Terminal area airspace sector dynamic division method based on multi-source data fusion, electronic equipment and medium

Abstract

The invention discloses a terminal area airspace sector dynamic division method based on multi-source data fusion, electronic equipment and a medium. The method comprises the following steps: rasterizing a terminal area airspace, mapping multi-source airspace data into a grid unit, constructing a multi-channel image feature matrix fusing traffic and meteorological information, and realizing multi-source information fusion representation; a potential vector of the feature matrix is extracted by using a variational auto-encoder, and typical operation scenes such as peak periods and severe weather are identified by using a clustering method; the method comprises the following steps: based on a typical operation scene, evaluating control load under each scene, constructing a scene-driven sector division set segmentation model, and decomposing an original problem into a main problem and a pricing sub-problem; and designing a row and column generation algorithm to solve the model, continuously searching a new path in combination with a branch and bound algorithm and a heuristic algorithm, and finally outputting an optimal sector division scheme. According to the method, fusion representation of multi-source airspace data can be realized, a high-flexibility sector division scheme can be quickly generated, the model complexity is reduced, the calculation efficiency is effectively improved, the airspace capacity utilization rate is improved, and the workload of a controller is reduced.

Description

Technical Field

[0001] This application relates to the field of sector partitioning, specifically to a method for dynamic partitioning of terminal area airspace sectors based on multi-source data fusion. Background Technology

[0002] The sector partitioning problem aims to balance controller workload and improve airspace resource utilization efficiency while ensuring air traffic control safety, through scientific spatial division and dynamic adjustment mechanisms. Its core objectives include balancing traffic load between sectors, optimizing route connectivity, and responding to the need for flexible expansion of airspace capacity in emergency scenarios.

[0003] Existing sector partitioning technologies are mainly divided into two categories:

[0004] (1) The sector partitioning technology based on Voronoi diagrams is based on the construction of a proximity relationship topology through dynamic generator location selection and weighted distance function. This method uses navigation stations and airway intersections as spatial base points, dynamically adjusts generator weights according to real-time traffic complexity and control load, drives the adaptive evolution of Voronoi polygon boundaries, and realizes dynamic matching between sector capacity and traffic demand.

[0005] (2) Based on the sector division of the gridded airspace, the airspace is discretized into grid units, and mixed integer programming or heuristic clustering algorithms are used to optimize the control load balance through discrete modeling and dynamic weighting mechanism. Combined with branch and bound and clustering algorithms, multi-objective collaborative decision-making under the airspace topology connectivity constraint is realized.

[0006] The following defects exist in the existing technology for processing sector division problems: (1) Traditional sector division models rely solely on traffic data, airspace structure data, and airspace complexity for sector division, but ignore the impact of meteorological elements such as thunderstorms and wind shear on sector control load, and lack a dynamic correlation mechanism between meteorological disturbance elements and control load; (2) The high-dimensional feature space and complex nonlinear coupling relationship of airspace operation data significantly restrict the scene recognition capability of clustering algorithms, making it difficult for traditional clustering algorithms to effectively retain key airspace operation mode information during feature dimensionality reduction; (3) The sector division generator weight adjustment mechanism based on Voronoi diagrams is limited to fixed landmarks such as route intersections, and lacks refinement and flexibility; (4) Sector division models based on gridded airspace require the construction of a large number of decision variables and constraints, and the time complexity of its solution increases exponentially with the airspace scale, making it difficult to meet the real-time optimization needs of large-scale airspace. Summary of the Invention

[0007] This application provides a method, electronic device, and medium for dynamic sector partitioning of terminals based on multi-source data fusion, which solves the technical problem that the existing technology ignores the fusion of multi-source information data and has a high complexity of sector partitioning model.

[0008] According to one aspect of this application, a method for dynamically dividing terminal area airspace sectors based on multi-source data fusion is provided, comprising the following steps:

[0009] Step 1: Rasterize the airspace of the terminal area, mapping multi-source airspace data, including at least airspace, flow and meteorological information, into raster cells to construct a multi-channel image feature matrix;

[0010] Step 2: Based on the variational autoencoder, extract features from the multi-channel image feature matrix to generate a potential spatial feature vector representing the airspace operation mode. Use a clustering method to divide scene clusters and identify at least typical airspace operation scenarios that are distinguished by traffic flow and weather.

[0011] Step 3: Based on the typical airspace operation scenarios and flight data, evaluate the control load of the typical airspace operation scenarios, construct a scenario-driven sector partitioning set segmentation model, and decompose the original problem of the sector partitioning set segmentation model into a main problem and a pricing sub-problem.

[0012] Step 4: Design a row and column generation algorithm to solve the sector partitioning set segmentation model. Combine the branch and bound algorithm and the heuristic algorithm to continuously search for new paths and finally output the optimal sector partitioning scheme.

[0013] Furthermore, in this application, the multi-source airspace data containing airspace, traffic flow, and meteorological information includes:

[0014] Structured airspace data, including sector boundaries and restricted airspace information;

[0015] Traffic data, including ADS-B track data and flight schedule data;

[0016] Meteorological data, including data from Fengyun meteorological satellites and Doppler radar data.

[0017] Furthermore, in this application, the feature extraction from the multi-channel image feature matrix based on the variational autoencoder includes:

[0018] An encoder is used to map the input data to the mean and variance of a latent distribution;

[0019] Reparameterization technique sampling of the latent space;

[0020] A decoder is used to reconstruct the data from the latent variables;

[0021] In the above process, the loss function is a weighted sum of reconstruction loss, KL divergence loss and VGG perception loss.

[0022] Furthermore, in this application, the method of using clustering to divide scene clusters includes:

[0023] Calculate the Davidson-Bolding index and the sum of squared errors within clusters curves corresponding to different cluster number K values. Select the K value corresponding to the minimum Davidson-Bolding index as the optimal number of clusters. At the same time, verify whether the inflection point of the sum of squared errors within clusters curve at the elbow point is consistent with the Davidson-Bolding index result, to ensure that the cluster number K can simultaneously meet the requirements of low intra-cluster dispersion and high inter-cluster separation.

[0024] When classifying typical operating scenarios based on clustering results, category labels for traffic scenarios and weather scenarios are marked according to each scenario cluster and cluster center.

[0025] Furthermore, in this application, the control load under typical airspace operation scenarios is evaluated, and a scenario-driven sector partitioning set segmentation model is constructed, including:

[0026] The gridded control load function is constructed, and the mapping from the typical airspace operation scenario to the control load is formed as follows:

[0027] F(N,J,W)=(ω N ·N+ω J ·J)(1+λW)

[0028] in,

[0029] N is the number of flights, ω N ω is the weighting coefficient for flight load, J is the number of route junctions, and ω is the weighting coefficient for flight load. J This is a weighting coefficient for flight load;

[0030] (1+λW) is the weather correction factor, where W=0 when the weather is good and the load is only the number of flights, and W=1 when there is severe weather and the weather intensity λ factor corrects the overall load;

[0031] Furthermore, in this application, the objective function of the main problem is as follows:

[0032]

[0033] Among them, C r For the cost of sector path r, C p Let x be the penalty coefficient for the slack variable P. r Choose a variable for the sector path. For slack variable P ij The penalty coefficient is R, which is the set of all sector paths, ij is the edge in the sector, and E is the set of edges in the sector;

[0034] P and P ij All are slack variables, where:

[0035] The constraints of the main problem include path quantity balance constraints and edge flow conservation constraints:

[0036] The path number balance constraint is as follows:

[0037] ∑ r∈R x r +P=K

[0038] The edge flow conservation constraint is as follows:

[0039] ∑ (i,j)∈E (θ ij, r·x r +θ ji ,r·x r )+P ij =0

[0040] Where θ ij,r This indicates the flow direction of path r on edge (i,j).

[0041] Furthermore, in this application, the method further includes: calculating the shadow value of resources by maximizing the dual problem model of the main problem to obtain the marginal benefits of airspace capacity and regulatory load, as shown in the following formula:

[0042]

[0043] Where, π K The dual value of the path number constraint; π ij This is the dual value of the edge flow balance constraint.

[0044] Furthermore, in this application, the design row and column generation algorithm solves the sector partitioning set segmentation model, continuously searching for new paths by combining branch and bound algorithms and heuristic algorithms, and finally outputs the optimal sector partitioning scheme, including:

[0045] First, an initial solution is selected from a finite subset of feasible sector partitioning paths, a linear relaxation model is constructed and solved to obtain the dual value of the path number constraint and the edge flow conservation constraint.

[0046] Subsequently, based on the dual value, the test number is calculated, and a new sector configuration scheme that satisfies the airspace capacity and control load is generated through heuristic search. Feasible solutions with negative test numbers are added to the main problem.

[0047] Finally, the main problem and subproblems are optimized through iterative loops until all test numbers are non-negative. At this point, the main problem reaches the optimal solution of linear relaxation, and the optimal sector partitioning scheme is obtained by branch and bound method.

[0048] A second aspect of this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it causes the electronic device to perform the method described in the first aspect of this application.

[0049] A third aspect of this application provides a computer-readable storage medium for storing a computer program that, when run on a computer, causes the computer to perform the method described in the first aspect of this application.

[0050] Beneficial effects:

[0051] As can be seen from the above technical solutions, the technical solution of the present invention provides a dynamic sector partitioning method for the terminal area airspace based on multi-source data fusion. First, the terminal area airspace is divided into grid cells based on a rasterization method, and multi-source airspace data is mapped into the grid cells to construct a multi-channel image feature matrix that integrates traffic and meteorological information. Then, a variational autoencoder is used to perform nonlinear dimensionality reduction on the multi-channel images to extract high-dimensional latent space vectors, and clustering is performed using a clustering method to identify typical operating scenarios such as severe weather and peak hours. On this basis, a sector partitioning model is constructed based on a row and column generation algorithm. The main problem uses integer programming to model sector path selection and a slack variable penalty mechanism, and the sub-problems use a greedy algorithm to integrate convexity preservation constraints and clockwise priority to generate new paths. Finally, the model is iteratively optimized based on a branch and bound algorithm to output the optimal sector partitioning scheme that satisfies airspace capacity, control load, and geometric compliance. This not only improves the problem of low sector freedom caused by sector partitioning technology based on Voronoi diagrams, but also solves the problem of excessive model complexity caused by grid-based sector partitioning, thereby meeting the needs of large-scale real-time airspace optimization. Attached Figure Description

[0052] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0053] Figure 1 This embodiment describes a method for dynamically dividing terminal area airspace sectors based on multi-source data fusion.

[0054] Figure 2 This is a flowchart of the multi-source data fusion described in this embodiment.

[0055] Figure 3 This is a flowchart illustrating the identification of a typical operating scenario as described in this embodiment.

[0056] Figure 4 This is a flowchart illustrating the row and column generation algorithm described in this embodiment.

[0057] Figure 5 This is a flowchart for solving the sub-problem described in this embodiment.

[0058] Figure 6 This is a schematic diagram of the sub-problem search path described in this embodiment.

[0059] Figure 7 This is a structural block diagram of the terminal area airspace sector dynamic partitioning device based on multi-source data fusion described in this embodiment.

[0060] Figure 8 This is a structural block diagram of the electronic device described in this embodiment. Detailed Implementation

[0061] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0062] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0063] Existing technologies for addressing the highly complex problem of airspace sector partitioning in terminal areas often only consider static sector partitioning, neglecting the need for flexible airspace resource allocation due to dynamic factors such as traffic flow fluctuations and sudden weather changes. This makes it difficult for partitioning schemes to adapt to the nonlinear changes in the spatiotemporal distribution of flight density. Furthermore, sector partitioning models have long solution times, and using solvers to solve mixed-integer programming models for sector partitioning suffers from low iterative efficiency, failing to meet the demands of large-scale real-time airspace optimization. A dynamic sector partitioning method for terminal areas based on multi-source data fusion is proposed. This method drives model iteration through real-time fusion of multi-source data, enabling the partitioning scheme to dynamically respond to traffic flow peaks and airspace status changes. Simultaneously, a column generation algorithm is used to quickly obtain the sector partitioning scheme, effectively improving airspace resource utilization.

[0064] like Figure 1 The diagram shown is a flowchart of a method for dynamically dividing the spatial sector of a terminal area based on multi-source data fusion, according to an embodiment of the present invention. Figure 1 It can be seen that the method for dynamically dividing the terminal area airspace sector based on multi-source data fusion includes the following steps:

[0065] Step 1: Rasterize the airspace of the terminal area, mapping multi-source airspace data, including at least airspace, traffic, and meteorological information, into raster cells to construct a multi-channel image feature matrix that integrates traffic and meteorological information, thereby achieving multi-source information fusion representation. The multi-source airspace data includes: structured airspace data, specifically sector boundary and restricted airspace information; traffic data, specifically ADS-B track data and flight schedule data; and meteorological data, specifically Fengyun meteorological satellite data and Doppler radar data.

[0066] Step 2: Based on the variational autoencoder, extract features from the multi-channel image feature matrix obtained in Step 1 to generate a potential spatial feature vector representing the airspace operation mode. Use a clustering method to divide scene clusters and identify at least typical airspace operation scenarios that are distinguished by traffic flow and weather. These typical operation scenarios include peak hours, severe weather, etc.

[0067] Step 3: Based on Step 2, evaluate the control load of the typical airspace operation scenarios and flight data, construct a scenario-driven sector partitioning set segmentation model, and decompose the original problem of the sector partitioning set segmentation model into a main problem and a pricing sub-problem.

[0068] Step 4: Design a row and column generation algorithm to solve the sector partitioning set segmentation model. Combine the branch and bound algorithm and the heuristic algorithm to continuously search for new paths and finally output the optimal sector partitioning scheme.

[0069] In some preferred embodiments, such as Figure 2 As shown, the terminal area's spatial domain is divided into several raster cells based on a rasterization method, with each raster cell having a spatial resolution of 0.05 degrees. Due to the heterogeneity of multi-source data, traditional methods struggle to fuse the data. Utilizing the characteristics of raster cells, structured data, flow data, and meteorological data from multiple sources are mapped to independent feature channels within each raster cell, with each raster cell representing a multi-level structure. Channel cascading technology is then used to cascade these independent feature channels together, ultimately constructing a multi-channel image feature matrix that fuses flow and meteorological information. Specifically, channel cascading technology generates a C×H×W fusion matrix (where H and W are the number of raster rows and columns, and C is the number of channels), and data quality is optimized through preprocessing and enhancement.

[0070] To address the mismatch between traffic flow data and meteorological data in the spatiotemporal dimensions, a time window method is used to match the temporal dimension, and a linear interpolation method is used to match the spatial dimension. To resolve the issues of different dimensions and imbalanced sample categories in multi-source heterogeneous data, channel normalization and oversampling are performed on the multi-channel feature matrix to optimize data quality, improve model generalization ability, and facilitate subsequent feature extraction.

[0071] The following analysis focuses on channel normalization and oversampling:

[0072] In convolutional neural networks, channel normalization eliminates dimensional differences between different channels by processing the data of each input channel independently. Specifically, for each channel of the input data, min-max normalization is used for standardization, as shown in the following formula:

[0073]

[0074] Where X represents the original channel data, X min and X max These are the minimum and maximum values ​​for the respective channels. This method linearly maps the data for each channel to the [0,1] interval, thus unifying the numerical scale. For example, in flight data, different original data ranges vary significantly; normalization prevents certain features from dominating model training due to their excessive magnitude, while also accelerating the convergence process of gradient descent.

[0075] To address the class skew problem caused by scarce samples during low-density flight periods, the SMOTE (Synthetic Minority Oversampling Technique) algorithm can be used for oversampling. This algorithm increases the amount of data during low-density periods by synthesizing minority class samples. The specific steps are as follows:

[0076] (1) Nearest neighbor selection: For each minority class sample S g Calculate its nearest neighbor samples (usually using Euclidean distance).

[0077] (2) Sample generation: Randomly select the nearest neighbor sample S f Along S g With S f The direction of the connection generates new samples:

[0078] S new =S g +rand(0,1)·(S f -S g )

[0079] Rand(0,1) is a random number in the interval [0,1], which controls the generation position of new samples.

[0080] In some preferred embodiments, such as Figure 3 As shown, in step 2, the multi-channel feature matrix is ​​input into the variational autoencoder. The encoder maps the input data to the mean and variance of the latent distribution, and then performs reparameterization trick sampling on the latent space. The decoder then reconstructs the data from the latent variables.

[0081] In the above process, the loss function of the variational autoencoder consists of three parts: reconstruction loss, KL divergence loss, and VGG perception loss.

[0082] The reconstruction loss measures the similarity between the decoder-generated data and the original input data. It takes the multi-channel feature matrix as input and the reconstructed multi-channel image matrix as output. The difference in image pixel values ​​is the reconstruction loss, and its expression is:

[0083]

[0084] Where, d θ (z) represents the decoder output, t is the input data, and z is the latent variable. q is the mean square error. φ (z|t) is the probability distribution of the encoder. For the latent variable z in its probability distribution q φ Expected value under (z|t).

[0085] KL divergence loss is used to constrain the latent distribution q of the encoder output. φ (z|t) and standard normal distribution The similarity is specifically manifested in the difference between the distribution of flow and meteorological characteristics and the standard normal distribution, and its expression is:

[0086]

[0087] Where, μ k and σ k and are the mean and standard deviation of the k-th dimension of the encoder output, respectively, where k is the k-th dimension of the latent vector z.

[0088] VGG perceptual loss extracts deep features from the input and reconstructed data using a pre-trained VGG network and calculates the Euclidean distance between feature maps. Specifically, this loss reflects the detailed representational ability of the latent feature vectors that characterize the features of flow and meteorological data. Its expression is as follows:

[0089]

[0090] Where, φ u,v W represents the feature map of the u-th and v-th convolutional layers of the VGG network. u,v H represents the width of the feature map of the u-th and v-th convolutional layers. u,v Let be the height of the feature map of the vth convolutional layer in the uth layer, and l and c represent the layer index and channel index of the feature map in the VGG network, respectively.

[0091] The total loss function is the weighted sum of the above three parts:

[0092]

[0093] Here, β and λ are balance coefficients used to adjust the influence weights of different loss terms.

[0094] In some optional embodiments, when using a variational autoencoder (VAE) to extract features from multimodal data, the flight operation matrix (containing temporal features such as flight quantity and delay status) and the meteorological parameter matrix (such as spatial distribution data of temperature, wind speed, and precipitation probability) are independently input into two parallel VAE encoders. This separate input design avoids feature coupling interference between modes, while capturing the latent distribution patterns of the two types of data through the probabilistic modeling capabilities of the VAE. Specifically, the flight matrix is ​​mapped by the encoder to a low-dimensional latent vector representing traffic fluctuations, and the mean and variance parameters output by the encoder are used to generate latent features following a Gaussian distribution through reparameterization techniques; while the meteorological matrix extracts the spatial correlation features of the meteorological field through another encoder, generating a latent vector reflecting the dynamic evolution of the meteorological system, and its training process uses KL divergence constraints to align the latent space with the standard normal distribution. This dual-channel architecture not only preserves the high dynamism of flight time-series data, but also effectively characterizes the complex spatial correlations of the meteorological field.

[0095] In some preferred embodiments, after generating latent space feature vectors through a variational autoencoder, a clustering method is used to perform cluster analysis on the latent space feature vectors to distinguish typical operating scenarios. The number of clusters K in the clustering method is determined by combining the Davies-Bouldin (DB) index and the sum of squared errors within clusters (SSE), specifically including the following steps:

[0096] By calculating the DB index and SSE curve corresponding to different K values, the K value corresponding to the minimum DB index is selected as the optimal number of clusters. At the same time, it is verified whether the inflection point of the SSE curve at the elbow is consistent with the DB index result, so as to ensure that the number of clusters K can simultaneously meet the requirements of low intra-cluster dispersion and high inter-cluster separation.

[0097] In some preferred embodiments, step 2 involves using a clustering method to divide scenario clusters. Specifically, this includes: dividing typical operational scenarios based on the clustering results; labeling traffic scenarios and weather scenarios with category tags based on each scenario cluster and its cluster center; classifying traffic scenarios into peak and off-peak periods based on flight quantity thresholds; and classifying weather scenarios into severe and favorable weather conditions based on radar echo intensity, wind speed, and precipitation. The typical operational scenarios obtained through clustering provide a data-driven decision-making basis for the dynamic division of airspace sectors.

[0098] like Figure 4 As shown, the construction of the scenario-driven sector partitioning set segmentation model includes:

[0099] Based on spatial domain data, a dynamic column generation framework for sector configuration schemes is constructed. The process mainly includes three stages: solving the main problem, solving the pricing sub-problems, and dynamic iterative optimization.

[0100] Step 4 describes a row and column generation algorithm that optimizes airspace resources by dynamically generating feasible sector partitioning schemes. The process includes three stages: solving the initial master problem, solving the pricing subproblem, and dynamic iterative optimization. First, an initial solution is selected from a finite subset of feasible sector partitioning paths. A linear relaxation model is constructed and solved to obtain the dual value of the path quantity constraint and the edge flow conservation constraint. Then, a test number is calculated based on the dual value. A new sector configuration scheme that satisfies airspace capacity and control load is generated through heuristic search, and feasible solutions with negative test numbers are added to the master problem. Finally, the master problem and subproblems are iteratively optimized until all test numbers are non-negative. At this point, the master problem reaches the optimal solution of linear relaxation, and the optimal sector partitioning scheme is obtained using the branch and bound method.

[0101] The first stage is solving the main problem. The objective function of the main problem is divided into two parts: the first part is the cost of the sector path to achieve sector compactness, and the second part is the penalty for sector number deviation and traffic imbalance, as shown in the following formula:

[0102]

[0103] Among them, C r For the operating cost of sector path r, C p x is the penalty coefficient for the sector path. r Select variables for binary sector paths. Let be the penalty coefficient for edge flow conservation, R be the set of all sector paths, ij be the edges in the sector, and E be the set of edges in the sector;

[0104] Slack variables P and P ij These are used to absorb the deviations of the path quantity constraint and the edge flow balance constraint, respectively.

[0105] The constraints of the main problem include the path number balance constraint and the edge flow conservation constraint.

[0106] The path number balance constraint:

[0107] ∑ r∈R x r +P=K

[0108] The total number of sector paths is consistent with the preset number of sectors K. The slack variable P makes the constraint a loose constraint, allowing the model to generate an initial feasible solution. On the other hand, it imposes a penalty on the number of sectors that exceed the limit.

[0109] The edge flow conservation constraint:

[0110] ∑ (i,j)∈E (θ ij,r ·x r +θ ji,r ·x r )+P ij =0

[0111] Where θ ij,r The slack variable P represents the flow direction of path r on edge (i,j). ij Used to impose penalties on edges that do not meet flow balance requirements.

[0112] In some optional embodiments, the dual problem model of the main problem described in step 3 reflects the marginal benefits of airspace capacity and regulatory load by maximizing the shadow value of resources, as shown in the following formula:

[0113]

[0114] Where, π K This is the dual value of the path quantity constraint, reflecting the impact of increasing or decreasing the number of unit paths on the total cost; π ij Let be the dual value of the edge flow balance constraint, representing the marginal cost of flow imbalance at edge (i,j). The solution set of the dual problem provides the basis for sensitivity analysis of the primal problem; for example, when π... K >C r This indicates that the current airspace resource allocation is excessively redundant, and resource utilization needs to be improved by expanding sector paths.

[0115] After constructing the main problem, the fixed counterclockwise path is taken as the initial feasible solution, and the variables of the model are relaxed. Then, the linear relaxation model is solved based on the branch and bound method to obtain the relaxed objective function value and the dual value of the path number constraint and the edge flow conservation constraint.

[0116] After obtaining the dual value, the second stage is to construct and solve the pricing subproblem. The objective function of the subproblem is to minimize the test number, which is defined as:

[0117]

[0118] Among them, C r π represents the cost of path r. K The dual value of the path quantity constraint represents the impact of adding or removing paths on the total cost; π ij This is the dual value of the edge flow balancing constraint. This function filters candidate sector configuration schemes that can significantly reduce global costs by minimizing the difference between the path cost and the penalty term adjusted by the dual value.

[0119] like Figure 5 As shown, the solution process for the subproblem includes several steps:

[0120] The first step is to construct the graph network structure, adding all the nodes and edges, as well as virtual nodes;

[0121] Secondly, set the edge weights, which are the path cost minus the dual value;

[0122] Next, a greedy algorithm is used to select the edge with the lowest weight in each search to obtain the path with the minimum test number, while also checking for convexity and clockwise constraints.

[0123] Finally, a closed new path is generated. If the test number of the new path is negative, the new path is added to the main problem; otherwise, it is discarded.

[0124] In the design of subproblem constraints, key constraints include convexity preservation constraints, clockwise path constraints, and path load constraints.

[0125] Convexity preservation constraints and clockwise path constraints are achieved by combining vector cross product and dot product. Specifically, let P be three consecutive points in the path. i-1 ,P i ,P i+1 Its corresponding vector is and The direction of the turn is determined by calculating the sign of the cross product of the vectors. If the cross product is negative, the path satisfies local convexity and is clockwise. When the cross product is zero, the direction of the turn is further ensured by ensuring the dot product is positive. and The lines must be in the same direction to avoid collinear but opposite directions. This condition can be formalized into a mathematical constraint:

[0126] (P i -P i-1 )×(P i+1 -P i )≤0

[0127] (P i -P i-1 )·(P i+1 -P i )>0

[0128] Path load constraints are implemented by relating geometric area to path load. At each step of the path search, the area enclosed by the current path is calculated to prevent excessive path load in advance. After obtaining a new path, the overall load is calculated to further ensure balanced sector load. The Shoelace formula is used to calculate the area of ​​the polygon enclosed by the path:

[0129]

[0130] Where n is the number of vertices of the polygon, x n+1 =x1 and y n+1=y1. Each term x i y i+1 -x i+1 y i The symbolic result of the area of ​​the trapezoid formed by the vectors formed by two adjacent points and the coordinate axes is represented by a positive value indicating that the region expands outward and a negative value indicating that it contracts inward.

[0131] The specific path search process is as follows: Figure 6 As shown, starting from the virtual node, arbitrarily select a node as the starting node. Then, for each subsequent node connected to it, determine whether the convexity preservation constraint and clockwise path constraint are satisfied by combining the vector cross product and dot product to obtain a feasible set of nodes. Then, sort the weight values ​​of the edges and select the node corresponding to the edge with the smallest weight to add to the path. Repeat the above steps until returning to the virtual node to obtain a new feasible path.

[0132] The third stage is dynamic iterative optimization, which involves repeatedly executing constraints on solving the main problem and generating a series of subproblems until all test numbers are non-negative. At this point, the main problem reaches the optimal solution of linear relaxation, and finally the optimal sector partitioning scheme is obtained by using the branch and bound method.

[0133] The core mechanism of the branch-and-bound method for restoring integer constraints relies on the dynamic partitioning of the solution space and the co-optimization of the bounds. Its core process consists of two stages: relaxation problem iteration and integer recovery.

[0134] First, a linear programming model of the main problem is constructed by relaxing integer constraints. The relaxed solution is then solved using the simplex method or column generation technique, and the test number is calculated. If non-integer variables exist in the relaxed solution, the key variable is selected for branching. Mutually exclusive constraints are added to decompose the original problem into two subproblems, forming a tree-like search structure. The relaxed solution of each subproblem provides an upper bound for the objective function value, while the current best integer solution serves as the lower bound. Pruning is achieved by comparing the upper and lower bounds: if the upper bound of a subproblem is lower than the current lower bound, the branch is directly pruned to reduce the search space.

[0135] During the dynamic iteration process, the main problem and subproblems form a feedback loop. The main problem generates a dual value through relaxation solutions, and the subproblems generate new columns or add cutting plane constraints based on this information, gradually tightening the feasible region. When all test numbers are non-negative, the main problem reaches the optimal state of linear relaxation, at which point the integer property of the solution needs to be verified. If the relaxation solution satisfies the integer constraint, the output is the global optimum; otherwise, a new round of branching operations is triggered until all branches are pruned or a solution satisfying the integer constraint is found.

[0136] Based on the same inventive concept as the above-described method embodiments, this application also provides a device for dynamic partitioning of terminal area airspace sectors based on multi-source data fusion, such as... Figure 7The diagram shown is a schematic of a device for dynamic partitioning of terminal area airspace sectors based on multi-source data fusion, according to this embodiment. Figure 7 It can be seen that the terminal area airspace sector dynamic partitioning device based on multi-source data fusion includes the following modules:

[0137] The multi-source information fusion module is used to rasterize the airspace of the terminal area, mapping multi-source airspace data, including at least airspace, traffic, and meteorological information, onto raster cells to construct a multi-channel image feature matrix that fuses traffic and meteorological information, thereby achieving multi-source information fusion representation. The multi-source airspace data, including airspace, traffic, and meteorological information, includes: structured airspace data, specifically sector boundary and restricted airspace information; traffic data, specifically ADS-B track data and flight schedule data; and meteorological data, specifically Fengyun meteorological satellite data and Doppler radar data.

[0138] The operation scene recognition module is used to extract features from the multi-channel image feature matrix based on the variational autoencoder, generate a potential spatial feature vector representing the airspace operation mode, divide scene clusters using a clustering method, and identify at least typical airspace operation scenarios that are distinguished by traffic flow and weather. These typical operation scenarios include peak hours, severe weather, etc.

[0139] The model building module is used to evaluate the control load of the typical airspace operation scenarios based on the typical airspace operation scenarios and flight data, build a scenario-driven sector partitioning set segmentation model, and decompose the original problem of the sector partitioning set segmentation model into a main problem and a pricing sub-problem.

[0140] The partitioning scheme generation module is used to design row and column generation algorithms to solve the sector partitioning set segmentation model. It combines branch and bound algorithms and heuristic algorithms to continuously search for new paths and finally outputs the optimal sector partitioning scheme.

[0141] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0142] Based on the same inventive concept as the above method embodiments, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it enables the electronic device to implement the control method described in the above embodiments.

[0143] In one embodiment, the electronic device may be a server, and in this embodiment, the structure of the electronic device may be as follows: Figure 3 As shown, it includes a memory 2001, a communication module 2003, and one or more processors 2002.

[0144] The memory 2001 is used to store computer programs executed by the processor 2002. The memory 2001 may mainly include a program storage area and a data storage area. The program storage area may store the operating system and programs required to run instant messaging functions, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0145] Memory 2001 may be volatile memory, such as random-access memory (RAM); memory 2001 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 2001 may be any other medium capable of carrying or storing a desired computer program having the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 2001 may be a combination of the above-mentioned memories.

[0146] Processor 2002 may include one or more central processing units (CPUs) or digital processing units, etc. Processor 2002 is used to implement the above-mentioned audio data processing method when calling computer programs stored in memory 2001.

[0147] The communication module 2003 is used to communicate with terminal devices and other servers.

[0148] This application embodiment does not limit the specific connection medium between the memory 2001, communication module 2003, and processor 2002. This application embodiment... Figure 8 The memory 2001 and the processor 2002 are connected via a bus 2004, which is in... Figure 8 The connections between other components are illustrated with arrows and are for illustrative purposes only, not as limiting information. The Bus 2004 can be divided into address bus, data bus, control bus, etc. For ease of description, Figure 8 The text uses only one arrow to describe it, but does not indicate that there is only one bus or one type of bus.

[0149] Based on the same inventive concept as the above-described method embodiments, embodiments of the present invention also provide a computer-readable storage medium for storing a computer program. When the computer program is run on a computer, it enables the electronic device to implement the control method described in the above embodiments. The computer-readable storage medium can be a readable signal medium or a readable storage medium. A readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0150] Based on the same inventive concept as the above-described method embodiments, embodiments of the present invention also provide a computer program product, which includes a computer program that, when run on an electronic device, causes the electronic device to perform the steps of the control methods described above according to various exemplary embodiments of this application. The program product may take the form of any combination of one or more readable media. These computer program commands can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the commands executed by the processor of the computer or other programmable data processing device generate a process for implementing... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0151] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

Claims

1. A method for dynamically dividing terminal area spatial sectors based on multi-source data fusion, characterized in that, Includes the following steps: Step 1: Rasterize the airspace of the terminal area, mapping multi-source airspace data, including at least airspace, flow and meteorological information, into raster cells to construct a multi-channel image feature matrix; Step 2: Based on the variational autoencoder, extract features from the multi-channel image feature matrix to generate a potential spatial feature vector representing the airspace operation mode. Use a clustering method to divide scene clusters and identify at least typical airspace operation scenarios that are distinguished by traffic flow and weather. Step 3: Based on the typical airspace operation scenarios and flight data, evaluate the control load of the typical airspace operation scenarios, construct a scenario-driven sector partitioning set segmentation model, and decompose the original problem of the sector partitioning set segmentation model into a main problem and a pricing sub-problem. Step 4: Design a row and column generation algorithm to solve the sector partitioning set segmentation model. Combine the branch and bound algorithm and the heuristic algorithm to continuously search for new paths and finally output the optimal sector partitioning scheme.

2. The method for dynamic partitioning of terminal area spatial sectors based on multi-source data fusion according to claim 1, characterized in that, The multi-source airspace data, including airspace, traffic, and meteorological information, includes: Structured airspace data, including sector boundaries and restricted airspace information; Traffic data, including ADS-B track data and flight schedule data; Meteorological data, including data from Fengyun meteorological satellites and Doppler radar data.

3. The method for dynamic partitioning of terminal area spatial sectors based on multi-source data fusion according to claim 2, characterized in that, The feature extraction from the multi-channel image feature matrix based on the variational autoencoder includes: An encoder is used to map the input data to the mean and variance of a latent distribution; Reparameterization technique sampling of the latent space; A decoder is used to reconstruct the data from the latent variables; In the above process, the loss function is a weighted sum of reconstruction loss, KL divergence loss and VGG perception loss.

4. The method for dynamic partitioning of terminal area spatial sectors based on multi-source data fusion according to claim 3, characterized in that, The method of dividing scene clusters using clustering includes: Calculate the Davidson-Bolding index and the sum of squared errors within clusters curves corresponding to different cluster number K values. Select the K value corresponding to the minimum Davidson-Bolding index as the optimal number of clusters. At the same time, verify whether the inflection point of the sum of squared errors within clusters curve at the elbow point is consistent with the Davidson-Bolding index result, to ensure that the cluster number K can simultaneously meet the requirements of low intra-cluster dispersion and high inter-cluster separation. When classifying typical operating scenarios based on clustering results, category labels for traffic scenarios and weather scenarios are marked according to each scenario cluster and cluster center.

5. The method for dynamic partitioning of terminal area spatial sectors based on multi-source data fusion according to claim 4, characterized in that, Assess the control load under typical operational scenarios for each airspace, and construct a scenario-driven sector partitioning set segmentation model, including: The gridded control load function is constructed, and the mapping from the typical airspace operation scenario to the control load is formed as follows: F(N,J,W)=(ω N ·N+ω J ·J)(1+λW) in, N is the number of flights, ω N ω is the weighting coefficient for flight load, J is the number of route junctions, and ω is the weighting coefficient for flight load. J This is a weighting coefficient for flight load; (1+λW) is the weather correction factor, where W=0 when the weather is good and the load is only the number of flights, and W=1 when there is severe weather and the weather intensity λ factor corrects the overall load.

6. The method for dynamically dividing terminal area airspace sectors according to claim 4, characterized in that, The objective function of the main problem is as follows: Among them, C r For the cost of sector path r, C p Let x be the penalty coefficient for the slack variable P. r Choose a variable for the sector path. For slack variable P ij The penalty coefficient is R, which is the set of all sector paths, ij is the edge in the sector, and E is the set of edges in the sector; P and P ij All are slack variables, where: The constraints of the main problem include path quantity balance constraints and edge flow conservation constraints: The path number balance constraint is as follows: ∑ r∈R x r +P=K The edge flow conservation constraint is as follows: ∑ (i,j)∈E (i ij,r ·x r +θ ji,r ·x r )+P ij =0 Where θ ij,r This indicates the flow direction of path r on edge (i,j).

7. The method for dynamic partitioning of terminal area spatial sectors based on multi-source data fusion according to claim 6, characterized in that, The method further includes: calculating the shadow value of resources by maximizing the dual problem model of the main problem to obtain the marginal benefits of airspace capacity and regulatory load, as shown in the following formula: Where, π K The dual value of the path number constraint; π ij This is the dual value of the edge flow balance constraint.

8. The method for dynamic partitioning of terminal area spatial sectors based on multi-source data fusion according to claim 6, characterized in that, The designed row and column generation algorithm solves the sector partitioning set segmentation model, continuously searching for new paths by combining branch and bound algorithms and heuristic algorithms, and finally outputs the optimal sector partitioning scheme, including: First, an initial solution is selected from a finite subset of feasible sector partitioning paths, a linear relaxation model is constructed and solved to obtain the dual value of the path number constraint and the edge flow conservation constraint. Subsequently, based on the dual value, the test number is calculated, and a new sector configuration scheme that satisfies the airspace capacity and control load is generated through heuristic search. Feasible solutions with negative test numbers are added to the main problem. Finally, the main problem and subproblems are optimized through iterative loops until all test numbers are non-negative. At this point, the main problem reaches the optimal solution of linear relaxation, and the optimal sector partitioning scheme is obtained by branch and bound method.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it causes the electronic device to implement the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program that, when run on a computer, causes the computer to perform the method as described in any one of claims 1 to 8.

Images