Region-level aviation flow prediction method based on Mamba-GCN

Abstract

The invention provides a region-level aviation flow prediction method based on Mamb-GCN, and belongs to the technical field of air traffic flow prediction, and the method comprises the steps: constructing a Mamb-GNC collaborative network model; historical flight path data of a target airspace is collected, the target airspace is divided into space grids, the number of aircrafts in each space grid is counted, and the time feature and the space feature of each aircraft are coded to construct a space-time tensor; constructing a dynamic adjacency matrix and a dynamic weight map based on the 8-neighborhood topology of the space grid; inputting the space-time tensor and the dynamic weight graph into a Mamba-GCN collaborative network model for training, and optimizing model parameters; and preprocessing the aviation trajectory data of the target airspace acquired in real time, and inputting the preprocessed aviation trajectory data into the trained Mamba-GCN collaborative network model to obtain an aviation flow prediction result. According to the method, the long-time dependence of the aviation flow in the time dimension and the grid correlation in the space dimension can be captured, and the prediction efficiency is high.

Description

Technical Field

[0001] This application belongs to the technical field of air traffic flow prediction, and particularly relates to a regional air traffic flow prediction method based on Mamba-GCN. Background Art

[0002] With the continuous growth of global air passenger traffic, the accurate prediction of air traffic flow has become the key to optimizing airspace resource allocation and preventing air congestion. Traditional air traffic flow prediction mainly uses statistical methods, traditional machine learning methods, and deep learning methods.

[0003] Statistical methods mainly adopt time series analysis models, such as autoregressive moving average model ARMA, autoregressive integrated moving average model ARIMA, and seasonal autoregressive integrated moving average model SARIMA, etc. Although these time series analysis models can capture the periodic characteristics in time series data, they ignore the spatial dependence of adjacent regions and are difficult to effectively capture the spatio-temporal coupling relationship of flight flow in regional airspace. With the development of artificial intelligence, traditional machine learning methods such as wavelet neural network, support vector machine, and Bayesian network have been introduced into air traffic flow prediction. Although these machine learning methods have improved the prediction ability to a certain extent, they are difficult to reveal the potential load spatio-temporal correlation when dealing with large-scale data, and the prediction performance is limited. Deep learning methods have also been applied in traffic flow prediction. Models such as spatio-temporal graph convolutional network STGCN, GraphWaveNet, and adaptive graph convolutional recurrent network AGCRN realize the joint modeling of spatio-temporal characteristics of traffic data by combining graph convolutional network GCN and time convolutional network TCN or recurrent neural network RNN. However, these deep learning methods have the following problems: First, the graph convolutional network GCN depends on a fixed adjacency matrix and cannot capture the dynamic changes of the spatio-temporal domain topology with the flight flow. Second, the receptive field of the time convolutional network TCN is limited and it is difficult to capture long-term dependencies; although Transformer can model long-term dependencies, its quadratic computational complexity limits its application in large-scale grid scenarios.

[0004] In response to the above problems, existing research has tried to introduce a dynamic graph generation mechanism (such as DGCRN and D2STNN) or improve the pre-training framework (such as STDMAE), but there are still problems of high computational complexity or loss of spatial relationship accuracy, which cannot meet the requirements of air traffic flow prediction. Summary of the Invention

[0005] In a first aspect, an embodiment of this application provides a regional air traffic flow prediction method based on Mamba-GCN, including the following steps: S1. Use the Mamba model and the GCN model to construct a Mamba-GNC collaborative network model; S2. Collect the historical ADS-B flight track data of the target airspace, divide the target airspace into spatial grids according to the preset spatial resolution, count the number of aircraft in each spatial grid within the preset time granularity, and encode the time features and spatial features of each aircraft to construct a spatio-temporal tensor; S3. Define the basic connection relationship between grids based on the 8-neighborhood topology of the spatial grid, map the geographical distance to the initial connection weight through the Gaussian kernel function, and then dynamically adjust the initial connection weight in combination with the real-time traffic correlation to generate a dynamic adjacency matrix and a dynamic weight map; S4. Input the spatio-temporal tensor and the dynamic weight map into the Mamba-GCN collaborative network model for training to optimize the model parameters; S5. After preprocessing the aviation track data of the target airspace obtained in real time, input it into the trained Mamba-GCN collaborative network model to obtain the aviation traffic prediction result within the set time period of the target airspace.

[0006] Further, in step S1, the Mamba-GNC collaborative network model is constructed by interacting and stacking the Mamba model and the GCN model and introducing an auxiliary layer; The Mamba-GNC collaborative network model is sequentially provided with a first Mamba model layer, a GCN model layer, a first linear transformation layer, a Relu activation layer, a second Mamba model layer, a temporal convolutional layer, and a second linear transformation layer from the input side to the output side; The first Mamba model layer, the GCN model layer, the first linear transformation layer, the Relu activation layer, the second Mamba model layer, the temporal convolutional layer, and the second linear transformation layer; The first Mamba model layer is used to model the long-term dependence relationship of the time dimension of the input spatio-temporal tensor through a selective state space model and output a first hidden state sequence; The GCN model layer is used to aggregate the spatial features of the first hidden state sequence based on the dynamic adjacency matrix and perform neighborhood information propagation using Chebyshev polynomials to output spatially enhanced features; The first linear transformation layer is used to adjust the dimension of the spatially enhanced features to match the Relu activation layer; The Relu activation layer is used to enhance the expression ability of the spatially enhanced features after dimension adjustment using a non-linear transformation; The second Mamba model layer is used to perform secondary time dependence modeling on the spatially enhanced features with enhanced expression ability and extract high-order temporal features; The temporal convolutional layer is used to expand the time receptive field through one-dimensional dilated convolution, perform local time dimension feature extraction and fusion on the extracted high-order temporal features, and obtain local temporal features; A second linear transformation layer for mapping local temporal features to the output dimension.

[0007] Furthermore, in step S1, the first Mamba model layer and the second Mamba model layer use a selective state space model to construct a state transition equation and an output equation, and model the input as a time series as a hidden state; the specific process is as follows: The first Mamba model layer and the second Mamba model layer use a selective state space model to construct a state transition equation, construct an introduced state transition matrix and an input projection matrix, and convert the spatio-temporal tensor into a hidden state; The first Mamba model layer and the second Mamba model layer use a selective state space model to construct an output equation, and convert the spatio-temporal tensor and the hidden state into an output; Reshape the spatio-temporal tensor to match the requirements of the selective state space model; Capture historical time node information from the reshaped spatio-temporal tensor through a forward selective state space model to obtain a forward hidden state sequence; Capture future time node information from the reshaped spatio-temporal tensor through a subsequent selective state space model to obtain a backward hidden state sequence; Linearly concatenate the forward hidden state sequence and the backward hidden state sequence in the feature dimension to obtain a bidirectional concatenated hidden state, and then use a dimensionality-reducing linear transformation operation to map the bidirectional concatenated hidden state to a space matching the output dimension to obtain an output hidden state; Use the output equation to extract the predicted traffic value from the spatio-temporal tensor and the output hidden state.

[0008] Furthermore, in step S1, the GCN model layer performs spatial feature aggregation on the first hidden state sequence based on a dynamic adjacency matrix, and uses Chebyshev polynomials for neighborhood information propagation. The specific process of outputting spatially enhanced features is as follows: Introduce an identity matrix to the dynamic adjacency matrix to add self-loops; Calculate the degree of each grid node in the dynamic adjacency matrix after adding self-loops to obtain a degree matrix, and then calculate the dynamic square root inverse of the degree matrix; Use the dynamic adjacency matrix after adding self-loops and the dynamic square root inverse of the degree matrix to construct a dynamic normalized aggregation matrix; Obtain the node features at each time step from the spatio-temporal tensor to obtain a node feature matrix at each time step; Combine the node feature matrix with the dynamic adjacency matrix, and use the dynamic normalized aggregation matrix for aggregation to obtain a feature matrix, completing spatial feature enhancement; Introduce a weight matrix in the first linear transformation layer and multiply it with the aggregated feature matrix to complete the linear transformation; In the Relu activation layer, the activation function is used to perform nonlinear activation on the linearly transformed feature matrix to obtain the node feature matrix of the next time step.

[0009] Furthermore, the specific steps of step S2 are as follows: S21. Determine the target spatial area, the preset spatial resolution and the preset time granularity; S22. Obtain the ADS-B historical track data of the target airspace and perform data cleaning to parse out the number of aircraft, longitude and latitude, and time; S23. Divide the target space area into spatial grids according to the preset spatial resolution and by two-dimensional plane projection, and record the number of grids N; S24. Sum the number of aircraft in each spatial grid at each moment according to the preset time granularity; S25. Determine the position of each moment in the timeline of a day according to the time granularity as a moment-day feature, and encode it to obtain a moment-day feature code tod; S26. Determine the position of the day to which each moment belongs in the timeline of a week according to the time granularity as a day-week feature, and encode it to obtain a day-week feature code dow; S27. After normalizing the longitude and latitude of each grid at each moment in the ADS-B historical track data, the longitude feature and the latitude feature are obtained, and they are encoded respectively to obtain the longitude encoding feature lon and the latitude encoding feature lat; S28. The time-day feature code tod and the day-week feature code dow are used as time features, and the longitude feature code lon and the latitude feature code lat are used as spatial features; S29. The number of aircraft in the target space area at each moment, the time-day feature tod coding, the day-week feature dow coding, the longitude feature lon coding and the latitude feature lat coding are used as a spatiotemporal tensor.

[0010] Furthermore, the specific steps of data cleaning in step S22 are as follows: S221. Obtain the ADS-B historical track data of the target airspace, wherein the ADS-B historical track data includes the aircraft icao number, latitude and longitude, altitude, ground speed and time; S222. Set threshold constraints according to ICAO standards, mark the track data with acceleration exceeding the upper threshold of acceleration as abnormal track points and remove them, and mark the track data with vertical speed exceeding the upper threshold of vertical speed as abnormal altitude change data and remove them; S223. The original second-level track data is resampled to minute granularity using cubic spline interpolation to complete the alignment of the track data in the time dimension and obtain a standardized track data set.

[0011] Further, the specific steps of step S3 are as follows: S31. Define the connection relationship of each grid according to the 8-neighborhood rule for the spatial grid of the target airspace to obtain the initial topology; S32. Based on the initial topology, with each grid as a node and the actual geographical distance between grids as an edge, define the graph edge structure; S33. For the associated grids connected to each grid, calculate the initial connection weight through the Gaussian kernel function and add the initial connection weight to the edge of the graph edge structure; S34. Take the number of aircraft in the spatio-temporal tensor of the target airspace as historical traffic data, and calculate the traffic time series correlation of the rows and columns where the associated grids are located based on the historical traffic data; S35. Dynamically adjust the connection weight on the edge of the graph edge structure according to the actual geographical distance and traffic time series correlation between each grid and the corresponding associated grid to obtain the dynamic adjacency matrix, and generate a dynamic weight graph of N×N dimensions for each time step; where N represents the number of grids in the target airspace.

[0012] Further, the specific steps of step S4 are as follows: S41. Construct a loss function based on the difference between the predicted air traffic and the actual air traffic of the Mamba-GCN collaborative network model; S42. Input the spatio-temporal tensor and the dynamic weight graph of the target airspace into the Mamba-GCN collaborative network model for training, and optimize the parameters of the Mamba-GCN collaborative network model with the goal of minimizing the loss function during the training process; S43. Until the loss function converges or reaches the preset number of iterations, complete the training.

[0013] Further, the specific steps of step S5 are as follows: S51. Take the generation process of the spatio-temporal tensor and the generation process of the dynamic weight graph as the input network; S52. Provide the air traffic trajectory data of the target airspace obtained in real time to the input network for preprocessing to obtain the spatio-temporal tensor and the dynamic weight graph; S53. Input the spatio-temporal tensor and the dynamic weight graph into the trained Mamba-GCN collaborative network model to predict the air traffic within the set time period of the target airspace; S54. Map and convert the predicted air traffic into a predicted traffic heat map using the output network.

[0014] Further, the prediction result of the air traffic in step S53 is characterized in the form of a spatio-temporal tensor and a dynamic weight graph; In step S54, after denormalizing the spatial features in the spatio-temporal tensor of the predicted air traffic flow, perform spatial coordinate mapping and generate a heat map of aircraft flow for each predicted time step using a visualization output method.

[0015] As can be seen from the above technical solutions, the present application has the following advantages: In the regional air traffic flow prediction method based on Mamba-GCN provided by the present application, by constructing a Mamba-GCN collaborative network model and integrating spatio-temporal tensors and dynamic weight maps, the problems of insufficient dynamic spatial topology modeling and low long-term dependence capture efficiency in regional air traffic flow prediction are solved; the present application can predict air traffic flow and provide support for optimizing airspace resource allocation and preventing air congestion; through the collaborative effect of the Mamba model and the GCN model, both the long-term dependence in the time dimension and the dynamic topological structure in the space dimension are considered, improving the accuracy and efficiency of prediction; using ADS-B historical flight track data for model training can make full use of existing actual flight data and enhance the reliability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the present application, the drawings required for description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a schematic flow chart of the regional air traffic flow prediction method based on Mamba-GCN of the present invention.

[0018] Figure 2 It is a schematic diagram of the network structure of the Mamba-GNC collaborative network model of the present invention.

[0019] Figure 3 It is a schematic diagram of the grid coding rule in the embodiment of the present invention.

[0020] Figure 4 It is a heat map of aircraft flow at a certain moment in the dataset of the present invention.

[0021] Figure 5 It is a schematic diagram of the aircraft flow statistics at each time step in a day of the present invention.

[0022] Figure 6 It is a schematic diagram of the aircraft flow on each day of a week of the present invention.

[0023] Figure 7Schematic diagram for generating the graph topology structure of the present invention; where (a) represents the schematic diagram of the spatial grid, (b) represents the schematic diagram of defining nodes and edges, and (c) represents the schematic diagram of the topology structure.

[0024] Figure 8 Schematic diagram of the input network of the present invention.

[0025] Figure 9 Schematic diagram of the output network of the present invention.

[0026] Figure 10 Schematic diagram of the overall processing framework of the present invention. Detailed implementation manners

[0027] In the following, the specific steps of the regional-level air traffic flow prediction method based on Mamba-GCN will be described in detail, and various embodiments of the present disclosure will be described more comprehensively. The present disclosure can have various embodiments, and adjustments and changes can be made therein. However, it should be understood that there is no intention to limit the various embodiments of the present disclosure to the specific embodiments disclosed herein, but the present disclosure should be understood to cover all adjustments, equivalents, and / or alternative solutions that fall within the spirit and scope of the various embodiments of the present disclosure.

[0028] Exemplarily speaking, in the context of the continuous increase in global air passenger volume, accurately predicting air traffic flow has become increasingly crucial for optimizing airspace resource allocation and preventing air congestion. Traditionally, air traffic flow prediction methods mainly rely on statistical methods, traditional machine learning methods, and deep learning methods.

[0029] Statistical methods usually rely on time series analysis models, such as autoregressive moving average model (ARMA), autoregressive integrated moving average model (ARIMA), and seasonal autoregressive integrated moving average model (SARIMA). Although these models can capture the periodic characteristics in time series data, they have limitations in dealing with the spatio-temporal coupling relationship in regional airspace because they ignore the spatial dependence between adjacent regions. With the development of artificial intelligence technology, traditional machine learning methods such as wavelet neural networks, support vector machines, and Bayesian networks have also been applied to air traffic flow prediction. Although these methods have improved the prediction ability to some extent, when facing large-scale data, they are not sufficient in revealing potential spatio-temporal correlations, thus limiting the prediction performance. Deep learning methods have also been widely used in the field of traffic flow prediction, such as models like spatio-temporal graph convolutional network STGCN, GraphWaveNet, and adaptive graph convolutional recurrent network AGCRN. They achieve joint modeling of spatio-temporal features of traffic data by combining graph convolutional network GCN and temporal convolutional network TCN or recurrent neural network RNN. However, these deep learning methods also have some problems: on the one hand, the graph convolutional network GCN relies on a fixed adjacency matrix, which makes it unable to capture the spatio-temporal topological structure that changes dynamically with flight flows; on the other hand, the receptive field of the temporal convolutional network TCN is limited and it is difficult to capture long-term dependencies, while although the Transformer model can model long-term dependencies, its quadratic computational complexity limits its application in large-scale grid scenarios.

[0030] To address these problems, existing research has attempted to introduce dynamic graph generation mechanisms (such as DGCRN and D2STNN) or improve the pre-training framework (such as STDMAE). However, these methods still have some defects, such as excessively high computational complexity or loss of spatial relationship accuracy, which makes it difficult for them to meet the high-precision requirements of air traffic flow prediction.

[0031] To solve the above problems, this embodiment provides a regional-level air traffic flow prediction method based on Mamba-GCN. By constructing a Mamba-GCN collaborative network model and integrating spatio-temporal tensors and dynamic weight graphs, it effectively solves the problems of insufficient dynamic spatial topology modeling and low long-term dependence capture efficiency in regional-level air traffic flow prediction.

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0033] Please refer to Figure 1 The following is a flowchart of a regional air traffic flow prediction method based on Mamba-GCN in a specific embodiment. The method includes the following steps: S1. Use the Mamba model and the GCN model to construct a Mamba-GNC collaborative network model; It should be noted that constructing the Mamba-GCN collaborative network model integrates the advantages of time series modeling and spatial feature aggregation. By using the Mamba model and the GCN model, it is possible to capture both the long-term dependence relationship in the time dimension and the complex topological structure in the spatial dimension, improving the model's ability to model the spatio-temporal coupling relationship in air traffic flow data and providing a basis for subsequent prediction; S2. Collect the historical ADS-B flight track data of the target airspace, divide the target airspace into spatial grids according to a preset spatial resolution, count the number of aircraft in each spatial grid within a preset time granularity, and encode the time features and spatial features of each aircraft to construct a spatio-temporal tensor; It should be noted that collecting the historical ADS-B flight track data and constructing the spatio-temporal tensor to capture the spatio-temporal features in the data; dividing the target airspace into spatial grids and counting the number of aircraft, while encoding the time features and spatial features, can comprehensively describe the spatio-temporal distribution of air traffic flow, providing high-quality input data for model training and improving the prediction accuracy; S3. Define the basic connection relationship between grids based on the 8-neighborhood topology of the spatial grid, map the geographical distance to the initial connection weight through the Gaussian kernel function, and then dynamically adjust the initial connection weight in combination with the real-time traffic correlation to generate a dynamic adjacency matrix and a dynamic weight map; It should be noted that generating the dynamic adjacency matrix and the dynamic weight map can capture the dynamic correlation of traffic between grids; defining the basic connection relationship based on the 8-neighborhood topology of the spatial grid and dynamically adjusting the weight in combination with the geographical distance and real-time traffic correlation can timely reflect the change trend and spatial association of flight traffic in the airspace, enabling the model to adapt to the dynamic air traffic environment; S4. Input the spatio-temporal tensor and the dynamic weight map into the Mamba-GCN collaborative network model for training to optimize the model parameters; It should be noted that training the Mamba-GCN collaborative network model ensures the optimization of model parameters and the improvement of prediction performance; by inputting the spatio-temporal tensor and the dynamic weight map into the model for training and optimizing the model parameters with the goal of minimizing the loss function, the model can fully learn the spatio-temporal features and traffic patterns in the data, making the prediction results accurate and reliable; S5. After preprocessing the aviation trajectory data of the target airspace obtained in real time, input it into the trained Mamba-GCN collaborative network model to obtain the aviation flow prediction result within the set time period of the target airspace; It should be noted that the real-time prediction of the aviation flow in the target airspace realizes the accurate prediction of air traffic flow with high prediction efficiency; after preprocessing the aviation trajectory data obtained in real time and inputting it into the trained model, the aviation flow prediction result within the target time period can be quickly output, providing timely decision support for the aviation management department and facilitating the optimization of airspace resource allocation and the prevention of air congestion.

[0034] In this embodiment, by constructing a Mamba-GNC collaborative network model and using the time series modeling ability of the Mamba model and the spatial feature aggregation ability of the GCN model, the capture of the spatio-temporal characteristics of aviation flow is realized; combined with ADS-B data processing, spatio-temporal tensor construction, and dynamic adjacency matrix generation, the actual aviation data is converted into effective information suitable for model training, enabling the model to learn the complex laws of aviation flow in the real scenario and providing a basis for prediction.

[0035] Further, as a refinement and extension of the specific implementation manner of the above embodiment, in order to fully illustrate the specific implementation process in this embodiment, another regional-level aviation flow prediction method based on Mamba-GCN is provided, and this method includes the following steps: S1. Use the Mamba model and the GCN model to construct a Mamba-GNC collaborative network model; in step S1, the Mamba-GNC collaborative network model is constructed by interacting and stacking the Mamba model and the GCN model and introducing an auxiliary layer; The Mamba-GNC collaborative network model is sequentially provided with a first Mamba model layer, a GCN model layer, a first linear transformation layer, a Relu activation layer, a second Mamba model layer, a temporal convolutional layer, and a second linear transformation layer from the input side to the output side; The first Mamba model layer, the GCN model layer, the first linear transformation layer, the Relu activation layer, the second Mamba model layer, the temporal convolutional layer, and the second linear transformation layer; It should be noted that the Mamba model layer is good at capturing time dimension dependencies, while the GCN model layer focuses on spatial dimension modeling. The Mamba-GCN collaborative network adopts the strategy of interacting and stacking Mamba and GCN to deeply integrate spatio-temporal features. The network structure of the Mamba-GNC collaborative network model is as Figure 2 shown; S2. Collect the ADS-B historical track data of the target airspace, divide the target airspace into spatial grids according to the preset spatial resolution, count the number of aircraft in each spatial grid within the preset time granularity, and encode the time and space characteristics of each aircraft to construct a space-time tensor; the specific steps of step S2 are as follows: S21. Determine the target spatial area, the preset spatial resolution and the preset time granularity; S22. Obtain the ADS-B historical track data of the target airspace and perform data cleaning to parse out the number of aircraft, longitude and latitude, and time; the specific steps of data cleaning in step S22 are as follows: S221. Obtain the ADS-B historical track data of the target airspace, wherein the ADS-B historical track data includes the aircraft icao number, latitude and longitude, altitude, ground speed and time; It should be noted that the icao number is the four-letter identification code for the airport and airline; S222. Set threshold constraints according to ICAO standards, mark the track data with acceleration exceeding the upper threshold of acceleration as abnormal track points and remove them, and mark the track data with vertical speed exceeding the upper threshold of vertical speed as abnormal altitude change data and remove them; Specifically, threshold constraints are set according to ICAO standards, abnormal track points with acceleration exceeding 4.9m / s² are eliminated, and abnormal altitude change data with vertical speed exceeding 500ft / min are removed to ensure the accuracy and reliability of the data; S223. The original second-level track data is resampled to minute granularity using cubic spline interpolation method, and the track data is aligned in the time dimension to obtain a standardized track data set; For example, in terms of spatial dimension, the target airspace (22°-32°N, 112°-122°E) is divided into Divide the grid cells (about 27.5km×27.5km) and generate a unique spatial identifier for each grid according to the Earth Space Grid Coding Rules to achieve multi-level scalable indexing under the global spatial reference system. The grid coding rules are as follows: Figure 3 As shown; In the time dimension, with a 1-minute statistical window, calculate the number of ICAO numbers in each grid as the traffic characterization value to form a spatio-temporal traffic matrix; to adapt to the input requirements of the traffic prediction model, the data is stored in the standard format of the PEMs dataset to form an aircraft traffic data file and an adjacency matrix definition file respectively; the aircraft traffic data file stores the aircraft traffic data of all grids at all time steps, and the data dimension is (10080, 1600, 1), where 10080 represents the total time step length of the data, 1600 represents the total number of grids, and 1 represents 1 channel feature of each grid at each time step, that is, the aircraft traffic; visualize the aircraft traffic heat map corresponding to a certain moment in the dataset as Figure 4 shown; It should be noted that the discreteness of the track data in the time series is overcome by the spline interpolation method. Each record contains time information, icao number, longitude and latitude, and ground speed information, and the time resolution is uniformly 1 minute; The cleaned track data is shown in Table 1: Table 1: Track Data Table

[0036] S23. Divide the target space area into spatial grids according to the preset spatial resolution and using the two-dimensional plane projection method, and record the number of grids N; S24. Sum the number of aircraft in each spatial grid at each moment according to the preset time granularity; S25. Determine the position of each moment in the time line of a day according to the time granularity as the moment-day feature, and encode it to obtain the moment-day feature encoding tod; S26. Determine the position of the day to which each moment belongs in the time line of a week according to the time granularity as the day-week feature, and encode it to obtain the day-week feature encoding dow; S27. After normalizing the longitude and latitude of each grid at each moment in the ADS-B historical track data, obtain the longitude feature and latitude feature, and encode them respectively to obtain the longitude encoding feature lon and the latitude encoding feature lat; S28. Use the moment-day feature encoding tod and the day-week feature encoding dow as time features, and use the longitude feature encoding lon and the latitude feature encoding lat as spatial features; S29. Use the number of aircraft in the target space area at each moment, the moment-day feature tod encoding, the day-week feature dow encoding, the longitude feature lon encoding, and the latitude feature lat encoding as the spatio-temporal tensor; Exemplarily, the target area is divided into spatial grids with a resolution of 0.25°×0.25°, and the number of aircraft in each grid at each moment is counted according to a time granularity of 1 minute. In this way, the ADS-B track data is converted into a 3D tensor with the shape of T×N×D, where T represents the historical time step, N represents the number of grids in the target area, and D represents the number of features of each grid, that is, the aircraft flow; After being converted into the spatio-temporal data format, spatio-temporal feature encoding is performed according to the spatio-temporal characteristics of the ADS-B data; in terms of time features, first, based on the periodicity of the track data, such as Figure 5 and Figure 6 shown, according to the time granularity, calculate the position of each time step on the time line of a day, and then calculate the position of the day corresponding to each time step in a week according to the time granularity, and represent them with tod (time of day) and dow (day of week) respectively. Both tod and dow belong to [0,1], and tod and dow are used as the 2nd and 3rd dimensional features of each grid respectively; in terms of spatial features, normalize the longitude and latitude of the center point of each grid, and represent them with lon and lat respectively. Both lon and lat belong to [0,1], and lon and lat are used as the 4th and 5th dimensional features of each grid respectively; S3. Define the basic connection relationship between grids based on the 8-neighborhood topology of the spatial grid, map the geographical distance to the initial connection weight through the Gaussian kernel function, and then dynamically adjust the initial connection weight in combination with the real-time traffic correlation to generate a dynamic adjacency matrix and a dynamic weight map; the specific steps of step S3 are as follows: S31. Define the connection relationship of each grid according to the 8-neighborhood rule for the spatial grid of the target airspace to obtain the initial topology; S32. Based on the initial topology, and taking each grid as a node and the actual geographical distance between grids as an edge, define the graph edge structure; S33. For the associated grids connected to each grid, calculate the initial connection weight through the Gaussian kernel function and add the initial connection weight to the edge of the graph edge structure; Specifically, pre-define the basic connection relationship according to the grid 8-neighborhood topology, such as Figure 7 shown, the value of the edge between nodes represents the actual geographical distance of the grid center point; when performing dynamic weight adjustment, what is adjusted is the connection strength of the edge, without changing the connection relationship between nodes. The weight value is usually inversely proportional to the distance. The closer the distance, the greater the weight. Therefore, first use the Gaussian kernel function to map the geographical distance to the weight; the specific mapping formula is as follows:

[0037] Among them, represents the grid And The geographical Euclidean distance from the center point; Represents a learnable bandwidth parameter that controls the geographical attenuation rate, Represents the initial connection weight, which is a static value; S34. Use the number of aircraft in the spatio-temporal tensor of the target airspace as historical traffic data, and calculate the traffic time-series correlation of the rows and columns where the associated grid is located based on the historical traffic data; Specifically, the formula for calculating the traffic time-series correlation is as follows:

[0038] Among them, Represents the correlation coefficient of the traffic time series between grid i and grid j at time step t, which characterizes the correlation of the traffic time series between grids; Represents the grid i At time step τ The traffic observation value of, Represents the grid i In the time window To t The mean of the traffic observation values within; Represents the grid j At time step τ The traffic observation value of, Represents the grid j In the time window To t The mean of the traffic observation values within; τ represents the time step variable, which is used to traverse the time steps from To t; t represents the current time step; T represents the length of the time window, which is used to determine the number of historical time steps considered when calculating the correlation; S35. Dynamically adjust the connection weights on the edges of the graph edge structure according to the actual geographical distance and traffic time-series correlation of each grid and its corresponding associated grid, obtain a dynamic adjacency matrix, and generate a dynamic weight graph of N×N dimensions for each time step; where N represents the number of grids in the target airspace; Specifically, the formula for dynamically adjusting the connection weights is as follows:

[0039] Among them, Represents the dynamically adjusted connection weight, which is a dynamic value; Is a learnable weight coefficient that satisfies ; S4. Input the spatio-temporal tensor and the dynamic weight graph into the Mamba-GCN collaborative network model for training to optimize the model parameters; The specific steps of step S4 are as follows: Construct a loss function based on the difference between the predicted air traffic flow and the actual air traffic flow of the Mamba-GCN collaborative network model; Exemplarily, taking the target airspace divided into spatial grids as an example, the traffic observation value of each grid at time step is ; for a given historical observation sequence of time steps , where , the goal of the traffic prediction task is to learn the mapping function to predict the traffic sequence of the next time steps:

[0040] where represents the predicted traffic of all grids at the time step; The model is optimized by minimizing the loss between the predicted value and the true value, and the loss function is defined as follows:

[0041] S42. Input the spatio-temporal tensor and dynamic weight map of the target airspace into the Mamba-GCN collaborative network model for training, and optimize the parameters of the Mamba-GCN collaborative network model with the goal of minimizing the loss function during the training process; S43. Keep training until the loss function converges or reaches the preset number of iterations; S5. After preprocessing the real-time obtained air traffic trajectory data of the target airspace, input it into the trained Mamba-GCN collaborative network model to obtain the air traffic flow prediction result within the set time period of the target airspace; The specific steps of step S5 are as follows: S51. Take the generation process of the spatio-temporal tensor and the generation process of the dynamic weight map as the input network; Exemplarily, the input network is as Figure 8 shown; S52. Provide the real-time obtained air traffic trajectory data of the target airspace to the input network for preprocessing to obtain the spatio-temporal tensor and the dynamic weight map; S53. Input the spatio-temporal tensor and the dynamic weight map into the trained Mamba-GCN collaborative network model to predict the air traffic flow within the set time period of the target airspace; S54. Map the predicted air traffic flow to a predicted traffic heat map using the output network; The prediction result of the air traffic flow in step S53 is characterized in the form of a spatio-temporal tensor and a dynamic weight map; In step S54, after the spatial features in the spatio-temporal tensor of the predicted air traffic flow are de-normalized, spatial coordinate mapping is performed, and a heat map of aircraft flow for each predicted time step is generated using a visualization output method; Exemplarily, the output network is as Figure 9 shown; Combining the input network, the Mamba-GNC collaborative network model, and the output network, a schematic diagram of the overall processing framework is as Figure 10 shown; The first Mamba model layer in step S1 is used to model the long-term temporal dependencies in the time dimension of the input spatio-temporal tensor through a selective state space model, and output a first hidden state sequence; The GCN model layer is used to aggregate the spatial features of the first hidden state sequence based on a dynamic adjacency matrix, and use Chebyshev polynomials for neighborhood information propagation, and output spatially enhanced features; The first linear transformation layer is used to adjust the dimension of the spatially enhanced features to match the Relu activation layer; The Relu activation layer is used to enhance the expressive power of the spatially enhanced features after dimension adjustment using a non-linear transformation; The second Mamba model layer is used to perform secondary temporal dependency modeling on the spatially enhanced features with enhanced expressive power, and extract high-order temporal features; The time convolution layer is used to expand the temporal receptive field through one-dimensional dilated convolution, perform local temporal dimension feature extraction and fusion on the extracted high-order temporal features, and obtain local temporal features; The second linear transformation layer is used to map the local temporal features to the output dimension; In step S1, the first Mamba model layer and the second Mamba model layer use a selective state space model to construct a state transition equation and an output equation, and model the input as a time series as a hidden state; The specific process is as follows: The first Mamba model layer and the second Mamba model layer use a selective state space model to construct a state transition equation, introduce a state transition matrix and an input projection matrix, and convert the spatio-temporal tensor into a hidden state; The first Mamba model layer and the second Mamba model layer use a selective state space model and introduce a state projection matrix and an output projection matrix to construct an output equation, and convert the spatio-temporal tensor and the hidden state into an output; Reshape the spatio-temporal tensor to match the requirements of the selective state space model; Capture historical time node information from the reshaped spatio-temporal tensor through a forward selective state space model to obtain a forward hidden state sequence; Capture future time node information from the reshaped spatio-temporal tensor through a subsequent selective state space model to obtain a backward hidden state sequence; Linearly concatenate the forward hidden state sequence and the backward hidden state sequence in the feature dimension to obtain a bidirectional concatenated hidden state, and then use a dimensionality-reducing linear transformation operation to map the bidirectional concatenated hidden state to a space matching the output dimension to obtain an output hidden state; Extract the predicted flow value from the spatio-temporal tensor and the output hidden state using the output equation; Specifically, the Mamba model layer captures the long-range dependencies of time series through the selective state space model SSM. Compared with the local convolution of TCN, it can model global long-term time dependencies; compared with Transformer, the time complexity is reduced from to , with higher computational efficiency, effectively improving the training and inference capabilities of the model; The state space model SSM models the time series as hidden states through the state equation. The state transition equation is as follows:

[0042] The output equation is as follows:

[0043] Among them, is the state transition matrix, , , are the input projection matrix, the state projection matrix, and the output projection matrix respectively, represents the spatio-temporal tensor at the t-th time step, represents the hidden state at the t-th time step; When the Mamba module models the time series, it first needs to reshape the spatio-temporal tensor into , then capture the historical and future time node information through the forward and backward SSMs respectively to obtain bidirectional hidden states, and finally linearly concatenate and reduce the dimensionality of the bidirectional hidden states and output as shown; Among them, the forward SSM is expressed as follows:

[0044] The backward SSM is expressed as follows:

[0045] The linear concatenation of the bidirectional hidden states is expressed as follows:

[0046] Among them, is a dynamic parameter dependent on the input; In step S1, the GCN model layer performs spatial feature aggregation on the first hidden state sequence based on the dynamic adjacency matrix and uses Chebyshev polynomials for neighborhood information propagation. The specific process of outputting the spatially enhanced features is as follows: Introduce the identity matrix to the dynamic adjacency matrix to add self-loops; Through self-loops, ensure that each node incorporates its own features into the aggregation scope; Calculate the degree of each grid node in the dynamic adjacency matrix after adding self-loops to obtain the degree matrix, and then calculate the dynamic square root inverse of the degree matrix; Use the dynamic adjacency matrix after adding self-loops and the dynamic square root inverse of the degree matrix to construct the dynamic normalized aggregation matrix; Obtain the node features of each time step from the spatio-temporal tensor to get the node feature matrix for each time step; Combine the node feature matrix with the dynamic adjacency matrix and use the dynamic normalized aggregation matrix for aggregation to obtain the feature matrix, completing the spatial feature enhancement; Introduce a weight matrix in the first linear transformation layer and multiply it with the aggregated feature matrix to complete the linear transformation; In the Relu activation layer, use the activation function to perform non-linear activation on the feature matrix after linear transformation to obtain the node feature matrix for the next time step; Specifically, after calculating the dynamic weights, use Chebyshev polynomials to achieve efficient spatial feature aggregation. The specific formula is as follows:

[0047] Among them, represents the node features at the t-th time step, , is the number of nodes, is the feature dimension of the nodes; represents the degree matrix of the nodes, , is a diagonal matrix, and the elements on the diagonal represent the degree of each node after adding self-loops (i.e., the number of connected edges, including self-loops); , is the dynamic weighted adjacency matrix after adding self-loops; represents the weight matrix at the t-th time step; represents the activation function; represents the dynamic square root inverse of the degree matrix, taking the reciprocal of the square root of each diagonal element of D ; represents the dynamic normalized aggregation matrix; It should be noted that when constructing the Mamba-GNC collaborative network model, the parameters of the first Mamba model layer and the second Mamba model layer are initialized respectively, including the state transition matrix, the input projection matrix, the state projection matrix, and the output projection matrix; the graph convolution kernel weight matrix of the GCN model layer is initialized; the weights and biases of the first linear transformation layer and the second linear transformation layer are initialized; the convolution kernel weights of the temporal convolution layer are initialized. After the output of the first Mamba model layer is subjected to channel number transformation through the first linear transformation layer, it is input to the ReLU layer for non-linear activation to obtain the activated feature map; the activated feature map is input to the GCN model layer, and the predefined initial adjacency matrix is used for spatial feature aggregation to obtain the spatially aggregated feature map; the spatially aggregated feature map is input to the second Mamba model layer to capture the time dimension dependence and output the time series feature map; the time series feature map is input to the temporal convolution layer for local feature extraction in the time dimension to obtain the prediction result. A first linear transformation layer is introduced between the first Mamba model layer and the GCN model layer to map the high-dimensional time series features output by the Mamba model to the feature space matching the input requirements of the GCN model, aligning the time features and the spatial features in terms of dimension to ensure the smooth flow of data between different model layers; the ReLU activation layer receives the output of the first linear transformation layer, performs non-linear activation on the data, and enhances the non-linear expression ability of the model, and its output is used as the input of the GCN model layer; a temporal convolution layer is introduced after the second Mamba model layer to perform local feature extraction and fusion on the time series features output by the Mamba model in the local time dimension, capture the local correlation and trend in the time series, and its output is mapped to the dimension space of the prediction target through the second linear transformation layer to obtain the final air traffic flow prediction result.

[0048] The regional air traffic flow prediction method based on Mamba-GCN provided in the embodiments of this application can be applied to an electronic device. Those skilled in the art can understand that the structure of the electronic device involved in the embodiments of the present invention does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine some components, or have different component arrangements. In the embodiments of the present invention, the electronic device includes, but is not limited to, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described herein and / or claimed.

[0049] An electronic device may include a processor, an external memory interface, an internal memory, a Universal Serial Bus (USB) interface, a charging management module, a power management module, a battery, a wireless communication module, an audio module, a speaker, a microphone, a sensor module, buttons, a camera, a display screen, and a Subscriber Identity Module (SIM) card interface, etc.

[0050] It can be understood that the structure illustrated in the embodiments of this application does not constitute a specific limitation on the electronic device. In other embodiments of this application, the electronic device may include more or fewer components than those illustrated, or combine certain components, or split certain components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0051] The processor may include one or more processing units. For example, the processor may include a Central Processing Unit (CPU), etc., an Application Processor (AP), a modem processor, a Graphics Processing Unit (GPU), an Image Signal Processor (ISP), a controller, a memory, a video codec, a Digital Signal Processor (DSP), a baseband processor, and / or a Neural-Network Processing Unit (NPU), etc. Among them, different processing units may be independent devices or integrated in one or more processors.

[0052] Among them, the processor may be the nerve center and command center of the electronic device. The controller may generate operation control signals according to the instruction operation code and timing signals to complete the control of fetching and executing instructions.

[0053] A memory may also be provided in the processor for storing instructions and data. In some embodiments, the memory in the processor is a cache memory. This memory can save the instructions or data that the processor has just used or recycled. If the processor needs to use the instruction or data again, it can directly call it from this memory. This avoids repeated accesses, reduces the waiting time of the processor, and thus improves the system efficiency.

[0054] The above-mentioned electronic device implements the technical solution of the regional-level air traffic flow prediction method based on Mamba-GCN in this application, which uses the Mamba model and the GCN model to construct a Mamba-GNC collaborative network model; collects the ADS-B historical track data of the target airspace, divides the target airspace into spatial grids according to the preset spatial resolution, counts the number of aircraft in each spatial grid within the preset time granularity, and encodes the time features and spatial features of each aircraft to construct a spatio-temporal tensor; defines the basic connection relationship between grids based on the 8-neighborhood topology of the spatial grid, maps the geographical distance to the initial connection weight through the Gaussian kernel function, and then dynamically adjusts the initial connection weight in combination with the real-time traffic correlation to generate a dynamic adjacency matrix and a dynamic weight map; inputs the spatio-temporal tensor and the dynamic weight map into the Mamba-GCN collaborative network model for training to optimize the model parameters; preprocesses the real-time obtained air traffic track data of the target airspace and inputs it into the trained Mamba-GCN collaborative network model to obtain the air traffic flow prediction result within the set time period of the target airspace. By integrating the Mamba model and the GCN model to construct a collaborative network, combining data processing and dynamic weight adjustment, the problem of complex spatio-temporal features and dynamic data changes in air traffic flow prediction is solved; it can more accurately capture the long-term dependence relationship of air traffic flow in the time dimension and the grid correlation characteristics in the spatial dimension, thereby improving the accuracy and timeliness of prediction; at the same time, the generation of the dynamic adjacency matrix and the weight map enables the model to adapt to the real-time changing air traffic flow data with beneficial effects.

[0055] In the storage medium provided by this application, there is a program product that can implement the regional-level air traffic flow prediction method based on Mamba-GCN.

[0056] The regional-level air traffic flow prediction method based on Mamba-GCN includes: using the Mamba model and the GCN model to construct a Mamba-GNC collaborative network model; collecting the ADS-B historical track data of the target airspace, dividing the target airspace into spatial grids according to the preset spatial resolution, counting the number of aircraft in each spatial grid within the preset time granularity, and encoding the time features and spatial features of each aircraft to construct a spatio-temporal tensor; defining the basic connection relationship between grids based on the 8-neighborhood topology of the spatial grid, mapping the geographical distance to the initial connection weight through the Gaussian kernel function, and then dynamically adjusts the initial connection weight in combination with the real-time traffic correlation to generate a dynamic adjacency matrix and a dynamic weight map; inputs the spatio-temporal tensor and the dynamic weight map into the Mamba-GCN collaborative network model for training to optimize the model parameters; preprocesses the real-time obtained air traffic track data of the target airspace and inputs it into the trained Mamba-GCN collaborative network model to obtain the air traffic flow prediction result within the set time period of the target airspace.

[0057] In some possible embodiments, the Mamba-GCN-based regional air traffic prediction method of the present disclosure may be implemented in the form of a program product, which includes program code. When the program product runs on a terminal device, the program code is used to cause the terminal device to execute the steps according to various exemplary embodiments of the present disclosure described in the "Exemplary Method" section above in this specification.

[0058] The storage medium of the present disclosure may adopt any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0059] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A regional air traffic forecasting method based on Mamba-GCN, characterized in that: The steps include: S1. Use the Mamba model and the GCN model to build the Mamba-GNC collaborative network model; S2. Collect the ADS-B historical track data of the target airspace, divide the target airspace into spatial grids according to the preset spatial resolution, count the number of aircraft in each spatial grid within the preset time granularity, and encode the time and space characteristics of each aircraft to construct a space-time tensor; S3. Based on the 8-neighborhood topology of the spatial grid, the basic connection relationship between grids is defined, the geographic distance is mapped to the initial connection weight through the Gaussian kernel function, and the initial connection weight is dynamically adjusted in combination with the real-time traffic correlation to generate a dynamic adjacency matrix and a dynamic weight graph; S4. Input the spatiotemporal tensor and dynamic weight map into the Mamba-GCN collaborative network model for training and optimizing model parameters; S5. After preprocessing the aviation trajectory data of the target airspace acquired in real time, input it into the trained Mamba-GCN collaborative network model to obtain the aviation traffic forecast result within the set time period of the target airspace.

2. The regional air traffic prediction method based on Mamba-GCN according to claim 1 is characterized in that: In step S1, the Mamba-GNC collaborative network model is constructed by interactively stacking the Mamba model and the GCN model and introducing an auxiliary layer; The Mamba-GNC collaborative network model is provided with a first Mamba model layer, a GCN model layer, a first linear transformation layer, a Relu activation layer, a second Mamba model layer, a temporal convolution layer and a second linear transformation layer in sequence from the input side to the output side; The first Mamba model layer, the GCN model layer, the first linear transformation layer, the Relu activation layer, the second Mamba model layer, the temporal convolution layer, and the second linear transformation layer; The first Mamba model layer is used to model the long-term dependency of the input spatiotemporal tensor in the time dimension through a selective state space model and output the first hidden state sequence; The GCN model layer is used to aggregate the spatial features of the first hidden state sequence based on the dynamic adjacency matrix, and use Chebyshev polynomials to propagate neighborhood information and output spatial enhancement features; The first linear transformation layer is used to adjust the dimension of the spatial enhancement features to match the ReLU activation layer; Relu activation layer, used to enhance the expressiveness of the spatial enhancement features after dimension adjustment using nonlinear transformation; The second Mamba model layer is used to perform secondary time dependency modeling on the spatial enhancement features after the expression ability is enhanced, and extract high-order temporal features; The temporal convolution layer is used to expand the temporal receptive field through one-dimensional dilated convolution, extract and fuse the local temporal dimension of the extracted high-order temporal features, and obtain local temporal features; The second linear transformation layer is used to map the local temporal features to the output dimension.

3. The regional air traffic prediction method based on Mamba-GCN according to claim 2 is characterized in that: In step S1, the first Mamba model layer and the second Mamba model layer use the selective state space model to construct the state transfer equation and the output equation, and model the input as a time series as a hidden state; the specific process is as follows: The first Mamba model layer and the second Mamba model layer use the selective state space model to construct the state transfer equation, construct the introduced state transfer matrix and the input projection matrix, and convert the space-time tensor into a hidden state; The first and second Mamba model layers use the selective state space model to construct the output equation, converting the spatiotemporal tensor and hidden state into output; Reshape the space-time tensor to match the requirements of the selective state-space model; The forward selective state space model is used to capture the historical time node information from the reshaped spatiotemporal tensor to obtain the forward hidden state sequence; The information of future time nodes is captured from the reshaped spatiotemporal tensor through a subsequent selective state space model to obtain a backward hidden state sequence; Linearly concatenate the forward hidden state sequence and the backward hidden state sequence in the feature dimension to obtain a bidirectional concatenated hidden state, and then use a dimensionality-reduced linear transformation operation to map the bidirectional concatenated hidden state to a space that matches the output dimension to obtain an output hidden state; The predicted flow value is extracted from the spatiotemporal tensor and the output hidden state using the output equation.

4. The regional air traffic prediction method based on Mamba-GCN according to claim 3 is characterized in that: In step S1, the GCN model layer aggregates the spatial features of the first hidden state sequence based on the dynamic adjacency matrix, and uses Chebyshev polynomials to propagate neighborhood information. The specific process of outputting spatial enhanced features is as follows: Introducing the identity matrix to add self-loops to the dynamic adjacency matrix; Calculate the degree of each grid node in the dynamic adjacency matrix after adding the self-loop, obtain the degree matrix, and then calculate the dynamic square root inverse of the degree matrix; Construct a dynamic normalized aggregation matrix using the dynamic adjacency matrix after adding self-loops and the dynamic square root inverse of the degree matrix; Obtain the node features of each time step from the spatiotemporal tensor to obtain the node feature matrix of each time step; Combine the node feature matrix with the dynamic adjacency matrix and use the dynamic normalized aggregation matrix to aggregate to obtain the feature matrix and complete the spatial feature enhancement; The weight matrix is ​​introduced into the first linear transformation layer and multiplied with the aggregated feature matrix to complete the linear transformation; In the Relu activation layer, the activation function is used to perform nonlinear activation on the linearly transformed feature matrix to obtain the node feature matrix of the next time step.

5. The regional air traffic prediction method based on Mamba-GCN according to claim 1, characterized in that: The specific steps of step S2 are as follows: S21. Determine the target spatial area, the preset spatial resolution and the preset time granularity; S22. Obtain the ADS-B historical track data of the target airspace and perform data cleaning to parse out the number of aircraft, longitude and latitude, and time; S23. Divide the target space area into spatial grids according to the preset spatial resolution and by two-dimensional plane projection, and record the number of grids N; S24. Sum the number of aircraft in each spatial grid at each moment according to the preset time granularity; S25. Determine the position of each moment in the timeline of a day according to the time granularity as a moment-day feature, and encode it to obtain a moment-day feature code tod; S26. Determine the position of the day to which each moment belongs in the timeline of a week according to the time granularity as a day-week feature, and encode it to obtain a day-week feature code dow; S27. After normalizing the longitude and latitude of each grid at each moment in the ADS-B historical track data, the longitude feature and the latitude feature are obtained, and they are encoded respectively to obtain the longitude encoding feature lon and the latitude encoding feature lat; S28. The time-day feature code tod and the day-week feature code dow are used as time features, and the longitude feature code lon and the latitude feature code lat are used as spatial features; S29. The number of aircraft in the target space area at each moment, the time-day feature tod coding, the day-week feature dow coding, the longitude feature lon coding and the latitude feature lat coding are used as a spatiotemporal tensor.

6. The regional air traffic prediction method based on Mamba-GCN according to claim 5, characterized in that: The specific steps of data cleaning in step S22 are as follows: S221. Obtain the ADS-B historical track data of the target airspace, wherein the ADS-B historical track data includes the aircraft icao number, latitude and longitude, altitude, ground speed and time; S222. Set threshold constraints according to ICAO standards, mark the track data with acceleration exceeding the upper threshold of acceleration as abnormal track points and remove them, and mark the track data with vertical speed exceeding the upper threshold of vertical speed as abnormal altitude change data and remove them; S223. The original second-level track data is resampled to minute granularity using cubic spline interpolation to complete the alignment of the track data in the time dimension and obtain a standardized track data set.

7. The regional air traffic prediction method based on Mamba-GCN according to claim 5, characterized in that: The specific steps of step S3 are as follows: S31. Define the connection relationship of each grid in the target airspace according to the 8-neighborhood rule to obtain an initial topology; S32. Based on the initial topology, each grid is used as a node and the actual geographical distance between grids is used as an edge to define a graph edge structure; S33. For the associated grids connected to each grid, an initial connection weight is calculated by a Gaussian kernel function, and the initial connection weight is added to the edge of the graph edge structure; S34. The number of aircraft in the space-time tensor of the target airspace is collected as historical traffic data, and the traffic time series correlation of the row and column of the associated grid is calculated based on the historical traffic data; S35. The connection weights on the edges of the graph edge structure are dynamically adjusted according to the actual geographical distance and traffic timing correlation between each grid and the corresponding associated grid to obtain a dynamic adjacency matrix, and an N×N-dimensional dynamic weight graph is generated for each time step; where N represents the number of grids in the target airspace.

8. The regional air traffic prediction method based on Mamba-GCN according to claim 7, characterized in that: The specific steps of step S4 are as follows: S41. Construct a loss function based on the difference between the predicted air traffic and the actual air traffic of the Mamba-GCN collaborative network model; S42. Input the spatiotemporal tensor and dynamic weight map of the target spatial domain into the Mamba-GCN collaborative network model for training, and optimize the parameters of the Mamba-GCN collaborative network model with the goal of minimizing the loss function during the training process; S43. The training is completed until the loss function converges or the preset number of iterations is reached.

9. The regional air traffic prediction method based on Mamba-GCN according to claim 8, characterized in that: The specific steps of step S5 are as follows: S51. The generation process of the spatiotemporal tensor and the generation process of the dynamic weight map are used as input networks; S52. Provide the aviation trajectory data of the target airspace acquired in real time to the input network for preprocessing to obtain a spatiotemporal tensor and a dynamic weight map; S53. Input the spatiotemporal tensor and dynamic weight map into the trained Mamba-GCN collaborative network model to predict the aviation traffic within the set time period of the target airspace; S54. Convert the predicted air traffic into a predicted traffic heat map using the output network map.

10. The regional air traffic prediction method based on Mamba-GCN according to claim 9, characterized in that: In step S53, the prediction result of the air traffic is represented in the form of a spatiotemporal tensor and a dynamic weight map; In step S54, after the spatial features in the spatiotemporal tensor of the predicted aviation traffic are denormalized, spatial coordinate mapping is performed, and a heat map of aircraft traffic at each predicted time step is generated using a visual output method.

Images