Flight scheduling method and system based on airspace partitioning data

Abstract

This invention discloses a flight scheduling method and system based on airspace partitioning data. The method includes: constructing an airspace partitioning database, determining regional parameters, constructing a planning cost function, performing flight scheduling, and generating a flight scheduling report. This invention belongs to the field of flight scheduling technology, specifically referring to a flight scheduling method and system based on airspace partitioning data. This solution binds dispersed airspace basic data with multi-dimensional location features, transforming dispersed multi-source data into a unified spatiotemporal shared feature matrix. Adaptive weight allocation is achieved through a task adaptive fusion layer, enabling head-enhanced training and parameter updates. The method couples the single-device planning cost function with the multi-device conflict cost, considering both local and global optima. Memoized recursive dynamic programming is used to decompose the total global cost into staged state values, meeting the real-time scheduling requirements of large-scale multi-device scenarios.

Description

Technical Field

[0001] This invention belongs to the field of flight scheduling technology, specifically a flight scheduling method and system based on airspace partitioning data. Background Technology

[0002] Flight scheduling based on airspace partitioning data is a method that uses an airspace partitioning database and intelligent prediction model to dynamically mine parameters such as flight hazard index and capacity load of airspace areas, constructs multi-dimensional planning cost functions and conflict avoidance rules, and combines dynamic programming algorithms to achieve accurate route allocation.

[0003] However, existing flight scheduling technologies suffer from several problems. First, the airspace data is fragmented and scattered from multiple sources. Second, multi-dimensional heterogeneous data is difficult to integrate into effective features. Third, traditional regional parameters are mostly manually calibrated and cannot be dynamically adjusted according to the spatiotemporal characteristics of the airspace. This results in low accuracy and poor adaptability of airspace regional parameter predictions, making it impossible to support refined and dynamic flight scheduling. Fourth, traditional scheduling focuses only on the cost of a single device and does not consider route conflicts and airspace resource competition among multiple devices. The combined scheduling of multiple devices, multiple airspaces, and multiple time slices involves a massive number of state variables, which can easily lead to the curse of dimensionality, resulting in frequent route conflicts and parameter misjudgments, thus restricting the safety and operational efficiency of collaborative operations of multiple flight devices. Summary of the Invention

[0004] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides a flight scheduling method and system based on airspace partitioning data. It addresses the technical problems of fragmented and dispersed multi-source airspace data, difficulty in fusing multi-dimensional heterogeneous data into effective features, and the fact that traditional regional parameters are mostly manually calibrated and cannot be dynamically adjusted according to the spatiotemporal characteristics of the airspace, resulting in low prediction accuracy and poor adaptability of airspace regional parameters, thus failing to support refined and dynamic flight scheduling. This invention binds dispersed airspace basic data with multi-dimensional positional features to achieve accurate correlation of multi-source data. A shared feature layer is designed to transform dispersed multi-source data into a unified spatiotemporal shared feature matrix, achieving deep fusion of multi-dimensional data. An adaptive weight allocation is achieved through a task-adaptive fusion layer, combined with task head-enhanced training and parameter update mechanisms, enabling the predicted regional parameters to dynamically adjust. This approach adapts to the spatiotemporal attributes of airspace and specific task scenarios to improve prediction accuracy. Addressing the technical challenge that traditional scheduling focuses solely on the cost of a single device, neglecting route conflicts and airspace resource contention among multiple devices, the combined scheduling of multiple devices, airspaces, and time slices involves massive state variables, easily falling into the curse of dimensionality. This leads to frequent route conflicts and parameter misjudgments, hindering the safety and operational efficiency of multi-flight equipment collaborative operations. The solution couples the single-device planning cost function with the multi-device conflict cost, balancing local and global optima. It employs memoized recursive dynamic programming to decompose the total global cost into staged state values, reducing computational complexity and efficiently extracting the airspace-time slice combination that minimizes the total global cost. This meets the real-time scheduling needs of large-scale multi-device scenarios, improving the safety and real-time performance of airspace scheduling.

[0005] The technical solution adopted by this invention is as follows: The flight scheduling method based on airspace partitioning data provided by this invention includes the following steps:

[0006] Step S1: Construct a spatial domain partitioning database;

[0007] Step S2: Determine the region parameters;

[0008] Step S3: Construct the planning cost function;

[0009] Step S4: Perform flight scheduling;

[0010] Step S5: Generate a flight scheduling report.

[0011] Further, in step S1, the construction of the spatial domain partitioning database includes the following steps:

[0012] Step S11: Flight route data collection. The pre-declared flight routes of each flight device are retrieved through the flight planning system, and the real-time position trajectory of the device is obtained by combining it with the ADS-B system to form dual route data of pre-planned route and real-time corrected route; the basic airspace delineation rules are collected simultaneously as the benchmark for airspace delineation.

[0013] Step S12: Airspace area matching and database construction. Based on dual-route data and basic airspace delineation rules, the spatial overlap analysis method is used to complete the airspace boundary calibration, obtain the airspace delineation database, and generate a unique airspace code.

[0014] Step S13: For airspaces with a matching coefficient that meets the standard and valid boundary verification, the system automatically generates a unique airspace code, which serves as the core index of the airspace in the database, thus completing the database integration.

[0015] Further, in step S2, determining the area parameters involves obtaining the area parameters for each airspace region based on the airspace partitioning database. These area parameters include four core indicators: flight hazard index, area capacity, area delay duration, and predicted traffic flow. This process includes the following steps:

[0016] Step S21: Query the database, using the unique airspace code in the airspace partitioning database as an index, retrieve the complete partitioning data of the target airspace, and preliminarily determine the segment to which the target airspace belongs;

[0017] Step S22: If there is no directly delineated data for the target airspace, perform associated region completion;

[0018] Step S23: Determine the location within the plate, clarify the spatial positioning of the target airspace within its belonging plate, and obtain multi-dimensional location features;

[0019] Step S24: Construct a regional parameter prediction model to predict regional parameters for the target spatial domain, including the following steps:

[0020] Step S241: Collect multidimensional location feature data, historical flight record data, and labeled area parameter data as a dataset; perform data preprocessing on the dataset;

[0021] Step S242: Design a spatiotemporal shared feature layer, including spatial domain correlation graph feature extraction, gated temporal feature extraction, and spatiotemporal attention fusion; transform scattered multi-source data into a spatiotemporal shared feature matrix;

[0022] Step S243: Construct a task adaptive fusion layer, perform adaptive weight allocation, and output the predicted values ​​of regional parameters based on the exclusive feature matrix selected from the spatiotemporal shared feature matrix;

[0023] Step S244: Task head augmentation training. Different loss functions are designed for different task characteristics of region parameters, and then integrated into the total loss through dynamic weights.

[0024] After the model outputs the predicted values ​​of regional parameters, it is connected to the air traffic control business rule engine to correct the predicted values ​​of flight hazard index and regional capacity, and obtain the final predicted values ​​of regional parameters.

[0025] Step S245: Model parameters updated.

[0026] Furthermore, in step S3, the construction of the planning cost function specifically involves, for each flight device, first sorting out all the airspace areas traversed by its flight routes to form a dedicated flight area set for that flight device; then, based on the regional parameters of the dedicated flight area set, constructing a multi-dimensional constrained planning cost function, wherein the multi-dimensional constraints include flight hazard index constraints, regional delay duration constraints, predicted traffic constraints, and regional capacity constraints.

[0027] Furthermore, in step S4, the flight scheduling process couples the planning cost function of a single device with the conflict cost of multiple devices. Based on the conflict avoidance rule of prioritizing high-traffic devices, a multi-device airspace route collaborative scheduling optimization model is designed, including the following steps:

[0028] Step S41: Define airspace conflict. If the planned routes of two or more flight devices overlap in the same airspace r and at the same time, it is determined to be an airspace conflict.

[0029] Step S42: Calculate the airspace conflict cost, which is used to quantify the airspace conflict cost generated by the i-th flight equipment in the planned route;

[0030] Step S43: Combining the planning cost function of a single device with the airspace conflict cost, build a global optimization framework for airspace route scheduling of multiple devices, and set three types of constraints: safety, airspace resources, and device priority, with the goal of minimizing the total global cost.

[0031] Step S44: Decompose the global total cost into staged state values ​​through memoized recursive dynamic programming, calculate the cost increment according to the airspace occupancy status, realize the staged optimal allocation of multi-device routes, and achieve the scheduling goal of minimizing the global total cost; calculate the state values ​​of memoized dynamic programming.

[0032] The calculated state values ​​of the memoized dynamic programming are stored in a cache to avoid repeated calculations and improve iteration efficiency. When the global total cost fluctuation value of three consecutive iterations is less than the preset precision or the number of iterations reaches the upper limit, the calculation stops. From the final state values ​​of the memoized dynamic programming, the airspace-time slice allocation combination that minimizes the global total cost is extracted to generate the final route for each flight device. At the same time, the passage time window of each flight device is marked to complete the flight scheduling.

[0033] Furthermore, in step S5, generating the flight scheduling report specifically involves overlaying the route planning results onto the airspace electronic map, marking equipment priorities and hazard index levels with different colors, displaying the iterative convergence process of the global total cost using a line graph, comparing the matching relationship between traffic and capacity in each airspace using a bar chart, and sending it to the air traffic controller to review the practical feasibility of the flight scheduling results and generate a standardized electronic report. Simultaneously, the core data in the scheduling report is archived into the airspace partitioning database and model dataset to form a data closed loop.

[0034] The flight scheduling system based on airspace partitioning data provided by the present invention includes a module for constructing an airspace partitioning database, a module for determining regional parameters, a module for constructing a planning cost function, a module for performing flight scheduling, and a module for generating flight scheduling reports.

[0035] The airspace partitioning database construction module specifically involves defining airspace boundaries based on dual-route data and basic airspace delineation rules, obtaining an airspace partitioning database, and generating a unique airspace code; the data is then sent to the region parameter determination module.

[0036] The module for determining regional parameters specifically involves using a unique spatial code as an index to retrieve the segmentation data of the target spatial domain to determine the segment to which the target spatial domain belongs, clarifying the spatial location of the target spatial domain within its segment, obtaining multi-dimensional location features, predicting the regional parameters of the target spatial domain based on these multi-dimensional location features, designing a shared feature layer to transform scattered multi-source data into a spatiotemporal shared feature matrix, constructing a task adaptive fusion layer, performing adaptive weight allocation, outputting the predicted regional parameter values ​​based on the exclusive feature matrix selected from the spatiotemporal shared feature matrix, performing task head enhancement training and updating the parameters, and sending the data to the module for constructing the planning cost function.

[0037] The module for constructing the planning cost function specifically constructs a multi-dimensional constrained planning cost function based on the regional parameters of a single device's dedicated flight area set; and sends the data to the flight scheduling module.

[0038] The flight scheduling module specifically couples the planning cost function of a single device with the conflict cost of multiple devices. Based on the conflict avoidance rule of prioritizing high-traffic devices, it designs a multi-device airspace and route collaborative scheduling optimization model. Through memoized recursive dynamic programming, it decomposes the global total cost into staged state values, extracts the airspace-time slice allocation combination that minimizes the global total cost, and completes the flight scheduling. The data is then sent to the flight scheduling report generation module.

[0039] The flight scheduling report generation module specifically overlays the route planning results onto the airspace electronic map and simultaneously archives the core data in the report into the airspace partitioning database and model dataset, forming a data closed loop.

[0040] The beneficial results achieved by the present invention using the above solution are as follows:

[0041] (1) To address the technical problems of fragmented and scattered multi-source airspace data, difficulty in integrating multi-dimensional heterogeneous data into effective features, and the fact that traditional regional parameters are mostly manually calibrated and cannot be dynamically adjusted according to the spatiotemporal characteristics of the airspace, resulting in low prediction accuracy and poor adaptability of airspace regional parameters, which cannot support refined and dynamic flight scheduling, we adopt the method of binding scattered airspace basic data with multi-dimensional location features to achieve accurate association of multi-source data. We design a shared feature layer to transform scattered multi-source data into a unified spatiotemporal shared feature matrix to achieve deep integration of multi-dimensional data. We achieve adaptive weight allocation through a task adaptive fusion layer and combine task head enhancement training and parameter update mechanism to enable the predicted value of regional parameters to dynamically adapt to the spatiotemporal attributes of the airspace and specific task scenarios, thereby improving prediction accuracy.

[0042] (2) To address the technical problem that traditional scheduling only focuses on the cost of a single device and does not consider route conflicts and airspace resource competition among multiple devices, the combined scheduling of multiple devices, multiple airspaces and multiple time slices has a large number of state variables, which is prone to the dimensionality curse, leading to frequent route conflicts and parameter misjudgments, thus restricting the safety and operational efficiency of multi-flight equipment collaborative operation, we adopt a method that couples the planning cost function of a single device with the conflict cost of multiple devices, taking into account both local and global optima. We use memoized recursive dynamic programming to decompose the total global cost into staged state values, reduce computational complexity, and efficiently extract the airspace-time slice combination that minimizes the total global cost, thereby meeting the real-time scheduling requirements of large-scale multi-device scenarios and improving the safety and real-time performance of airspace scheduling. Attached Figure Description

[0043] Figure 1 A flowchart illustrating the flight scheduling method based on airspace partitioning data provided by this invention;

[0044] Figure 2 This is a schematic diagram of a flight scheduling system based on airspace partitioning data provided by the present invention.

[0045] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0047] Example 1, see Figure 1 The present invention provides a flight scheduling method based on airspace partitioning data, which includes the following steps:

[0048] Step S1: Construct an airspace delineation database. Specifically, based on dual-route data and basic airspace delineation rules, the spatial overlap analysis method is used to complete the airspace boundary calibration, obtain the airspace delineation database, and generate a unique airspace code.

[0049] Step S2: Determine the regional parameters. Specifically, using the unique spatial code as an index, retrieve the target spatial domain division data to determine the segment to which the target spatial domain belongs, clarify the spatial location of the target spatial domain within its segment, obtain multi-dimensional location features, construct a regional parameter prediction model based on the multi-dimensional location features, design a shared feature layer to transform the scattered multi-source data into a spatiotemporal shared feature matrix, construct a task adaptive fusion layer, perform adaptive weight allocation, output the regional parameter prediction value based on the exclusive feature matrix selected from the spatiotemporal shared feature matrix, perform task head enhancement training, and update the parameters.

[0050] Step S3: Construct the planning cost function, specifically, based on the regional parameters of the dedicated flight area set of a single device, construct a multi-dimensional constrained planning cost function;

[0051] Step S4: Perform flight scheduling, specifically by coupling the planning cost function of a single device with the conflict cost of multiple devices. Based on the conflict avoidance rule of prioritizing high-traffic devices, design a multi-device airspace and route collaborative scheduling optimization model. Through memoized recursive dynamic programming, decompose the global total cost into staged state values, extract the airspace-time slice allocation combination that minimizes the global total cost, and complete the flight scheduling.

[0052] Step S5: Generate a flight scheduling report, which involves overlaying the route planning results onto the airspace electronic map and simultaneously archiving the core data in the report to the airspace partitioning database and model dataset to form a data closed loop.

[0053] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the construction of the spatial domain partitioning database includes the following steps:

[0054] Step S11: Flight route data collection. The pre-declared flight routes of each flight device are retrieved through the flight plan system, and the real-time position trajectory of the device is obtained by combining the ADS-B system to form dual route data of pre-planned route and real-time corrected route; Simultaneously, basic airspace delineation rules are collected. The basic airspace delineation rules include airspace function type, altitude level division standard, control area range threshold, etc., as the basis for airspace delineation.

[0055] Step S12: Airspace area matching and database construction. Based on dual-route data and basic airspace delineation rules, the spatial overlap analysis method is used to complete the airspace boundary calibration, obtain the airspace delineation database, and generate a unique airspace code.

[0056] The spatial overlap analysis method includes the following:

[0057] First, noise reduction filtering and outlier removal are performed on the pre-planned route coordinates (WGS-84) and ADS-B real-time trajectory coordinates of the flight equipment. Then, the coordinates are converted into a local plane coordinate system for airspace control through a seven-parameter conformal projection model to ensure that the plane position error is less than 0.5 meters and the vertical height error is less than 0.3 meters, providing a precise coordinate reference for airspace boundary calibration.

[0058] Based on the flight altitude layer, all flight route spatiotemporal coordinate points at the same altitude layer are selected. Combined with the airspace safety interval threshold, a 1km×1km horizontal grid and a 100m vertical grid are set. The DBSCAN density clustering algorithm is used to lock several airspace candidate regions. Each airspace candidate region is based on the minimum convex hull of the cluster coordinate points as the basic outline, and a safety interval of 5km horizontally and 200m vertically is superimposed to expand outward to form an initial airspace boundary of an irregular polygon.

[0059] For each candidate airspace region, the pre-planned route airspace contour and the real-time trajectory airspace contour at the same altitude are overlaid, and the intersection area ratio of the two contours is calculated. Regions with an intersection ratio of ≥70% are high-matching candidate regions and proceed to the subsequent boundary verification stage; regions with an intersection ratio of less than 70% are marked as low-matching regions and are fed back to the air traffic control system for review.

[0060] For the boundary coordinates of the candidate airspace region, the boundary closure is verified using the ray method. A ray is emitted from the boundary point to be verified in the due east direction, and the number of intersections between the ray and the boundary of the region itself is counted. If the number of intersections is odd, it is determined to be a valid boundary point, and if it is even, it is removed. At the same time, the distance between adjacent boundary points is checked. If the plane distance is less than 50 meters, it is marked as too close. Key nodes such as trajectory inflection points are retained, and collinear redundant points are removed.

[0061] An airspace matching coefficient is introduced (weighted by 40% flight route coverage, 30% safety interval compliance rate, and 30% functional type adaptability). When the airspace matching coefficient is ≥0.8, the boundary is confirmed to be valid. Finally, the boundary topology compliance (no self-intersection, no hanging points) and airspace rule compliance (no intrusion into restricted areas) are verified to confirm the final airspace boundary.

[0062] Step S13: For airspaces with a matching coefficient that meets the standard and valid boundary verification, the system automatically generates a unique airspace code, which serves as the core index of the airspace in the database, thus completing the database integration.

[0063] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the determination of area parameters involves obtaining area parameters for each airspace area based on the airspace partitioning database. The area parameters include four core indicators: flight hazard index, area capacity, area delay duration, and predicted traffic flow. The steps include:

[0064] Step S21: Query the database, using the unique airspace code in the airspace division database as an index, retrieve the complete division data of the target airspace, including basic information such as airspace function type, geographic boundary coordinates, and altitude level, and divide it into sections such as core control and routine operation according to the control level, and preliminarily determine the section to which the target airspace belongs.

[0065] Step S22: If there is no directly delineated data for the target airspace (e.g., a newly added temporary flight area), complete the associated area;

[0066] The formula used to calculate the regional similarity between a target airspace without directly delineated data and other airspace regions is as follows:

[0067] ;

[0068] In the formula, S represents the region similarity, and its value ranges from [0, 5]. This represents shape similarity based on Hausdorff distance normalization. This represents the shape weighting coefficient, which is fixed at 1.2. This indicates a target spatial domain where no direct data partitioning is possible. Indicates the area of ​​the comparative airspace. The area of ​​all airspaces in the airspace partitioning database represents the average area, used to normalize area differences. D represents the straight-line distance between the geometric centers of two airspaces. Indicates the distance attenuation factor;

[0069] The preset similarity threshold is 0.6. 3-5 related airspaces with a similarity to the target airspace exceeding the threshold are selected. The distribution of the associated airspaces is statistically analyzed. The airspace with the highest frequency is taken as the airspace to which the target airspace belongs. If there are multiple airspaces with the same frequency, the airspace function type of the target airspace is combined to finally confirm the airspace to which the target airspace without direct division data belongs.

[0070] Step S23: Determine the location within the plate, clarify the spatial positioning of the target airspace within its belonging plate, and obtain multi-dimensional location features, which include absolute coordinate features, relative location features, and hierarchical location features;

[0071] The absolute coordinate features include the latitude and longitude of the spatial geometric center, altitude, and boundary coordinate extreme values;

[0072] The relative positional characteristics include the distance and azimuth between the airspace and core landmarks within the area (such as airports and route intersections);

[0073] The hierarchical location features include the position of the airspace within the plate height hierarchy;

[0074] Step S24: Construct a regional parameter prediction model to predict regional parameters for the target spatial domain, including the following steps:

[0075] Step S241: Collect multidimensional location feature data, historical flight record data, and labeled area parameter data as a dataset; preprocess the dataset; integrate the dimensions of the preprocessed dataset to construct the core input tensor. Where T represents the time step, N represents the number of spatially associated nodes, and F represents the multi-source feature dimension;

[0076] Step S242: Design a spatiotemporal shared feature layer, including airspace association graph feature extraction, gated temporal feature extraction, and spatiotemporal attention fusion; transform scattered multi-source data such as airspace geographic features, historical flight time series features, and meteorological features into a unified feature representation, i.e., a spatiotemporal shared feature matrix; including the following:

[0077] Feature extraction of spatial domain association graph and construction of spatial domain association graph based on spatial domain topology. Where V represents each control zone node, E represents the route connectivity between zones, and A represents the adjacency matrix, with adjacent zones marked as 1 and non-adjacent zones marked as 0; the spatial domain association graph features between zones are learned by improving the graph convolutional network, and the formula used is as follows:

[0078] ;

[0079] In the formula, Represents the feature matrix of the spatial domain correlation graph. This represents the activation function. Represents an enhanced adjacency matrix. Let X be the degree matrix of the enhanced adjacency matrix, and let X be the input tensor. Represents the spatial feature weight matrix. Indicates spatial feature bias term;

[0080] Gated temporal feature extraction addresses the issues of temporal information redundancy and gradient vanishing over long sequences in traditional temporal convolutional networks by introducing a gating mechanism to construct a gated temporal convolutional network. The formula used is as follows:

[0081] ;

[0082] ;

[0083] In the formula, Represents the gated output matrix. Represents the gate weight matrix. This indicates a causal convolution operation. Indicates the gating bias term. It represents the Hadamah accumulation. This represents the hyperbolic tangent activation function. Represents the convolution weight matrix. Represents the convolution bias term. Indicates residual connection, Represents the time-series feature matrix;

[0084] Spatiotemporal attention fusion introduces a spatiotemporal attention mechanism, allowing the model to automatically focus on key spatiotemporal features. The formula used is as follows:

[0085] ;

[0086] ;

[0087] In the formula, Represents the spatiotemporal attention weights. This represents the time feature vector at time t1. This represents the spatial feature vector of the i-th controlled area. This represents the spatiotemporal feature matching score, where N represents the total number of controlled areas. Represents the spatiotemporal shared feature matrix;

[0088] Step S243: Construct a task-adaptive fusion layer, perform adaptive weight allocation, and output the predicted region parameters based on the exclusive feature matrix selected from the spatiotemporal shared feature matrix, including the following:

[0089] In air traffic control scenarios, area parameter prediction involves multiple parallel tasks with significant differences in operational priority and prediction difficulty (e.g., flight hazard index prediction has a much higher priority than area delay duration prediction, and its prediction error tolerance is lower). An adaptive weight allocation method is used to achieve a dynamic balance between task priority and prediction accuracy. The formula used is as follows:

[0090] ;

[0091] In the formula, This represents the adaptive weight of the k-th task, where k represents the task index. This represents the business priority coefficient of the kth task (which needs to be preset manually to reflect the core requirements of air traffic control). This represents the prediction error of the real-time validation set for the k-th task;

[0092] For task feature adaptation, since the feature requirements of the four core parameters differ, a unique feature mapping is designed for each task. The task-specific feature matrix is ​​selected and recombined from the spatiotemporal shared feature matrix, using the following formula:

[0093] ;

[0094] In the formula, This represents the feature matrix specific to the k-th task. This represents the feature adaptation weight matrix for the k-th task. This represents the feature adaptation bias term for the k-th task;

[0095] Differentiated prediction head design: Based on the characteristics of each task, dedicated prediction heads are designed for the four regional parameter prediction tasks to achieve the mapping from the dedicated feature matrix to the original predicted value. The formula used is as follows:

[0096] ;

[0097] In the formula, This represents the predicted flow rate. This represents the head weight matrix for traffic prediction. The feature matrix representing traffic prediction. This represents the bias term in the flow forecast header. ReLU represents the predicted regional capacity, and ReLU represents the activation function. This represents the regional capacity prediction head weight matrix. This represents the specific feature matrix for regional capacity prediction. This represents the regional capacity forecast head bias term. This represents the predicted regional delay duration. This represents the activation function used to achieve non-negativity constraints and smooth gradients. This represents the weight matrix for predicting regional delay duration. The feature matrix representing the prediction of regional delay duration. This indicates the head bias term for the regional delay duration prediction. This represents the predicted flight hazard index, which is divided into four levels: low, medium, high, and extremely high. A classification prediction head is used. This represents the classification prediction head weight matrix. This represents the classification prediction head bias term. argmax represents the dedicated matrix for predicting the flight hazard index, and argmax represents the selection of the highest probability level from the hazard level probability distribution P output by Softmax as the final predicted value of the flight hazard index.

[0098] Step S244: Task-head augmentation training. Differentiated loss functions are designed for the different task characteristics of the four parameters, and then dynamically weighted to form the total loss. This achieves multi-task collaborative optimization and improves the robustness of the regional parameter prediction model. ;in, The loss function for the k-th task is represented by: RMSE and quantile combined loss are used to calculate the loss for predicted traffic, regional capacity, and regional delay duration; cross-entropy loss and MAE combined loss are used to calculate the loss for the flight hazard index.

[0099] After the model outputs the predicted values ​​of regional parameters, it is connected to the air traffic control business rule engine to correct the predicted values ​​of flight hazard index and regional capacity, thus obtaining the final predicted values ​​of regional parameters. The formula used is as follows:

[0100] ;

[0101] ;

[0102] In the formula, This represents the final predicted value of the revised Flight Hazard Index. This represents the meteorological rule correction factor (manually preset). Indicates a high-risk weather warning indicator. This represents the corrected regional capacity forecast. This represents the original regional capacity prediction value. This indicates the upper limit of the area's safety capacity as approved by air traffic control.

[0103] Step S245: Model parameter update. The model parameters refer to the sum of all learnable weight matrices and bias terms in the entire regional parameter prediction model. Airspace operation data has strong time sensitivity (new routes, adjustments to control rules, seasonal traffic changes). If a full retraining is adopted, it will consume a lot of computing resources. Smooth incremental updates are adopted to ensure that regional parameter prediction can operate effectively in the long term under the dynamic business scenario of air traffic control. While retaining the adaptation to historical airspace patterns, new patterns in the newly added data (such as peak season traffic characteristics and the impact of new control rules) are absorbed. The formula used is as follows:

[0104] ;

[0105] In the formula, This represents the new parameters of the model after incremental updates. This indicates that the parameter is updated to the attenuation coefficient. This represents the old parameters of the model before the update. This indicates the parameter increment for training with newly added data.

[0106] By performing the above operations, the system binds dispersed airspace basic data with multi-dimensional location features to achieve accurate correlation of multi-source data. A shared feature layer is designed to transform the dispersed multi-source data into a unified spatiotemporal shared feature matrix, achieving deep fusion of multi-dimensional data. An adaptive weight allocation is achieved through a task-adaptive fusion layer. Combined with task head-enhanced training and parameter update mechanisms, the predicted regional parameters can dynamically adapt to the spatiotemporal attributes of the airspace and specific task scenarios, improving prediction accuracy. This solves the technical problems of dispersed and fragmented airspace multi-source data, difficulty in fusing multi-dimensional heterogeneous data into effective features, and the fact that traditional regional parameters are mostly manually calibrated and cannot be dynamically adjusted according to the spatiotemporal characteristics of the airspace, resulting in low prediction accuracy and poor adaptability of airspace regional parameters, which cannot support refined and dynamic flight scheduling.

[0107] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the construction of the planning cost function specifically involves, for each flight device, first sorting out all the airspace areas traversed by its flight routes to form a set of exclusive flight areas for that flight device; then, based on the regional parameters of the exclusive flight area set, constructing a planning cost function with multi-dimensional constraints, including flight hazard index constraints, regional delay duration constraints, predicted traffic constraints, and regional capacity constraints.

[0108] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the flight scheduling is performed by coupling the planning cost function of a single device with the conflict cost of multiple devices. Based on the conflict avoidance rule of prioritizing high-traffic devices, a multi-device airspace route cooperative scheduling optimization model is designed, including the following steps:

[0109] Step S41: Define airspace conflict. If the planned routes of two or more flight devices overlap in the same airspace r and at the same time, it is determined to be an airspace conflict.

[0110] Step S42: Calculate the airspace conflict cost, which is used to quantify the airspace conflict cost generated by the i-th flight equipment in the planned route. The formula used is as follows:

[0111] ;

[0112] In the formula, This represents the airspace conflict cost of the i-th flight equipment. The variable represents the collision penalty coefficient, and r represents a single spatial cell. Let represent the set of dedicated flight areas for the i-th flight device, and t represent a single time slice (e.g., 5 min / 10 min as a time slice). This represents the planned flight time interval for the i-th flight device. This indicates the occupancy status of airspace r in time slice t. A value of 1 indicates that the airspace is occupied by other flying equipment, and a value of 0 indicates that the airspace is idle. c represents the traffic priority weight, with a value range of [0.6, 0.9], used to ensure the passage priority of high-traffic equipment. Let M represent the reference flight traffic for the i-th device, M represent the set of all flight devices participating in route allocation, and j be the index used to traverse all flight devices. This represents the reference flight flow for the j-th flight device. This represents the maximum reference flight flow rate among all flight equipment;

[0113] Step S43: Combining the planning cost function of a single device with the airspace conflict cost, a global optimization framework for multi-device airspace route scheduling is established. The objective is to minimize the total global cost, and three types of constraints are set: safety, airspace resources, and device priority. The formulas used are as follows:

[0114] ;

[0115] In the formula, This represents the total global cost, which is the sum of the planning costs of all flight equipment and the costs of airspace conflicts. Let i represent the planning cost function for the i-th flight device. This represents the comprehensive risk index of the i-th flight equipment. This indicates the preset upper limit of the risk threshold. This indicates the priority determination threshold, when When necessary, the equipment is classified as a high-priority group, and the remaining equipment is classified as a normal-priority group. Routes are allocated to the high-priority group first, followed by processing of the normal-priority group, to prevent route preemption for high-priority equipment. This represents the set of all available airspace resources across the entire airspace.

[0116] Step S44: By using memoized recursive dynamic programming, the global total cost is decomposed into staged state values. The cost increment is calculated based on the differences in airspace occupancy status to achieve phased optimal allocation of multi-device routes and achieve the scheduling goal of minimizing the global total cost. This includes the following:

[0117] Memoized recursive calculus is used to calculate the state values ​​of memoized dynamic programming, using the following formula:

[0118] ;

[0119] ;

[0120] In the formula, Let represent the state value of the memoized dynamic programming, that is, the minimum global total cost accumulated when processing the i-th flight device, with airspace r in the corresponding occupied state during time slice t. This represents the minimum total global cost accumulated after processing the (i-1)th flight device, with airspace r in the corresponding occupied state during time slice t. This represents the minimum cost of the two occupancy states for each space-time slice, ensuring that the allocation of space-time slices is the optimal choice;

[0121] The calculated state values ​​of the memoized dynamic programming are stored in a cache to avoid repeated calculations and improve iteration efficiency. When the global total cost fluctuation value of three consecutive iterations is less than the preset precision (such as the preset precision is 0.01) or the number of iterations reaches the upper limit, the calculation stops. From the final state values ​​of the memoized dynamic programming, the airspace-time slice allocation combination that minimizes the global total cost is extracted to generate the final route for each flight device. At the same time, the passage time window of each flight device is marked to avoid time domain conflicts and complete flight scheduling.

[0122] By performing the above operations, the single-device planning cost function is coupled with the conflict cost of multiple devices, taking into account both local and global optima. Memoized recursive dynamic programming is used to decompose the total global cost into staged state values, reducing computational complexity and efficiently extracting the airspace-time slice combination that minimizes the total global cost. This meets the real-time scheduling requirements of large-scale multi-device scenarios, improves the safety and real-time performance of airspace scheduling, and solves the technical problem that traditional scheduling only focuses on the cost of a single device and does not consider route conflicts and airspace resource contention between multiple devices. The combined scheduling of multiple devices, multiple airspaces, and multiple time slices has a massive number of state variables, which is prone to the curse of dimensionality, leading to frequent route conflicts and parameter misjudgments, thus restricting the safety and operational efficiency of multi-flight equipment collaborative operation.

[0123] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, generating a flight scheduling report specifically involves overlaying the route planning results onto the airspace electronic map, marking equipment priorities and hazard index levels with different colors, displaying the iterative convergence process of the global total cost using a line graph, comparing the matching relationship between traffic and capacity in each airspace using a bar chart, and sending it to the air traffic controller to review the practical feasibility of the flight scheduling results and generate a standardized electronic report. Simultaneously, the core data in the scheduling report (such as optimized route data, parameter correction results, etc.) are archived into the airspace partitioning database and model dataset to form a data closed loop.

[0124] Example 7, see Figure 2This embodiment is based on the above embodiments. The flight scheduling system based on airspace partitioning data provided by the present invention includes a module for constructing an airspace partitioning database, a module for determining regional parameters, a module for constructing a planning cost function, a module for performing flight scheduling, and a module for generating flight scheduling reports.

[0125] The airspace partitioning database construction module specifically involves defining airspace boundaries based on dual-route data and basic airspace delineation rules, obtaining an airspace partitioning database, and generating a unique airspace code; the data is then sent to the region parameter determination module.

[0126] The module for determining regional parameters specifically involves using a unique spatial code as an index to retrieve the segmentation data of the target spatial domain to determine the segment to which the target spatial domain belongs, clarifying the spatial location of the target spatial domain within its segment, obtaining multi-dimensional location features, predicting the regional parameters of the target spatial domain based on these multi-dimensional location features, designing a shared feature layer to transform scattered multi-source data into a spatiotemporal shared feature matrix, constructing a task adaptive fusion layer, performing adaptive weight allocation, outputting the predicted regional parameter values ​​based on the exclusive feature matrix selected from the spatiotemporal shared feature matrix, performing task head enhancement training and updating the parameters, and sending the data to the module for constructing the planning cost function.

[0127] The module for constructing the planning cost function specifically constructs a multi-dimensional constrained planning cost function based on the regional parameters of a single device's dedicated flight area set; and sends the data to the flight scheduling module.

[0128] The flight scheduling module specifically couples the planning cost function of a single device with the conflict cost of multiple devices. Based on the conflict avoidance rule of prioritizing high-traffic devices, it designs a multi-device airspace and route collaborative scheduling optimization model. Through memoized recursive dynamic programming, it decomposes the global total cost into staged state values, extracts the airspace-time slice allocation combination that minimizes the global total cost, and completes the flight scheduling. The data is then sent to the flight scheduling report generation module.

[0129] The flight scheduling report generation module specifically overlays the route planning results onto the airspace electronic map and simultaneously archives the core data in the report into the airspace partitioning database and model dataset, forming a data closed loop.

[0130] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0131] 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.

[0132] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A flight scheduling method based on airspace partitioning data, characterized in that: The method includes the following steps: Step S1: Construct an airspace delineation database, specifically by marking airspace boundaries based on dual-route data and basic airspace delineation rules, obtaining the airspace delineation database and generating a unique airspace code; Step S2: Determine the regional parameters. Specifically, using the unique spatial code as an index, retrieve the target spatial domain division data to determine the segment to which the target spatial domain belongs, clarify the spatial location of the target spatial domain within its segment, obtain multi-dimensional location features, predict the regional parameters of the target spatial domain based on the multi-dimensional location features, design a shared feature layer to transform the scattered multi-source data into a spatiotemporal shared feature matrix, construct a task adaptive fusion layer, perform adaptive weight allocation, output the predicted regional parameter values ​​based on the exclusive feature matrix selected from the spatiotemporal shared feature matrix, perform task head enhancement training, and update the parameters. Step S3: Construct the planning cost function, specifically, based on the regional parameters of the dedicated flight area set of a single device, construct a planning cost function with multi-dimensional constraints; the multi-dimensional constraints include flight hazard index constraints, regional delay duration constraints, predicted traffic constraints, and regional capacity constraints; Step S4: Perform flight scheduling, specifically by coupling the planning cost function of a single device with the conflict cost of multiple devices. Based on the conflict avoidance rule of prioritizing high-traffic devices, design a multi-device airspace and route collaborative scheduling optimization model. Through memoized recursive dynamic programming, decompose the global total cost into staged state values, extract the airspace-time slice allocation combination that minimizes the global total cost, and complete the flight scheduling. Step S5: Generate a flight scheduling report, which involves overlaying the route planning results onto the airspace electronic map and simultaneously archiving the core data in the report to the airspace partitioning database and model dataset to form a data closed loop.

2. The flight scheduling method based on airspace partitioning data according to claim 1, characterized in that: In step S2, determining the area parameters involves obtaining the area parameters for each airspace region based on the airspace partitioning database. These area parameters include four core indicators: flight hazard index, area capacity, area delay duration, and predicted traffic flow. The steps include: Step S21: Query the database, using the unique airspace code in the airspace partitioning database as an index, retrieve the complete partitioning data of the target airspace, and preliminarily determine the segment to which the target airspace belongs; Step S22: If there is no directly delineated data for the target airspace, perform associated region completion; Step S23: Determine the location within the plate, clarify the spatial positioning of the target airspace within its belonging plate, and obtain multi-dimensional location features; Step S24: Construct a regional parameter prediction model to predict regional parameters for the target spatial domain, including the following steps: Step S241: Collect multidimensional location feature data, historical flight record data, and labeled area parameter data as a dataset; perform data preprocessing on the dataset; Step S242: Design a spatiotemporal shared feature layer, including spatial domain correlation graph feature extraction, gated temporal feature extraction, and spatiotemporal attention fusion; transform scattered multi-source data into a spatiotemporal shared feature matrix; Step S243: Construct a task adaptive fusion layer, perform adaptive weight allocation, and output the predicted values ​​of regional parameters based on the exclusive feature matrix selected from the spatiotemporal shared feature matrix; Step S244: Task head augmentation training. Different loss functions are designed for different task characteristics of region parameters, and then integrated into the total loss through dynamic weights. After the model outputs the predicted values ​​of regional parameters, it is connected to the air traffic control business rule engine to correct the predicted values ​​of flight hazard index and regional capacity, and obtain the final predicted values ​​of regional parameters. Step S245: Model parameters updated.

3. The flight scheduling method based on airspace partitioning data according to claim 1, characterized in that: In step S4, flight scheduling is performed by coupling the planning cost function of a single device with the conflict cost of multiple devices. Based on the conflict avoidance rule of prioritizing high-traffic devices, a multi-device airspace route cooperative scheduling optimization model is designed, including the following steps: Step S41: Define airspace conflict. If the planned routes of two or more flight devices overlap in the same airspace r and at the same time, it is determined to be an airspace conflict. Step S42: Calculate the airspace conflict cost, which is used to quantify the airspace conflict cost generated by the i-th flight equipment in the planned route; Step S43: Combining the planning cost function of a single device with the airspace conflict cost, build a global optimization framework for airspace route scheduling of multiple devices, and set three types of constraints: safety, airspace resources, and device priority, with the goal of minimizing the total global cost. Step S44: Decompose the global total cost into staged state values ​​through memoized recursive dynamic programming, calculate the cost increment according to the airspace occupancy status, realize the staged optimal allocation of multi-device routes, and achieve the scheduling goal of minimizing the global total cost; calculate the state values ​​of memoized dynamic programming. The calculated state values ​​of the memorized dynamic programming are stored in the cache. When the global total cost fluctuation value of three consecutive iterations is less than the preset precision or the number of iterations reaches the upper limit, the calculation stops. From the final state values ​​of the memorized dynamic programming, the airspace-time slice allocation combination that minimizes the global total cost is extracted to generate the final route for each flight device. At the same time, the passage time window of each flight device is marked to complete the flight scheduling.

4. The flight scheduling method based on airspace partitioning data according to claim 1, characterized in that: In step S1, constructing the spatial domain partitioning database includes the following steps: Step S11: Flight route data collection. The pre-declared flight routes of each flight device are retrieved through the flight planning system, and the real-time position trajectory of the device is obtained by combining it with the ADS-B system to form dual route data of pre-planned route and real-time corrected route; the basic airspace delineation rules are collected simultaneously as the benchmark for airspace delineation. Step S12: Airspace area matching and database construction. Based on dual-route data and basic airspace delineation rules, the spatial overlap analysis method is used to complete the airspace boundary calibration, obtain the airspace delineation database, and generate a unique airspace code. Step S13: For airspaces with a matching coefficient that meets the standard and valid boundary verification, the system automatically generates a unique airspace code, which serves as the core index of the airspace in the database, thus completing the database integration.

5. The flight scheduling method based on airspace partitioning data according to claim 1, characterized in that: In step S3, the construction of the planning cost function specifically involves, for each flight device, first sorting out all the airspace areas traversed by its flight routes to form a set of exclusive flight areas for that flight device; then, based on the regional parameters of the exclusive flight area set, constructing a planning cost function with multi-dimensional constraints.

6. The flight scheduling method based on airspace partitioning data according to claim 1, characterized in that: In step S5, generating a flight scheduling report specifically involves overlaying the route planning results onto an electronic airspace map, marking equipment priorities and hazard index levels with different colors, displaying the iterative convergence process of the global total cost using a line graph, comparing the matching relationship between traffic and capacity in each airspace using a bar chart, and sending the report to the air traffic controller. The practical feasibility of the flight scheduling results is reviewed, and a standardized electronic report is generated. Simultaneously, the core data in the scheduling report is archived into the airspace partitioning database and model dataset to form a data closed loop.

7. A flight scheduling system based on airspace partitioning data, used to implement the flight scheduling method based on airspace partitioning data as described in any one of claims 1-6, characterized in that: It includes modules for building an airspace partitioning database, determining regional parameters, building a planning cost function, performing flight scheduling, and generating flight scheduling reports.

8. The flight scheduling system based on airspace partitioning data according to claim 7, characterized in that: The airspace partitioning database construction module specifically involves defining airspace boundaries based on dual-route data and basic airspace delineation rules, obtaining an airspace partitioning database, and generating a unique airspace code; the data is then sent to the region parameter determination module. The module for determining regional parameters specifically involves using a unique spatial code as an index to retrieve the segmentation data of the target spatial domain to determine the segment to which the target spatial domain belongs, clarifying the spatial location of the target spatial domain within its segment, obtaining multi-dimensional location features, predicting the regional parameters of the target spatial domain based on these multi-dimensional location features, designing a shared feature layer to transform scattered multi-source data into a spatiotemporal shared feature matrix, constructing a task adaptive fusion layer, performing adaptive weight allocation, outputting the predicted regional parameter values ​​based on the exclusive feature matrix selected from the spatiotemporal shared feature matrix, performing task head enhancement training and updating the parameters, and sending the data to the module for constructing the planning cost function. The module for constructing the planning cost function specifically constructs a multi-dimensional constrained planning cost function based on the regional parameters of a single device's dedicated flight area set; and sends the data to the flight scheduling module. The flight scheduling module specifically couples the planning cost function of a single device with the conflict cost of multiple devices. Based on the conflict avoidance rule of prioritizing high-traffic devices, it designs a multi-device airspace and route collaborative scheduling optimization model. Through memoized recursive dynamic programming, it decomposes the global total cost into staged state values, extracts the airspace-time slice allocation combination that minimizes the global total cost, and completes the flight scheduling. The data is then sent to the flight scheduling report generation module. The flight scheduling report generation module specifically overlays the route planning results onto the airspace electronic map and simultaneously archives the core data in the report into the airspace partitioning database and model dataset, forming a data closed loop.

Images