Flight flow intelligent prediction and air traffic collaborative optimization system based on big data

Abstract

The invention discloses a flight flow intelligent prediction and air traffic collaborative optimization system based on big data, and relates to the technical field of air traffic management. Comprising the steps that a data acquisition module acquires multi-source data from a civil aviation database and aligns the multi-source data to generate a fusion data matrix; the traffic prediction module extracts traffic feature vectors, generates a feature similarity matrix by calculating distribution differences, and generates a basic traffic prediction result based on migration prediction model parameters; the disturbance correction module detects a sudden disturbance event based on real-time weather and airspace state data and corrects a basic prediction result; the capacity adjusting module generates a capacity adjusting scheme when the capacity deviation value exceeds a threshold value according to the correction result and the airport operation capability data; and the collaborative optimization module finally generates a multi-airport collaborative scheduling scheme through a game equilibrium algorithm based on the correction result, the capacity scheme and the regional coordination data. According to the invention, the accuracy of flight flow prediction and the efficiency of multi-airport collaborative management are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of air traffic management technology, specifically to a big data-based intelligent prediction system for flight traffic and a collaborative optimization system for air traffic. Background Technology

[0002] Air traffic management is central to ensuring the safety and efficiency of civil aviation operations, and its key lies in the accurate prediction of flight traffic and the coordinated allocation of airspace resources. In recent years, with the help of big data and artificial intelligence technologies, this field is developing towards intelligence and self-adaptation.

[0003] Existing technologies have numerous solutions dedicated to predicting short-term traffic flow and formulating scheduling plans by analyzing historical flight data, real-time weather information, and airport capacity. These methods are typically modeled based on independent airspace units or airports, providing valuable references under normal operating conditions and alleviating congestion and improving operational efficiency to some extent. However, these methods generally lack in-depth consideration of the dynamic correlations between nodes within a large-scale airspace network and the collaborative impact of sudden disturbances. Specifically, existing prediction models often view the traffic flow of each airport in isolation, failing to effectively capture the implicit dependencies between different airports caused by route connections and traffic interactions, resulting in insufficient accuracy of predictions in complex networked operating environments. Furthermore, when disturbances such as thunderstorms or military exercises occur, existing systems struggle to quickly and quantitatively assess their cross-regional cascading effects, leading to discrepancies between local adjustment schemes and the global optimal solution. These limitations mean that existing technologies require improvement in overall efficiency and robustness when facing large-scale, networked air traffic coordination optimization, and cannot fully meet the needs of integrated and intelligent operation of modern air traffic systems. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a big data-based intelligent flight flow prediction and air traffic collaborative optimization system.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] This invention discloses a big data-based intelligent flight flow prediction and air traffic collaborative optimization system, comprising:

[0007] The data acquisition module is used to obtain flight operation data, meteorological and airspace status data, airport operation capacity data and regional flight coordination data from the civil aviation database, and align them according to timestamps and spatial coordinates to generate a fused data matrix;

[0008] The traffic prediction module is used to extract traffic feature vectors of each airport from the fused data matrix, calculate the distribution differences of the traffic feature vectors between different airports to generate a feature similarity matrix, and transfer the prediction model parameters based on the feature similarity matrix to generate basic traffic prediction results.

[0009] The disturbance correction module is used to detect sudden disturbance events based on the real-time acquired meteorological and airspace status data, and to correct the basic traffic forecast results to generate corrected traffic forecast results.

[0010] The capacity adjustment module is used to calculate the capacity deviation value of each airspace sector based on the corrected traffic forecast result and the airport operation capacity data, and generate a capacity adjustment plan when the capacity deviation value exceeds a preset deviation threshold.

[0011] The collaborative optimization module is used to generate a multi-airport collaborative scheduling scheme based on the corrected traffic forecast results, the capacity adjustment scheme, and the regional flight coordination data through a game equilibrium algorithm.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0013] 1. This invention establishes a unified data foundation for high-precision analysis through spatiotemporal alignment and fusion of multi-source data. Furthermore, it utilizes a feature similarity matrix constructed from the distribution differences among traffic feature vectors to achieve intelligent migration of prediction model parameters between associated airports. This overcomes the limitations of isolated predictions for each airport in traditional models, enabling the keen capture of implicit traffic correlations and propagation patterns within the airspace network, thereby fundamentally improving the accuracy and generalization ability of basic traffic prediction results.

[0014] 2. This invention can proactively detect and quantify the impact of sudden disturbances based on dynamically changing meteorological and airspace conditions. It not only relies on historical patterns for prediction but also responds to instantaneous changes in the external environment, making timely and targeted corrections to the basic prediction results. This effectively enhances the system's robustness in the face of uncertainties such as thunderstorms and airspace control, ensuring consistency between prediction results and actual operational situations, and providing a more reliable basis for subsequent decision-making.

[0015] 3. This invention achieves integrated decision-making for airspace resource allocation and flight scheduling through dynamic assessment of capacity deviation and collaborative optimization of game equilibrium. The capacity adjustment scheme is triggered by accurate corrective predictions, avoiding blind resource scheduling; while treating each airport as a game participant for collaborative scheduling effectively balances the interests of multiple parties and seeks the optimal solution at the global level. This design tightly couples local capacity adjustment with global traffic allocation, improving the overall utilization efficiency of airspace resources and the smoothness of multi-airport collaborative operation. Attached Figure Description

[0016] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts. Wherein:

[0017] Figure 1 This is a system module connection diagram of the present invention;

[0018] Figure 2 This is a flowchart of the system modules of the present invention;

[0019] Figure 3 This is a flowchart illustrating the process of correcting the basic flow prediction results according to the present invention.

[0020] Figure 4 This is a flowchart illustrating the steps of the multi-airport collaborative scheduling scheme of the present invention. Detailed Implementation

[0021] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0022] Application Overview:

[0023] In existing technologies, flight traffic forecasting and airspace collaborative management largely rely on independent historical data analysis or static capacity models, which are insufficient to effectively address the dynamic correlations and sudden disturbances in large-scale airspace networks. Traditional methods, when dealing with multi-airport coordination issues, often treat each airport as an independent unit for forecasting and scheduling, failing to fully consider the implicit dependencies between airports arising from route connections and traffic interactions. This results in insufficient accuracy of forecasts in complex networked operating environments. Furthermore, when disturbances such as thunderstorms or military exercises occur, existing systems lack quantitative assessment and collaborative response mechanisms for the cross-regional cascading effects of disturbances. This leads to discrepancies between local adjustment plans and overall operational efficiency, failing to meet the integrated and intelligent operational requirements of modern air traffic systems.

[0024] To address the aforementioned issues, the study found an intrinsic correlation between the differences in traffic characteristic vector distribution among different airports and the overall airspace operational status. By establishing a model parameter transfer mechanism based on feature similarity, collaborative prediction at the airspace network level can be achieved. The research revealed that dynamically correcting the basic prediction results based on real-time disturbance events, combined with capacity deviation assessment triggering collaborative optimization, can improve the system's adaptability and robustness. Further system validation demonstrated that applying game equilibrium theory to multi-airport resource allocation decisions can effectively balance the operational interests of all parties and generate the optimal scheduling plan at the global level.

[0025] After introducing the basic concept of the present invention, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0026] Example:

[0027] like Figure 1 As shown, the big data-based intelligent flight flow prediction and air traffic collaborative optimization system includes:

[0028] The data acquisition module is used to obtain flight operation data, meteorological and airspace status data, airport operation capacity data and regional flight coordination data from the civil aviation database, and align them according to timestamps and spatial coordinates to generate a fused data matrix;

[0029] The traffic prediction module is used to extract traffic feature vectors of each airport from the fused data matrix, calculate the distribution differences of traffic feature vectors between different airports to generate a feature similarity matrix, and transfer the prediction model parameters based on the feature similarity matrix to generate basic traffic prediction results.

[0030] The disturbance correction module is used to detect sudden disturbance events based on real-time acquired meteorological and airspace status data, correct the basic flow forecast results, and generate corrected flow forecast results.

[0031] The capacity adjustment module is used to calculate the capacity deviation value of each airspace sector based on the corrected traffic forecast results and airport operation capacity data, and generate a capacity adjustment plan when the capacity deviation value exceeds the preset deviation threshold.

[0032] The collaborative optimization module is used to generate multi-airport collaborative scheduling schemes based on corrected traffic forecast results, capacity adjustment plans, and regional flight coordination data, using a game equilibrium algorithm.

[0033] like Figure 2As shown, the working principle of this application is as follows: The data acquisition module obtains flight operation data, meteorological and airspace status data, airport operation capacity data, and regional flight coordination data in real time from heterogeneous data sources such as civil aviation meteorological systems, air traffic control radar systems, and airport operation databases through data interfaces. It then calls the built-in data alignment engine to perform spatiotemporal alignment and fusion of all multi-source asynchronous data based on GPS timestamps and geospatial coordinates (such as latitude and longitude), ultimately generating a fused data matrix under a unified spatiotemporal framework.

[0034] Subsequently, the traffic prediction module runs a feature extraction algorithm from the fused data matrix to quantify the traffic intensity, heading distribution, and other characteristics of each airport within a specific time period, forming its traffic feature vector. Further calculations are made to determine the distribution differences of these feature vectors among different airports, constructing a feature similarity matrix that reflects the similarity of traffic patterns among nodes within the airspace network. Based on this feature similarity matrix, the system uses a parameter transfer engine to transfer the parameters of a well-trained prediction model from one airport to other airports with similar traffic patterns, thereby quickly generating a highly accurate basic traffic prediction result and effectively utilizing the overall characteristics of the airspace network.

[0035] The disturbance correction module continuously monitors the real-time data stream and uses its integrated disturbance event detector to identify phenomena such as thunderstorm area expansion in weather radar data and newly added closed areas in airspace status data. Once a sudden disturbance event is detected, the disturbance correction module activates a correction algorithm to assess the impact range and duration of the sudden disturbance event, and accordingly performs a weighted correction on the basic traffic forecast results, outputting a corrected traffic forecast result that better reflects the real-time situation.

[0036] Based on this, the capacity adjustment module compares the corrected forecast results with the inherent capacity (such as runway hourly capacity and maximum number of aircraft per sector) in the airport's operational capacity database, and calculates the capacity deviation between the forecasted traffic and the maximum carrying capacity of each airspace sector. When the capacity deviation exceeds a preset deviation threshold (within a safe range), the decision logic within the module automatically generates a capacity adjustment plan.

[0037] Finally, the collaborative optimization module inputs the corrected traffic forecast results, capacity adjustment plans, and regional flight coordination data (such as delay cost parameters negotiated by various airlines) into its game equilibrium solver. This game equilibrium solver simulates each airport as a game participant, constructs its payoff function, and calculates the Nash equilibrium point through an iterative algorithm. Based on this, it outputs a multi-airport collaborative scheduling scheme that balances the interests of all parties and achieves global approximate optimality.

[0038] At the hardware level, this system relies on server clusters deployed on cloud platforms or air traffic control data centers. Its core includes distributed storage and computing units for massive data storage and processing, network communication interfaces responsible for real-time data reception, and high-performance computing modules that support the operation of algorithm models.

[0039] The prediction model uses a three-layer network structure with 128 hidden layer units. The Adam optimizer is used during training with an initial learning rate of 0.001, a batch size of 32, and a total of 500 training iterations. Early stopping is performed on the validation set to prevent overfitting.

[0040] Through the aforementioned technical solutions, this application enhances the overall utilization efficiency of airspace resources. By accurately predicting traffic flow and dynamically allocating resources, it effectively alleviates congestion at key nodes. In the face of unforeseen circumstances such as sudden weather changes or airspace closures, it generates and executes recovery plans to ensure operational stability. Through globally optimized collaborative decision-making, a balance of interests is found among multiple participants, thereby improving flight punctuality while reducing operating costs and coordination burdens. This provides a more intelligent, reliable, and economical modern management paradigm for air traffic.

[0041] This application further proposes that the data obtained from the civil aviation database, including flight operation data, meteorological and airspace status data, airport operation capacity data, and regional flight coordination data, specifically include:

[0042] Flight operation data includes flight number, scheduled time, departure and arrival airports, waypoint sequence, and delay duration;

[0043] Meteorological and airspace status data include thunderstorm areas, wind speed and direction, airspace closure areas, and military exercise notices;

[0044] Airport operational capacity data includes runway capacity, sector load, and flight departure rate.

[0045] The flight operation data acquired by the system is a set of core parameters that accurately describe the flight plan and actual execution status. Specifically, it includes the flight number used to uniquely identify the flight, the planned takeoff and landing times strictly adhering to Coordinated Universal Time (UTC), the clear departure and destination airport codes, the spatial coordinate sequence of waypoints arranged in flight order, and the recorded actual delay time relative to the planned time. These data collectively characterize the individual and group movement features of air traffic flows.

[0046] Simultaneously, the meteorological and airspace status data acquired by the system provide crucial environmental information affecting flight safety and efficiency. This data, at a finer granular level, includes, but is not limited to: polygonal boundaries and intensity information of thunderstorm areas detected and digitally delineated by meteorological radar; real-time wind speed and direction vectors at specific airspace levels; airspace closure areas designated due to military activities or special missions and their effective periods; and official military exercise notification texts. This data serves as the direct basis for the system to perceive external disturbances and make dynamic adjustments.

[0047] In addition, the system integrates airport operational capacity data that characterizes the limits of infrastructure capacity. This data quantifies the processing capacity of key nodes and primarily includes: the number of aircraft takeoffs and landings a single runway can handle per unit time (typically one hour), i.e., runway capacity; the number of aircraft handled by a specific control sector within the airspace at a given moment, i.e., sector load; and the actual flight departure efficiency of the airport in ground operations, i.e., flight departure rate. This data provides benchmark thresholds for assessing system load and triggering capacity adjustments.

[0048] By systematically integrating and standardizing the aforementioned multi-source heterogeneous data, this application enables the entire system to make coherent logical reasoning and decisions based on consistent data facts, thereby improving the overall situational awareness and intelligent decision-making level of the air traffic management system.

[0049] This application further proposes that the specific methods for generating the fused data matrix by aligning according to timestamps and spatial coordinates include:

[0050] The system aligns planned timestamps in flight operation data with observation timestamps in meteorological and airspace status data. Specifically, the system first performs alignment along the time dimension. It extracts planned takeoff, landing, and waypoint overflight times, accurate to the second, from the flight operation data and matches them with the corresponding observation times (such as weather radar scan cycles and airspace status activation times) for each record in the meteorological and airspace status data. This process associates flight planning events within the same or adjacent time windows with meteorological and airspace status, ensuring temporal consistency in the analysis.

[0051] In the spatial dimension, the system aligns waypoint sequences in flight operation data with meteorological and airspace status data to determine meteorological impact areas. Specifically, the system overlays the waypoint trajectories (composed of latitude and longitude coordinate sequences) from flight operation data with thunderstorm and strong wind areas defined by geographic polygon boundaries in meteorological data, and closed zone boundaries in airspace status data. Through spatial geometric calculations, the system can accurately determine whether each flight's trajectory crosses or is adjacent to specific meteorological impact areas or restricted airspace at a given time, thus establishing a direct link between flight operations and spatial environmental disturbances.

[0052] After completing the above spatiotemporal alignment, the system integrates information entries from different data sources but within the same spatiotemporal framework, generating a structured fused data matrix for each analysis time slice. The rows of this fused data matrix typically represent different analysis units (such as a single flight or a specific airspace sector), while the columns represent the fused multi-dimensional attributes (such as flight schedule information, current weather conditions, and the capacity status of the sector).

[0053] Through the aforementioned refined spatiotemporal alignment and fusion mechanism, this application effectively solves the analytical bias problem caused by inconsistent spatiotemporal data benchmarks in traditional methods, effectively overcomes the information silos and cognitive limitations caused by dispersed data sources and inconsistent spatiotemporal benchmarks in traditional air traffic management, and enhances the overall predictability, accuracy and collaborative efficiency of air traffic management.

[0054] This application further proposes that, after the traffic prediction module generates traffic feature vectors for each airport, the system performs a feature similarity matrix calculation to quantify the inherent correlation between the operating modes of different airports within the airspace network. The specific steps for calculating the distribution differences of traffic feature vectors among different airports and generating the feature similarity matrix include:

[0055] Extract flight density distribution characteristics, time-segment structure characteristics, and weather impact characteristics from historical flight operation data of each airport;

[0056] Calculate the similarity weights of flight density distribution characteristics, time period structure characteristics, and weather impact characteristics among different airports;

[0057] Generate a feature similarity matrix based on similarity weights.

[0058] The above process first extracts multi-dimensional statistical features from the historical flight operation data of each airport to form a comprehensive feature vector. These features aim to characterize the core patterns of airport operations and mainly include: flight density distribution features, that is, the distribution of the number of arriving and departing flights at the airport in different time periods (such as hours and days), reflecting its basic traffic flow pattern; time period structure features, used to distinguish the differences in traffic flow composition at different time periods such as weekdays and weekends, and morning and evening peak hours; and weather impact features, which quantify the typical impact of historical meteorological conditions (such as visibility and wind speed) on airport traffic flow and delay levels.

[0059] After obtaining the above feature set, the system uses a weighted similarity calculation model to evaluate the overall similarity between any two airports. The formula is as follows:

[0060]

[0061] in: This represents the comprehensive feature similarity between airport i and airport j;

[0062] , These represent the flight density distribution characteristics, time-segment structure characteristics, and weather impact characteristics of airports i and j, respectively.

[0063] It is a function used to calculate the similarity between two feature vectors. Depending on the nature of the feature data, cosine similarity or Gaussian kernel function can be selected.

[0064] These are the similarity weights assigned to the three types of features: flight density, time period structure, and weather impact, and they satisfy the following conditions: Weight Optimization can be achieved through pre-defined domain knowledge or by utilizing historical data and machine learning methods (such as feedback tuning based on prediction performance). For example, when optimizing based on historical data, initial weights can be set to... =0.5, =0.3, =0.2.

[0065] The system calculates the overall similarity of each airport pair by traversing all airport pairs. Ultimately, these results are constructed into a The symmetric matrix, or feature similarity matrix, is given by N, where N is the total number of airports. This feature similarity matrix, from a data-driven perspective, clearly reveals the relationships between nodes in the airspace network in terms of their operational modes.

[0066] By generating a feature similarity matrix using the methods described above, this application transforms the complex, implicit operational relationships between airports into explicit, quantifiable mathematical expressions. It can draw upon the historical experience of similar airports, improving the accuracy of basic traffic prediction results, especially at airports with sparse data or those experiencing unforeseen events, thereby enhancing the robustness and generalization ability of the entire prediction system.

[0067] like Figure 3 As shown, this application further proposes specific steps for detecting sudden disturbance events based on real-time acquired meteorological and airspace status data and correcting the basic traffic forecast results, including:

[0068] The system scans and analyzes real-time meteorological and airspace status data to detect sudden disturbances that may disrupt normal traffic patterns.

[0069] Real-time monitoring of changes in thunderstorm areas and airspace closure status in meteorological and airspace status data; specifically, firstly, time-series analysis of radar image sequences is performed to monitor the movement path and area change trend of thunderstorm areas, while simultaneously listening to airspace status instructions issued by air traffic control departments to identify newly added closed areas or changes in their effective time.

[0070] When an expansion of a thunderstorm area or the closure of new airspace is detected, the module calculates the scope and duration of the disturbance's impact. Specifically, once an event such as the expansion of a thunderstorm area or the closure of new airspace is detected, the module immediately activates the impact assessment algorithm. This algorithm integrates the event's geographical location, intensity (e.g., thunderstorm level), and direction of movement, and uses an airspace network topology model to deduce the affected routes and sectors, thereby defining the scope of the disturbance's impact. Simultaneously, based on meteorological forecast data or the validity period of airspace instructions, the expected duration of the disturbance event is estimated.

[0071] The basic flow forecast results are weighted and corrected based on the scope and duration of the disturbance.

[0072] To quantify the impact of disturbances on the prediction results, a weighted correction model is used to dynamically adjust the basic flow prediction results. The formula for the corrected flow prediction results is as follows:

[0073]

[0074] in: This represents the corrected flow forecast result at a future time t;

[0075] This represents the basic flow prediction result at the same time t;

[0076] It is a function related to time t, representing the proportion of traffic flow for planned flights within the affected area of ​​the disturbance;

[0077] This represents the overlap duration of the predicted time t within the duration of the disturbance, given the current system time.

[0078] It is a time decay constant used to adjust the sensitivity to the effect of duration;

[0079] This is a correction factor used to control the overall magnitude of the correction. Its value can be configured according to the type of disturbance (such as thunderstorms or military exercises). The correction factor α is set according to the type of disturbance. For example, for thunderstorm disturbances, α can be 0.5 to 0.7; for normal flow fluctuations, α can be 0.1 to 0.3.

[0080] The weighted correction model is characterized by the fact that the wider the scope of the disturbance and the longer the overlap with the prediction time, the greater the downward adjustment of the system to the basic prediction result, thus more accurately reflecting the flow suppression effect caused by the disturbance event.

[0081] Through the aforementioned real-time detection and quantitative correction mechanism, this application dynamically adapts static basic traffic forecasts to rapidly changing operating environments. This effectively overcomes the shortcomings of traditional forecasting models, such as delayed or inaccurate responses to sudden disturbances, generating corrected traffic forecasts that better reflect actual airspace capacity. This provides a reliable and near-real-time decision-making basis for subsequent capacity adjustments and collaborative optimization, enhancing the system's decision-making resilience and practicality in uncertain environments.

[0082] This application further proposes that the specific steps for the capacity adjustment module to generate a capacity adjustment scheme include:

[0083] Based on the revised traffic forecast results and combined with the airspace topology, the predicted traffic density of each airspace sector is calculated, which is the number of aircraft expected to enter the sector per unit time.

[0084] To scientifically assess capacity risk, the predicted traffic density is compared with the maximum carrying capacity in the airport's operational capacity data, and the capacity deviation value is calculated.

[0085] The capacity deviation value directly reflects the extent to which the predicted flow exceeds the capacity limit. The calculation formula is quantified as follows:

[0086]

[0087] in, The capacity deviation value of sector i is a dimensionless ratio.

[0088] This represents the predicted flow density of sector i;

[0089] This represents the maximum carrying capacity of sector i, an inherent parameter determined by the airspace structure, control rules, and safety standards.

[0090] The system has a preset safe deviation threshold. When the capacity deviation value of a certain sector is detected satisfy When a sector is deemed to be at risk of overload, the module automatically triggers adjustment logic. At this point, the system selects and combines options from the policy library to generate a specific capacity adjustment plan based on the magnitude and duration of the deviation and surrounding airspace conditions. This plan is not a single instruction, but a set of actionable adjustment instructions. Core measures include: dynamically adjusting the number of operating seats to enhance the control service capacity for the sector; temporarily modifying the sector boundary to redistribute some load to adjacent sectors with larger available capacity; or reallocating route priorities to reshape and divert traffic for flights flying over the area. These instructions together constitute a comprehensive solution aimed at restoring the sector load to a safe level.

[0091] Specifically, during the initial deployment, the system loads historical flight traffic and airspace capacity data from the past year. Using a grid search method, it iterates within a preset reasonable range [0.10, 0.20] with a step size of 0.01 to find the δ value that maximizes the accuracy of airspace congestion warnings over the historical period as the initial threshold. For example, calculations show that δ=0.15 most effectively avoids frequent false alarms while triggering timely capacity adjustments; therefore, the initial preset deviation threshold δ is set to 0.15. Subsequently, this preset deviation threshold δ will be dynamically adjusted by the self-learning optimization module based on real-time performance.

[0092] Through the aforementioned automatic decision-making mechanism based on quantification deviation, this application can proactively identify potential capacity bottlenecks and generate accurate and timely control instructions, thereby effectively preventing airspace overload, improving the flexibility and security of overall airspace resource utilization, and laying a feasible resource constraint foundation for subsequent multi-airport collaborative optimization.

[0093] like Figure 4 As shown, this application further proposes to generate a multi-airport collaborative scheduling scheme through a game equilibrium algorithm. This process models each airport in the region as a player in a game, aiming to find a globally optimal solution acceptable to all parties through the algorithm. Specific steps include:

[0094] Based on the revised traffic forecast results and regional flight coordination data, and according to the airspace operation structure within the region, the affected airports are set as game participants; the decision variable for each game participant is its planned flight schedule sequence.

[0095] The payoff function for each game participant is constructed using delay cost parameters from regional flight coordination data and flight departure efficiency from airport operational capacity data. The core parameters of the payoff function are derived from delay cost parameters (such as the unit-time delay cost agreed upon by airlines) in the regional flight coordination data and flight departure efficiency from the airport operational capacity data. The goal of the payoff function is to improve departure efficiency while minimizing its own total delay cost.

[0096] However, airports do not make decisions independently. An airport's flight release sequence can affect the traffic reception of its downstream airports through the airway network. To address this, the system calculates the interrelationships of flight traffic between airports based on revised traffic forecasts, analyzes the connections and conflicts between flights, and generates a coordination cost matrix. This matrix quantifies the additional delays or efficiency losses caused by any airport's schedule adjustments to other airports.

[0097] After defining the payoff function and coordination cost matrix, the system solves for these using iterative algorithms (such as optimal response dynamics or virtual game theory). The iterative algorithm simulates the process by which each participant continuously optimizes its own strategy after considering the strategies of others, ultimately converging to a Nash equilibrium. During the iterative process of solving for the Nash equilibrium, the learning rate parameter can be set to 0.01 and the exploration rate parameter to 0.1 to ensure the convergence of the algorithm. In this equilibrium state, no single airport can gain greater benefits by unilaterally changing its scheduling scheme. The system outputs the strategy combination corresponding to this equilibrium point, thus forming the final multi-airport collaborative scheduling scheme.

[0098] By introducing a game theory model, this application transforms the complex multi-agent collaborative optimization problem into a structured mathematical solution problem. It overcomes the shortcomings of traditional centralized optimization, which struggles to balance the interests of all parties. It seeks an approximate optimal solution for overall operational efficiency while respecting the autonomy of each airport, thereby enhancing the scientific rigor and fairness of collaborative decision-making in large-scale airspace networks and achieving global optimization of airspace resource utilization and flight operation efficiency.

[0099] This application further proposes that, after the collaborative optimization module generates a multi-airport collaborative scheduling scheme through a game equilibrium algorithm, the system also includes a self-learning optimization module, which performs the following operations:

[0100] The self-learning optimization module begins working after the collaborative optimization module generates a scheduling plan and executes it for one cycle, continuously collecting multi-dimensional performance data generated by the system during actual operation. This data constitutes the system performance feedback dataset, whose core indicators include: prediction error data between traffic forecasts and actual observations, capacity utilization data calculated from the ratio of actual traffic to maximum capacity in each airspace sector, and average flight delay time data within the region.

[0101] A multi-objective reward function is constructed based on the system performance feedback dataset to quantitatively evaluate the overall operational efficiency of the system. The multi-objective reward function comprehensively reflects the weighted evaluation results of prediction accuracy, airspace resource utilization efficiency, and flight punctuality rate.

[0102] The mathematical expression of the multi-objective reward function is as follows:

[0103]

[0104] in: This represents the reward value obtained in this period, i.e., the weighted evaluation result. The higher the value, the better the system performance.

[0105] It is a metric for prediction accuracy, which can be represented by 1 minus the normalized mean absolute error;

[0106] It is a negative indicator of airspace resource utilization efficiency, used to measure the balance of capacity utilization in each sector, and to avoid some sectors being overloaded while others are idle.

[0107] It is the average delay time;

[0108] These are the positive weights assigned to the three sub-goals, used to adjust the relative importance of different performance goals.

[0109] The weighted evaluation results are used to dynamically adjust the parameters of each module through a reinforcement learning algorithm, generating an optimized system parameter set; the weighted evaluation results are used as the reward value. The data is fed into a reinforcement learning algorithm (such as policy gradient or Actor-Critic). The reinforcement learning algorithm aims to maximize long-term cumulative reward by automatically exploring and evaluating adjustment strategies for the corresponding parameters of each module in the system. In this embodiment, the reinforcement learning algorithm uses the Proximal Policy Optimization (PPO) algorithm, with a learning rate of 0.0003, a discount factor of 0.99, and a mini-batch size of 64 data points per iteration.

[0110] These key parameters to be adjusted include: the feature transfer weights used in the traffic prediction module to calculate the feature similarity matrix. The preset deviation threshold used to trigger alarms in the capacity adjustment module. The algorithm outputs a set of optimized system parameters, including the game strategy parameters (such as learning rate and exploration rate) of the game model in the collaborative optimization module. Through iterative optimization, the algorithm finally outputs a set of optimized system parameters.

[0111] By introducing the aforementioned reinforcement learning-based self-learning loop, this application enables the system to evolve from a static automated tool into an intelligent agent with continuous evolution capabilities. It can adaptively fine-tune its internal core parameters based on feedback from the real operating environment, thereby overcoming the performance degradation problem caused by improper initial parameter settings or long-term drift of the operating environment. This ensures that the system can maintain and continuously improve its comprehensive performance in traffic prediction, resource allocation, and collaborative optimization throughout its entire lifecycle.

[0112] This application further proposes that, before generating the fused data matrix, the system also includes a data verification and repair module, used to perform the following operations:

[0113] The data verification and repair module serves as the first line of defense for data entering the system. It is responsible for quality control of all raw input data obtained from the civil aviation database, including flight operation data, meteorological and airspace status data, airport operation capacity data, and regional flight coordination data.

[0114] First, a rigorous data verification process is implemented, scanning the input data stream using predefined business rules and statistical models. Completeness checks and outlier detection are performed on flight operation data, meteorological and airspace status data, airport operational capacity data, and regional flight coordination data obtained from the civil aviation database. This identifies missing or outlier data items that exceed preset reasonable ranges. Two main types of data quality issues are identified: first, completeness issues, which involve detecting missing data in key fields, such as empty flight waypoint sequences or consecutively missing meteorological observations; and second, reasonableness issues, which involve detecting whether data values ​​exceed preset reasonable ranges, such as negative flight delay durations or airport runway capacity far exceeding physical limits.

[0115] Instead of simply discarding identified anomalous data items, the module initiates a corresponding repair mechanism. The repair strategy is dynamically selected based on the type of data anomaly and the context: for data with obvious time-series patterns (such as airport hourly traffic), the module uses interpolation algorithms (such as time series prediction or spline interpolation) based on the corresponding historical data sequences for repair; for data that requires immediate verification (such as sudden changes in airspace status), the module attempts to perform cross-validation from other independent real-time data sources and logically reconstructs the original anomalous data based on the verification results to restore its most likely true value.

[0116] After completing the above verification and repair process, all data items are converted into a unified standardized data format. The standardized data, after verification and repair, is sent to the data acquisition module for data alignment and fusion. This data serves as reliable input for the module to align timestamps and spatial coordinates, ultimately generating a high-quality fused data matrix.

[0117] This application improves the quality and consistency of input data from the source, avoiding the risk of systematic bias or inaccurate decision-making in subsequent intelligent prediction and optimization modules. This mechanism ensures that the core functions of the system are built on a reliable data foundation, thereby enhancing the accuracy, stability, and reliability of the output results of the entire air traffic collaborative optimization system.

[0118] This application further proposes that the system also includes a closed-loop control module for coordinating data flow and execution timing between modules, specifically including:

[0119] One of the core functions of the closed-loop control module is to dynamically adjust the system's decision-making cycle. Based on system performance feedback data collected by the self-learning optimization module, it dynamically calculates the trigger execution interval for the next cycle. The formula is:

[0120]

[0121] in, This is the current trigger interval. This is the overall system performance score for the current period calculated based on the self-learning optimization module. For the normalized score in the interval [0,1], The preset target performance threshold, The adjustment coefficient, used to control the sensitivity to interval changes, is typically set to a small positive number, such as β = 0.1. This adjustment mechanism allows the system to adapt even when performance is poor ( Automatically shorten the decision-making cycle for rapid response, while when performance meets or exceeds standards ( If necessary, the cycle should be extended appropriately to save computing resources.

[0122] Meanwhile, the closed-loop control module has established a comprehensive version management system. After each trigger execution, a globally unique version identifier is generated to correct traffic prediction results, capacity adjustment plans, and multi-airport collaborative scheduling plans. The version identifier is then bound and stored with the generation timestamp and input data fingerprint to form a complete traceable data chain.

[0123] When the self-learning optimization module outputs the optimized system parameter set, it matches the corresponding data environment state based on the version identifier of the currently executed scheme, performs compatibility verification on the optimized parameters through the parameter verification interface, and loads them into the corresponding modules in batches according to the version identifier after the verification is passed.

[0124] By introducing intelligent periodic adjustments and strict version control, this application achieves dynamic matching between the system's operating rhythm and changes in the external environment, ensuring the stability and reliability of the system optimization process. This not only improves resource utilization efficiency but, more importantly, provides crucial technical support for the continuous optimization and safe evolution of complex air traffic management systems, enabling the system to maintain optimal decision-making quality and adaptability during long-term operation.

[0125] The following is a specific example of a big data-based intelligent flight flow prediction and air traffic collaborative optimization system:

[0126] A severe summer convective weather event occurred in the airspace of the Yangtze River Delta region, requiring airports A, B, and C to coordinate traffic control during the morning rush hour due to thunderstorm disturbances. The system activated intelligent prediction and collaborative optimization throughout the entire process, enabling dynamic allocation of airspace resources and improving flight operation efficiency.

[0127] The data acquisition module retrieves four types of core data in real time from the civil aviation comprehensive database:

[0128] Flight operation data: Waypoint sequence (latitude and longitude N31°10′ / E121°20′ to N30°20′ / E120°15′) and historical delay time (average 12 minutes in the past 30 days) for flight 1 at Airport A (scheduled departure at 08:30); International flight departure rate at Airport B (89% from 07:00 to 08:00).

[0129] Meteorological and airspace status data: The East China Air Traffic Management Bureau's radar detected a thunderstorm area (center N31°05′ / E121°30′, area expanded to 200km). 2 (Moving speed 15km / h), accompanied by level 8 crosswinds; military temporary airspace closure notice (08:00-10:00, affecting the northeast sector of Airport C).

[0130] Airport operational capacity data: Airport A runway capacity (45 takeoffs and landings per hour), Airport B north sector load (currently 32 aircraft / sector, maximum capacity 40 aircraft), Airport C flight departure rate (78% at 07:30).

[0131] Regional coordination data: Flight delay cost coefficients for airports A and B (12,000 yuan / minute and 18,000 yuan / minute respectively), and route priority agreements between airport C and the two airports.

[0132] A fused data matrix is ​​generated by aligning timestamps (UTC second-level synchronization) and matching spatial coordinates (based on waypoint latitude and longitude). For example, the scheduled time of flight 1 (08:30) is bound with the thunderstorm impact period (08:15-09:00) and the runway capacity data of Airport A to form a spatiotemporally correlated structured data unit.

[0133] The traffic prediction module extracts traffic feature vectors for the three airports from the fused data matrix, including:

[0134] Flight density distribution: Airport A has an average flow of 38 flights / hour during the morning peak (07:00-09:00), Airport B has 42 flights / hour, and Airport C has 28 flights / hour.

[0135] Time-of-day structure characteristics: Airport A has a weekday morning peak traffic fluctuation coefficient of 0.15 (0.08 on weekends), while international flights account for 45% of Airport B's traffic (compared to 20% for Airport A).

[0136] Weather impact characteristics: Under thunderstorm conditions, the traffic flow at Airport A decreased by 0.3 (historical average), while the traffic flow at Airport C decreased by 0.25 due to crosswinds.

[0137] The association of operational modes between airports is quantified using a weighted similarity formula:

[0138]

[0139] Among them, weight The similarity function used is cosine similarity. The calculation yields:

[0140] ;

[0141] .

[0142] Based on the similarity matrix, the parameters of the mature LSTM prediction model trained at Airport B are transferred to Airport A (similarity 0.76) to generate basic traffic prediction results: Airport A's traffic from 08:00 to 09:00 is 41 aircraft / hour (the original independent model prediction bias rate was 8%, which was reduced to 3% after migration).

[0143] The disturbance correction module monitored weather radar data in real time and found that the thunderstorm area expanded by 15% at 07:45 and moved towards the main landing runway of Airport A. The disturbance impact range was calculated using the airspace topology model: covering the approach sector of Airport A (N31°00′-31°15′ / E121°10′-121°30′), lasting for approximately 90 minutes (08:15-09:45).

[0144] The baseline forecast results are adjusted using a weighted correction model:

[0145] in, racks / hour (basic forecast); (Thunderstorm disturbance correction factor); (Percentage of flights affected); D(t) = 60 minutes (duration of overlap between predicted time t = 08:30 and duration of disturbance); minutes (time decay constant).

[0146] Substituting into the weighted correction model, we get: .

[0147] Revised traffic forecast: Airport A's actual capacity to handle traffic from 08:00 to 09:00 has decreased to 26 aircraft per hour, a reduction of 36.6% compared to the baseline forecast.

[0148] The capacity adjustment module will correct the comparison between traffic flow and airport operational capacity, for Airport A's approach sector. racks / hour (corrected forecast flow). racks / hour (maximum capacity during thunderstorms);

[0149] Calculate the spatial sector capacity deviation value: (Exceeds the preset deviation threshold δ=0.15).

[0150] The system triggered the following capacity adjustment measures: 1. Operational seat adjustment: Add 2 controllers to the approach sector of Airport A (increasing the number of seats from 3 to 5); 2. Sector boundary redrawing: Temporarily allocate 30% of the airspace in the northeast sector of Airport A to the north sector of Airport B (the north sector of Airport B currently has a load of 32 aircraft and a capacity of 40 aircraft, with some redundancy); 3. Route priority allocation: Prioritize international flights to use the western backup route and detour around the thunderstorm area.

[0151] The collaborative optimization module sets Airport A, Airport B, and Airport C as game participants, constructs a payoff function, and generates a coordination cost matrix by combining the delay cost parameters in the regional coordination data (Airport A: 12,000 yuan / minute, Airport B: 18,000 yuan / minute). The Nash equilibrium is then solved through an iterative algorithm.

[0152] Game theory solution process:

[0153] Airport A has a delay rate of 15%, Airport B has a delay rate of 10%, and Airport C has a delay rate of 8%. After iterative adjustments, Airport A reduced its delay rate to 9% by transferring 3 flights to Airport B's backup sector. After accepting the transfer, Airport B's release efficiency decreased from 89% to 85%, but due to the higher delay cost (18,000 yuan / minute), its overall revenue improved. The Nash equilibrium point was determined, with delay rates of 9%, 11%, and 7% for the three airports respectively. By minimizing the diagonal elements of the collaborative cost matrix, a multi-airport collaborative scheduling scheme was finally generated.

[0154] Airport A: Flight 1 is delayed until 09:50 and will use the North Sector route of Airport B instead;

[0155] Airport B: International flights are given priority, and the average delay of domestic flights is controlled within 12 minutes;

[0156] Airport C: Adjusting the intervals between inbound flights to avoid the period affected by thunderstorms, and activating temporary waiting areas.

[0157] During system operation, the self-learning optimization module collects performance feedback data: prediction error rate 3.2% (target threshold 5%), capacity utilization rate 89%, and average delay time 11 minutes. Parameters are dynamically adjusted based on a multi-objective reward function.

[0158] Feature transfer weights for traffic prediction module Increased from 0.4 to 0.45 (to enhance the effect of density characteristics);

[0159] Capacity adjustment preset deviation threshold Tighten from 0.15 to 0.12 (increase sensitivity).

[0160] The closed-loop control module scores based on current performance. Extend the trigger interval ( minute, ), to obtain the trigger execution interval for the next cycle. : (Since the performance meets the requirements, the decision-making cycle will be extended appropriately to save computing resources.)

[0161] In this embodiment of the severe convective weather event, the system captured early signals of thunderstorm expansion through multi-source data fusion. Based on the similarity of traffic characteristics between airports, it achieved cross-airport model transfer, reducing the traffic prediction error of Airport A from 15% to 3.2% compared to traditional methods. When the thunderstorm's impact area expanded, the disturbance correction module dynamically lowered the capacity prediction, and combined with capacity deviation assessment, triggered sector boundary redrawing and seat replenishment, avoiding airspace overload. Through game theory equilibrium algorithm to coordinate the scheduling of the three airports, the overall regional delay time was ultimately controlled within 11 minutes, a reduction of more than 40% compared to similar weather conditions in the same period in history. The system's self-learning mechanism continuously optimized parameters. In the following week of operation, the prediction response speed improved by 15%, and the airspace resource utilization rate stabilized in the range of 85%-90%, achieving fully closed-loop intelligent management.

[0162] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A big data-based intelligent flight flow prediction and air traffic collaborative optimization system, characterized by: include: The data acquisition module is used to obtain flight operation data, meteorological and airspace status data, airport operation capacity data and regional flight coordination data from the civil aviation database, and align them according to timestamps and spatial coordinates to generate a fused data matrix; The traffic prediction module is used to extract traffic feature vectors of each airport from the fused data matrix, calculate the distribution differences of the traffic feature vectors between different airports to generate a feature similarity matrix, and transfer the prediction model parameters based on the feature similarity matrix to generate basic traffic prediction results. The disturbance correction module is used to detect sudden disturbance events based on the real-time acquired meteorological and airspace status data, and to correct the basic traffic forecast results to generate corrected traffic forecast results. The capacity adjustment module is used to calculate the capacity deviation value of each airspace sector based on the corrected traffic forecast result and the airport operation capacity data, and generate a capacity adjustment plan when the capacity deviation value exceeds a preset deviation threshold. The collaborative optimization module is used to generate a multi-airport collaborative scheduling scheme based on the corrected traffic forecast results, the capacity adjustment scheme, and the regional flight coordination data through a game equilibrium algorithm.

2. The intelligent flight flow prediction and air traffic collaborative optimization system based on big data as described in claim 1, characterized in that: The data obtained from the civil aviation database includes flight operation data, meteorological and airspace status data, airport operation capacity data, and regional flight coordination data, specifically including: The flight operation data includes flight number, scheduled time, departure and arrival airports, waypoint sequence, and delay duration; The meteorological and airspace status data includes thunderstorm areas, wind speed and direction, airspace closure areas, and military exercise notices. The airport operational capacity data includes runway capacity, sector load, and flight departure rate.

3. The intelligent flight flow prediction and air traffic collaborative optimization system based on big data as described in claim 1, characterized in that: The specific methods for generating a fused data matrix by aligning it according to timestamps and spatial coordinates include: Align the scheduled times in the flight operation data with the observation times in the weather and airspace status data using timestamps; Align the waypoint sequence in the flight operation data with the meteorological impact area in the meteorological and airspace status data using spatial coordinates.

4. The intelligent flight flow prediction and air traffic collaborative optimization system based on big data as described in claim 1, characterized in that: The specific steps for calculating the distribution differences of traffic feature vectors among different airports and generating a feature similarity matrix include: Extract flight density distribution characteristics, time-segment structure characteristics, and weather impact characteristics from historical flight operation data of each airport; Calculate the similarity weights of the flight density distribution characteristics, time period structure characteristics, and weather impact characteristics among different airports; The feature similarity matrix is ​​generated based on the similarity weights.

5. The intelligent flight flow prediction and air traffic collaborative optimization system based on big data as described in claim 1, characterized in that: The specific steps for detecting sudden disturbance events based on the real-time acquired meteorological and airspace status data and correcting the basic traffic forecast results include: Real-time monitoring of changes in thunderstorm areas and airspace closure status in the aforementioned meteorological and airspace status data; When the expansion of the thunderstorm area or the closure of new airspace is detected, the range and duration of the disturbance are calculated. The basic flow prediction results are weighted and corrected based on the range of the disturbance and the duration of the disturbance.

6. The intelligent flight flow prediction and air traffic collaborative optimization system based on big data as described in claim 1, characterized in that: The specific steps by which the capacity adjustment module generates the capacity adjustment scheme include: Calculate the predicted flow density for each airspace sector based on the corrected flow prediction results; The predicted flow density is compared with the maximum carrying capacity in the airport's operational capacity data to calculate the capacity deviation value; When the capacity deviation value exceeds a preset deviation threshold, the capacity adjustment scheme is generated; wherein, the capacity adjustment scheme includes adjustment instructions for adjusting the number of operating seats, modifying sector boundaries, or reallocating route priorities.

7. The intelligent flight flow prediction and air traffic collaborative optimization system based on big data as described in claim 1, characterized in that: The specific steps for generating a multi-airport collaborative scheduling scheme using a game equilibrium algorithm include: Based on the revised traffic forecast results and the regional flight coordination data, each airport is set as a game participant; The delay cost parameter in the regional flight coordination data and the flight departure efficiency in the airport operation capacity data are used to construct the payoff function for each game participant. Based on the corrected traffic forecast results, the mutual influence relationship of flight traffic among airports is calculated, and a coordination cost matrix is ​​generated. The Nash equilibrium point is obtained by solving the revenue function and the coordination cost matrix through an iterative algorithm, and a multi-airport collaborative scheduling scheme based on the Nash equilibrium point is output.

8. The intelligent flight flow prediction and air traffic collaborative optimization system based on big data as described in claim 1, characterized in that: After the collaborative optimization module generates a multi-airport collaborative scheduling scheme through a game equilibrium algorithm, the system also includes a self-learning optimization module, which performs the following operations: Collect prediction error data, capacity utilization data, and average delay time data generated during system operation to generate a system performance feedback dataset. A multi-objective reward function is constructed based on the system performance feedback dataset. The multi-objective reward function comprehensively reflects the weighted evaluation results of prediction accuracy, airspace resource utilization efficiency and flight punctuality rate. The weighted evaluation results are used to dynamically adjust the corresponding parameters of each module through a reinforcement learning algorithm to generate an optimized system parameter set; wherein, the corresponding parameters of each module include the feature transfer weight in the traffic prediction module, the preset deviation threshold in the capacity adjustment module, and the game strategy parameters in the collaborative optimization module.

9. The intelligent flight flow prediction and air traffic collaborative optimization system based on big data as described in claim 1, characterized in that: Before generating the fused data matrix, the system also includes a data verification and repair module, used to perform the following operations: The integrity of the flight operation data, the meteorological and airspace status data, the airport operation capacity data and the regional flight coordination data obtained from the civil aviation database is checked and anomaly detection is performed to identify abnormal data items that are missing or exceed a preset reasonable range. For the identified abnormal data items, imputation and repair are performed based on the corresponding historical data sequence, or cross-validation and data reconstruction are performed in combination with real-time data sources; The standardized data, after verification and repair, is sent to the data acquisition module for data alignment and fusion to generate the fused data matrix.

10. The intelligent flight flow prediction and air traffic collaborative optimization system based on big data according to claim 8, characterized in that: The system also includes a closed-loop control module for coordinating data flow and execution timing between modules, specifically including: Based on the system performance feedback data collected by the self-learning optimization module, the trigger execution interval for the next cycle is dynamically calculated. The formula is: ,in, This is the current trigger interval. The overall system performance score for the current period is calculated based on the self-learning optimization module. The preset target performance threshold, For adjustment coefficients; After each trigger execution, a globally unique version identifier is generated for the corrected traffic prediction result, the capacity adjustment scheme and the multi-airport collaborative scheduling scheme, and the version identifier is bound and stored with the generation timestamp and input data fingerprint; When the self-learning optimization module outputs the optimized system parameter set, it matches the corresponding data environment state based on the version identifier of the currently executed scheme, performs compatibility verification on the optimized parameters through the parameter verification interface, and loads them into the corresponding modules in batches according to the version identifier after the verification is passed.

Images