediction head, and a third prediction head. The first prediction head maps the spatiotemporal coding features of airport nodes to airport takeoff and landing count prediction sequences with an interval of 1 hour and a time frame within the next 24 hours. The second prediction head maps the spatiotemporal coding features of airspace sector nodes to sector aircraft count prediction sequences with an interval of 15 minutes and a time frame within the next 6 hours. The third prediction head maps the spatiotemporal coding features of key waypoint nodes to waypoint transit traffic prediction sequences with an interval of 5 minutes and a time frame within the next 3 hours.
6. The AI-based method for predicting and managing air traffic flow for large-scale sporting events according to claim 1, characterized in that: The dynamic guidance strategy includes a macro-level resource pre-allocation strategy and specific instructions, and the hierarchical reinforcement learning method includes an upper-level meta-controller and a lower-level executor.
7. The AI-based method for predicting and managing air traffic flow for large-scale sporting events according to claim 6, characterized in that: The upper-level meta-controller outputs a macro-resource pre-allocation strategy with an hourly decision cycle based on the airport takeoff and landing count prediction sequence in the multi-scale traffic prediction results. The lower-level actuator generates specific instructions with a minute-based decision cycle based on the sector aircraft number prediction sequence and waypoint transit traffic prediction sequence in the multi-scale traffic prediction results, under the constraint of the macro-resource pre-allocation strategy.
8. The AI-based method for predicting and managing air traffic flow for large-scale sporting events according to claim 6, characterized in that: The macro-resource pre-allocation strategy includes flight schedules, core flight capacity control parameters, and a target airspace sector list. The specific instructions include flight rerouting instructions, flight delay instructions, and flight reordering instructions.
9. The AI-based method for predicting and managing air traffic flow for large-scale sporting events according to claim 1, characterized in that: The airspace operation rules are obtained based on the standard operating procedures and interval standards of air traffic control. The simulation verification includes simulating the implementation of dynamic diversion strategies under different complex scenario variables to generate adaptive results, comparing the analysis of adaptive results with preset target values to generate strategy simulation results, and calibrating the dynamic diversion strategies based on strategy simulation results and multi-scale traffic prediction results to generate flight scheduling schemes.
10. The AI-based method for predicting and managing air traffic flow for large-scale sporting events according to claim 9, characterized in that: The preset target values are set according to airspace capacity standards and event support requirements. The complex scenario variables include spectator traffic, airline flight schedules, sudden weather changes, waypoint density, and airspace resource occupancy.