Method for identifying key nodes in manned-unmanned aircraft integrated airspace based on GCN

Abstract

The present application relates to a kind of GCN-based manned unmanned aerial vehicle fusion operation airspace key node identification method, comprising the following steps: manned and unmanned aerial vehicle fusion operation airspace section network construction, manned and unmanned aerial vehicle heterogeneous traffic flow section network weighted assignment, section network node importance index calculation, key node identification model construction and training based on graph convolutional neural network, key node identification is carried out based on the graph convolutional neural network of training completion.The key node identification method of the present application is based on the two factors of manned / unmanned aerial vehicle flow and manned / unmanned traffic flow running time between sections under the fusion operation airspace, the weight of the edge in the section network is valued, the effective construction of the undirected weighted section network of fusion airspace can be realized;The key node identification model based on the graph convolutional neural network proposed can reasonably calculate the importance of node by combining a variety of importance indexes;Key nodes in fusion airspace network can be quickly identified.

Description

Technical Field

[0001] This invention relates to the field of airspace management technology for the integrated operation of manned and unmanned aircraft, and in particular to a method for identifying key nodes in the airspace for the integrated operation of manned and unmanned aircraft based on GCN. Background Technology

[0002] In manned and unmanned integrated airspace operations, air transportation systems exist as complex networks, with nodes representing different entities and edges representing the connections between them. For complex networks with varying topologies, the influence of each node on the overall network differs; nodes that play a crucial role in network performance are called critical nodes. In airway networks, critical airway nodes are vital for maintaining the network's structure and function, significantly impacting the safe and orderly operation of air traffic. If a sudden event causes a critical airway node to fail, its throughput capacity will be reduced or even lost, affecting surrounding nodes and triggering cascading failures, leading to large-scale airspace delays and a significant decline in network performance. Identifying critical nodes and implementing targeted protection to ensure their connectivity and resilience can enhance the reliability of the entire integrated airspace network, providing a reference for rational resource allocation and focusing on managing weak links. Against this backdrop, identifying critical nodes in airway networks has become a pressing challenge for the aviation industry and academia.

[0003] While existing technologies have achieved some results, they still have the following shortcomings: First, current methods for identifying key nodes in air traffic networks mainly focus on single manned aircraft traffic operation modes, and cannot identify key nodes in integrated manned and unmanned airspace networks; second, current methods for identifying key network nodes mostly start from static indicators describing the network in topology, ignoring the traffic distribution characteristics presented after the network carries traffic, and cannot well reflect the traffic effect on the integrated airspace network structure; third, current methods for identifying key network nodes mostly use a single indicator for evaluation, and the key nodes obtained from different indicators will differ, making it difficult to weigh and select in practical use. Summary of the Invention

[0004] This invention provides a method for identifying key nodes in the airspace for the integrated operation of manned and unmanned aircraft based on GCN.

[0005] This invention provides a method for identifying key nodes in the airspace for the integrated operation of manned and unmanned aerial vehicles (UAVs) based on graph convolutional neural networks (GCNs), comprising: step 1, constructing a segment network for the integrated operation of manned and UAVs; step 2, assigning weighted values ​​to the segment network of heterogeneous traffic flows between manned and UAVs; step 3, calculating the importance index of nodes in the segment network; step 4, constructing and training a key node identification model based on a graph convolutional neural network; and step 5, identifying key nodes based on the trained graph convolutional neural network.

[0006] Furthermore, in step 1, the method for constructing the airspace segment network for the integrated operation of manned and unmanned aircraft includes: constructing a segment network based on the topological relationship of segments in real data, with waypoints as network nodes and segments as edges.

[0007] Furthermore, in step 2, the method for weighting the heterogeneous traffic flow segment network of manned and unmanned aircraft includes: assigning weights to the edges in the segment network based on two factors: the manned / unmanned aircraft flow f between segments in the fused operating airspace and the manned / unmanned traffic flow operation time t, using the adjacency matrix A = (a ij ) N×N storage;

[0008]

[0009] Among them, a ij Let w represent the edge relationship from any node i to node j, N be the number of nodes in the network, and w be the edge relationship from node i to node j. ij Let be the weight of the edge connecting any two adjacent nodes i and j in the flight segment network;

[0010]

[0011] Among them, w ij As for the weight of road segments, For the time-of-day or all-day manned aircraft traffic of flight segment ij, This refers to the average operating time of manned aircraft on flight segment ij, divided by time period or throughout the day. For the time-sharing or all-day unmanned aerial vehicle traffic of flight segment ij, This refers to the average operating time of unmanned aerial vehicles in flight segment ij, divided into time periods or throughout the day.

[0012] Furthermore, in step 3, the calculation of the importance index of the network nodes of the flight segment includes: calculating the importance index of the nodes based on the flight segment network and the edge weights in the network, including weighted degree centrality, weighted betweenness centrality, compact centrality, weighted cycle ratio, and node bridging value;

[0013] For any node i in the network, its weighted degree k i for:

[0014]

[0015] Among them, a ji Let j be the edge relationship from node j to node i;

[0016] For any node i in the network, its weighted betweenness centrality BC i for:

[0017]

[0018] Among them, g st It is the sum of the edge weights on the shortest path from point s to point t; It is the sum of the edge weights on the path that passes through point i in the shortest path from point s to point t;

[0019] For any node i in the network, its compact centrality CC i for:

[0020]

[0021] Where, d ij Let n be the distance between node i and node j, and n be the number of nodes in the network.

[0022] For any node i in the network, its weighted cyclotron ratio r i for:

[0023]

[0024] Where node j and node h are any two nodes other than node i, r i Let k be the weighted circle ratio of node i. i A is the number of edges connected to node i, i.e., its degree. ij This represents the connection status between node i and node j; if there is a connection, it is 1, otherwise it is 0.

[0025] For any node i in the network, its bridging value Vc i The calculation includes the following steps:

[0026] Step A: Initialize the community. Treat each node as a community, initialize the community to which each node belongs, and calculate the modularity Q of the initial network.

[0027]

[0028] Among them, e vw Let a be the proportion of the edges connecting communities v and w in the entire network. v Let V be the proportion of edges in the entire network where only one point is inside community v.

[0029]

[0030]

[0031] Among them, C i and C j Let C represent the communities that node i and node j belong to in the network. i If v, then δ(C) i If ,v) is 1, otherwise δ(C) i If C(v) is 0, then C(v) is 0. j For w, then δ(C) j If δ(C,w) is 1, then δ(C) is 1. j w) is 0, M is half the sum of the weights of all edges in the network, k i Let i be the degree of node i;

[0032] Step B: For each node i, calculate the modularity gain ΔQ after moving it to the adjacent community J. If ΔQ>0, move node i to the adjacent community J that maximizes the modularity gain, update the community where node i is located, and repeat step B until the modularity can no longer be increased by moving nodes.

[0033] Step C: Based on the final community division results, assign nodes to their respective communities; the final bridging value Vc for node i. i for:

[0034] Vc i =∑ J Q iJ ;

[0035] Among them, Q iJ Q represents the value indicating whether node i belongs to community J. If node i belongs to community J, then Q... iJ =1, otherwise 0.

[0036] Furthermore, in step 4, the method for constructing and training the key node recognition model based on the graph convolutional neural network includes: step 4.1 using a propagation model to calculate the true importance value of each node in the trained nodes; step 4.2 constructing a graph convolutional neural network model; and step 4.3 training the key node recognition model of the graph convolutional neural network.

[0037] Furthermore, in step 4.1, the calculation of the true importance value of each node in the trained nodes using the propagation model includes: setting the SIR model parameters according to the established flight segment network, gradually simulating the propagation process when each node acts as a source of infection, and recording the number of other nodes it infects as the true importance value label y of the node.

[0038] Furthermore, in step 4.2, the constructed graph convolutional neural network model includes: an input layer, a graph convolutional layer, a deactivated layer following the aircraft, and an output layer.

[0039] Furthermore, the input layer of the graph convolutional neural network includes: taking the network nodes of the flight segment network, the adjacency matrix A, the feature matrix X composed of the node importance index, and the node importance true value label y as the input of the graph convolutional neural network.

[0040] Furthermore, the graph convolutional layer in the graph convolutional neural network includes: receiving input from the input layer, learning the vector representation of each node using the graph convolutional neural network, capturing the features of the node and its neighbors, and outputting a new feature matrix Z;

[0041]

[0042] Where σ is the activation function, using the ReLU function. It is about the angle matrix. W is the weight matrix formed by adding self-connections to the adjacency matrix A.

[0043] Furthermore, the deactivation layer in the graph convolutional neural network includes: receiving the feature matrix output by the graph convolutional layer, setting the output of a portion of the neurons to zero with the aircraft, and finally obtaining matrix H;

[0044] H = Ms⊙Z;

[0045] Where Z is the output of the deactivated layer of the aircraft, Ms is a binary mask matrix with the same shape as Z, and ⊙ is an element-wise multiplication operation.

[0046] Furthermore, the output layer of the graph convolutional neural network includes: receiving the output of the deactivated layer of the aircraft, performing a combination of feedforward and backpropagation, adjusting the network parameters by calculating the gradient of the loss function, and finally outputting the result.

[0047]

[0048] W out It is the weight matrix of the fully connected layer, b out It is the bias of the fully connected layer.

[0049] Furthermore, in step 4.3, training the key node identification model based on a graph convolutional neural network includes: inputting the adjacency matrix A of the graph structure of the flight segment network, the feature matrix X composed of node importance indices, and the true importance value labels y of the nodes into the graph convolutional neural network, and predicting values ​​during training. Training ends when the loss value converges with the true value label y.

[0050] Furthermore, in step 5, the method for identifying key nodes based on the trained graph convolutional neural network includes: constructing a segment network model for the segment network for which key nodes need to be identified; assigning weighted values ​​to the edges of the segment network; calculating the node importance index of the segment network; inputting the node importance into the trained graph convolutional neural network model for node importance prediction; sorting the node importance values ​​in reverse order; and selecting the top 5% as key nodes.

[0051] The beneficial effects of this method are:

[0052] (1) The proposed method assigns weights to edges in the segment network based on two factors: manned / unmanned aircraft traffic flow and manned / unmanned traffic flow operation time between segments under the integrated airspace, which can effectively construct an undirected weighted segment network in the integrated airspace.

[0053] (2) The proposed key node identification model based on graph convolutional neural network can reasonably calculate the importance of nodes by combining multiple importance indicators;

[0054] (3) The proposed method for identifying key nodes in the integrated airspace of manned and unmanned aircraft based on GCN can quickly identify key nodes in the integrated airspace network.

[0055] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0056] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0057] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0058] Figure 1 This invention illustrates the relationship between the components of the method for identifying key nodes in the airspace of manned and unmanned aerial vehicles based on GCN, according to an embodiment of the present invention.

[0059] Figure 2 The diagram illustrates the process of constructing and training a key node recognition model based on a graph convolutional neural network according to an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] In this embodiment, the relationships between the components of the method for identifying key nodes in the fusion operational airspace of manned and unmanned aerial vehicles based on graph convolutional neural networks (GCN) are as follows: Figure 1 As shown, the historical data required for constructing the airspace segment network for integrated manned and unmanned aerial vehicle (UAV) operations, weighting the heterogeneous traffic flow segment network for manned and UAVs, and calculating the importance index of segment network nodes is the segment network traffic data from a certain region in China on April 1, 2019. The process of constructing and training a key node identification model based on a graph convolutional neural network can be described as follows: Figure 2 As shown.

[0062] In this embodiment, the method for identifying key nodes in the airspace for the integrated operation of manned and unmanned aircraft based on GCN includes: Step 1, constructing a segment network for the integrated operation airspace of manned and unmanned aircraft; Step 2, assigning weighted values ​​to the segment network of heterogeneous traffic flows of manned and unmanned aircraft; Step 3, calculating the importance index of nodes in the segment network; Step 4, constructing and training a key node identification model based on a graph convolutional neural network; Step 5, identifying key nodes based on the trained graph convolutional neural network.

[0063] Further, in step 1, the airspace segment network for the integrated operation of manned and unmanned aerial vehicles is constructed by using waypoints as network nodes and segments as edges based on the topological relationships of segments in real data.

[0064] Further, in step 2, the heterogeneous traffic flow segment network for manned and unmanned aircraft is weighted. The method is as follows: based on two factors—the manned / unmanned aircraft flow rate *f* and the manned / unmanned traffic flow operation time *t* between segments in the integrated operating airspace—the weights of the edges in the segment network are assigned, using the adjacency matrix A = (a ij ) N×N Storage; In specific application scenarios, an undirected weighted network can be constructed based on waypoints and flight segments in actual data. Based on 45,120 flight segment traffic data points, 261 waypoints are extracted as network nodes. Simultaneously, weights are assigned to the edges in the flight segment network, using an adjacency matrix A = (a ij ) 261×261 store;

[0065]

[0066] Among them, a ij Let w represent the edge relationship from any node i to node j, N be the number of nodes in the network, and w be the edge relationship from node i to node j. ij Let be the weight of the edge connecting any two adjacent nodes i and j in the flight segment network;

[0067]

[0068] Among them, w ij As for the weight of road segments, For the time-of-day or all-day manned aircraft traffic of flight segment ij, This refers to the average operating time of manned aircraft on flight segment ij, divided by time period or throughout the day. For the time-sharing or all-day unmanned aerial vehicle traffic of flight segment ij, This refers to the average operating time of unmanned aerial vehicles in flight segment ij, divided into time periods or throughout the day.

[0069] Further, in step 3, the importance index of the flight segment network nodes is calculated. This is done by calculating the weighted degree centrality, weighted betweenness centrality, compact centrality, weighted cycle ratio, and node bridging value of the flight segment network based on the constructed undirected weighted network and its edge weights.

[0070] For any node i in the network, its weighted degree centrality k i for:

[0071]

[0072] Among them, a ji Let j be the edge relationship from node j to node i;

[0073] For any node i in the network, its weighted betweenness centrality BC i for:

[0074]

[0075] Among them, g st It is the sum of the edge weights on the shortest path from point s to point t; It is the sum of the edge weights on the path that passes through point i in the shortest path from point s to point t;

[0076] For any node i in the network, its compact centrality CC i for:

[0077]

[0078] Where, d ij Let n be the distance between node i and node j, and n be the number of nodes in the network.

[0079] For any node i in the network, its weighted cyclotron ratio r i for:

[0080]

[0081] Where node j and node h are any two nodes other than node i, r i Let k be the weighted circle ratio of node i. i A is the number of edges connected to node i, i.e., its degree. ij This represents the connection status between node i and node j; if there is a connection, it is 1, otherwise it is 0.

[0082] For any node i in the network, its bridging value Vc i The calculation includes the following steps:

[0083] Step A: Initialize the community. Treat each node as a community, initialize the community to which each node belongs, and calculate the modularity Q of the initial network.

[0084]

[0085] Among them, e vw Let a be the proportion of the edges connecting communities v and w in the entire network. v Let V be the proportion of edges in the entire network where only one point is inside community v.

[0086]

[0087]

[0088] Among them, C i and C j Let C represent the communities that node i and node j belong to in the network. i If v, then δ(C) i If ,v) is 1, otherwise δ(C) i If C(v) is 0, then C(v) is 0. j For w, then δ(C) j If δ(C,w) is 1, then δ(C) is 1. j w) is 0, M is half the sum of the weights of all edges in the network, k i Let i be the degree of node i;

[0089] Step B: For each node i, calculate the modularity gain ΔQ after moving it to the adjacent community J. If ΔQ>0, move node i to the adjacent community J that maximizes the modularity gain, update the community where node i is located, and repeat step B until the modularity can no longer be increased by moving nodes.

[0090] Step C: Based on the final community division results, assign nodes to their respective communities; the final bridging value Vc for node i. i for:

[0091] Vc i =∑ J Q iJ ;

[0092] Among them, Q ij Q represents the value indicating whether node i belongs to community J. If node i belongs to community J, then Q... iJ =1, otherwise 0.

[0093] Furthermore, in step 4, the key node recognition model based on graph convolutional neural networks is constructed and trained, and the method is as follows:

[0094] 4.1 Calculate the true importance value of each node in the training nodes using the propagation model;

[0095] 4.2 Construct a graph convolutional neural network model, which includes: an input layer, a graph convolutional layer, an aircraft-inactivated layer, and an output layer;

[0096] 4.3 Training the key node recognition model of the graph convolutional neural network.

[0097] Further, in step 4.1, in a specific application scenario, the SIR propagation model is run to calculate the importance value of each node (the number of other nodes that the node can influence) as the data true value label y for model training.

[0098] Further, in step 4.2, a graph convolutional neural network model is constructed, including: an input layer, a graph convolutional layer, a deactivated layer following the aircraft, and an output layer.

[0099] Furthermore, the network structure diagram of the flight segment (network nodes, adjacency matrix A) and the importance index of each node (weighted degree centrality k) will be further analyzed. i Weighted betweenness centrality BC i Close centrality CC i Weighted circle ratio r i Bridge value Vc i The feature matrix X, composed of the node importance values ​​and the true labels y, are used as the input layer of the graph convolutional neural network.

[0100] Furthermore, in the graph convolutional neural network, the graph convolutional layer receives input from the input layer, uses the graph convolutional neural network to learn the vector representation of each node, and outputs a new feature matrix Z;

[0101]

[0102] Where σ is the activation function, using the ReLU function. It is about the angle matrix. The adjacency matrix A is formed by adding self-connections, and W is the weight matrix;

[0103] Furthermore, in the graph convolutional neural network, the feature matrix output by the graph convolutional layer is received by the deactivated layer along with the aircraft, and the output of a portion of the neurons is set to zero along with the aircraft to obtain matrix H;

[0104] H = Ms⊙Z;

[0105] Where Z is the output of the deactivated layer of the aircraft, Ms is a binary mask matrix with the same shape as Z, and ⊙ is an element-wise multiplication operation;

[0106] Furthermore, in the graph convolutional neural network, the output layer receives the output from the deactivated layer of the aircraft, performs a combination of feedforward and backpropagation, and adjusts the network parameters by calculating the gradient of the loss function in conjunction with the true value label y, ultimately outputting the final result.

[0107]

[0108] W out It is the weight matrix of the fully connected layer, b out It is the bias of the fully connected layer.

[0109] Predicted values ​​during training Training ends when the loss value converges with the true value label y; the node prediction values ​​are sorted in descending order, and the top 5% of nodes are selected as key nodes.

[0110] Further, in step 4.3, a key node identification model based on a graph convolutional neural network is trained. The method is as follows: the adjacency matrix A of the graph structure of the flight segment network, the feature matrix X composed of node importance indices, and the true importance value labels y of the nodes are input into the graph convolutional neural network. During training, the predicted values... Training ends when the loss value converges with the true value label y.

[0111] Further, step 5 involves identifying key nodes based on the trained graph convolutional neural network, including: constructing a segment network model for the segment network for which key nodes need to be identified, assigning weighted values ​​to the edges of the segment network, calculating the node importance index of the segment network, inputting the node importance into the trained graph convolutional neural network model for node importance prediction, sorting the node importance values ​​in reverse order, and selecting the top 5% as key nodes.

[0112] In summary, the method for identifying key nodes in the airspace of manned and unmanned aircraft based on graph convolutional neural networks (GCN) adopted in this invention can integrate different traffic operation modes of manned and unmanned aircraft into the same network and load traffic distribution onto the network; it can also comprehensively calculate the node importance value by combining multiple node importance indicators, which can effectively promote the identification of key nodes in a more reasonable airway network.

[0113] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A method for identifying key nodes in the airspace for the integrated operation of manned and unmanned aerial vehicles based on GCN, characterized in that, include: Step 1: Construction of airspace segment network for integrated operation of manned and unmanned aerial vehicles; Step 2: Weighted assignment of heterogeneous traffic flow segments between manned and unmanned aircraft; Step 3: Calculate the importance index of network nodes in the flight segment; Step 4: Construction and training of a key node recognition model based on graph convolutional neural networks; Step 5: Identify key nodes based on the trained graph convolutional neural network; In step 1, the method for constructing the airspace segment network for the integrated operation of manned and unmanned aircraft includes: constructing a segment network based on the topological relationship of segments in real data, with waypoints as network nodes and segments as edges; In step 2, the method for weighted assignment of heterogeneous traffic flow segment networks for manned and unmanned aircraft includes: Based on the flow of manned / unmanned aircraft in the sub-segments of the integrated operation airspace Manned / unmanned traffic flow travel time Two factors are used to assign weights to edges in the segment network, using an adjacency matrix. storage; ； in, For any node To the node Edge relationship, The number of nodes in the network. For any adjacent node in the flight segment network and The weight of the connected edges; ； in, As for the weight of road segments, For the segment Manned aircraft traffic in different time periods or throughout the day, For manned aircraft operating on flight segments at different times or throughout the day Average runtime For the segment Traffic flow of unmanned aerial vehicles in different time periods or throughout the day. For unmanned aerial vehicles to operate in different time periods or throughout the day on different flight segments The average runtime.

2. The key node identification method according to claim 1, characterized in that, In step 3, the calculation of the importance index of the network nodes of the flight segment includes: The importance of nodes is calculated based on the segment network and the edge weights in the network, including weighted degree centrality, weighted betweenness centrality, compact centrality, weighted cycle ratio, and node bridging value. For any node in the network Its weighted degree centrality for: ； in, For nodes To the node Edge relations; For any node in the network Its weighted betweenness centrality for: ； in, For point Time The sum of edge weights on the shortest path; For point Time Points passed through in the shortest path The sum of the edge weights on the path; For any node in the network Its close centrality for: ； in, For nodes and nodes The distance between them The number of nodes in the network; For any node in the network Its weighted circle ratio for: ； Among them, nodes and nodes To remove nodes Any two external nodes, For nodes The weighted circle ratio, For nodes The number of connected edges, also known as degree. For nodes and nodes The connection status is 1 if there is a connection, otherwise 0. For any node in the network Its bridging value The calculation includes the following steps: Step A: Initialize communities. Treat each node as a community, initialize the community to which each node belongs, and calculate the modularity of the initial network. ; ； in, For the club and The proportion of edges connecting the edges in the entire network. There is only one point in the club The proportion of internal edges in the entire network. ； ； in, and They are nodes With nodes In online communities, if for ,but =1, otherwise If it is 0, for ,but =1, otherwise =0, It is half the sum of the edge weights in the network. For nodes The degree; Step B, for each node Calculate how to move it to an adjacent community Modularity gain after ,if Then the node Move to the neighboring community that maximizes the modularity gain. Update nodes For the club you belong to, repeat step B until you can no longer increase the modularity by moving nodes; Step C: Based on the final community division results, assign the nodes to their respective communities; final nodes Bridge value for: ； in, As representative node Does it belong to a club? The value of the node Belongs to the club ,but Otherwise, it is 0.

3. The key node identification method according to claim 2, characterized in that, In step 4, the method for constructing and training the key node recognition model based on graph convolutional neural networks includes: Step 4.1 Use the propagation model to calculate the true importance value of each node in the trained nodes; Step 4.2 Construct a graph convolutional neural network model; Step 4.3 Train a key node recognition model based on graph convolutional neural networks.

4. The key node identification method according to claim 3, characterized in that, In step 4.1, calculating the true importance value of each node in the trained nodes using the propagation model includes: Based on the established flight segment network, SIR model parameters are set, and the propagation process of each node as a source of infection is simulated step by step. The number of other nodes infected by each node is recorded as the true value label of the node's importance. .

5. The key node identification method according to claim 4, characterized in that, In step 4.2, constructing the graph convolutional neural network model includes: Input layer, graph convolutional layer, aircraft-inactivated layer, output layer; The input layer of the graph convolutional neural network includes: network nodes and adjacency matrix of the flight segment network. The feature matrix consists of node importance indicators. Node importance true value label As input to a graph convolutional neural network; The graph convolutional layer in the graph convolutional neural network includes: receiving input from the input layer, learning the vector representation of each node using the graph convolutional neural network, capturing the features of the node and its neighbors, and outputting a new feature matrix. ; ； in, For the activation function, use function, It is about the angle matrix. Adjacency matrix Adding the matrix formed by self-connection, It is a weight matrix; The deactivation layer in the graph convolutional neural network includes: receiving the feature matrix output by the graph convolutional layer, setting the output of a portion of the neurons to zero according to the aircraft, and finally obtaining the matrix. ; ； in, As the output of the aircraft's deactivated layer, To and Binary mask matrices with the same shape, This is an element-wise multiplication operation; The output layer of the graph convolutional neural network includes: receiving the output of the deactivated layer of the aircraft, performing a combination of feedforward and backpropagation, and then... (the sentence is incomplete and likely refers to a function that integrates with node labels). By combining the calculation of the gradient of the loss function to adjust the network parameters, the final output is obtained. ; ； It is the weight matrix of the fully connected layer. It is the bias of the fully connected layer.

6. The key node identification method according to claim 5, characterized in that, In step 4.3, training the key node recognition model based on a graph convolutional neural network includes: The adjacency matrix of the flight segment network The feature matrix consists of node importance indices. Node importance true value label In a convolutional neural network with an input graph, the predicted value during training. With true value label Training ends when the loss value converges.

7. The key node identification method according to claim 6, characterized in that, In step 5, the method for key node identification based on the trained graph convolutional neural network includes: For the flight segment network that needs to identify key nodes, a flight segment network model is constructed, the edges of the flight segment network are weighted and assigned, the node importance index of the flight segment network is calculated, and the node importance is predicted by inputting it into the trained graph convolutional neural network model. The node importance values ​​are sorted in reverse order, and the top 5% are selected as key nodes.

Images