A traffic flow pattern recognition-driven refined traffic flow prediction method

Abstract

This invention discloses a traffic flow pattern recognition-driven refined traffic flow prediction method, relating to the fields of air traffic situation awareness and traffic flow management. By efficiently extracting spatiotemporal features from flight trajectories, accurately identifying local flow patterns, and constructing a graphical model based on representative flow patterns, fine-grained prediction of traffic flow within the airspace is achieved, thereby significantly improving the precision and dynamic response capability of local area control. It is particularly suitable for deep analysis of flight trajectory data within airspace sectors, providing accurate auxiliary decision support for air traffic control, route planning, and airspace design.

Description

Technical Field

[0001] This invention relates to the field of air traffic situational awareness and flow management technology, and in particular to a method for refined flow prediction driven by air traffic flow pattern recognition. Background Technology

[0002] With the continuous growth of global air traffic and the increasing density of flights within airspace, flight trajectory clustering and traffic prediction technologies in terminal areas have, to some extent, met the needs of flight scheduling and safety supervision. However, existing flight trajectory clustering algorithms mainly target relatively fixed and standardized flight trajectories within terminal areas. These algorithms are typically based on pre-defined morphological assumptions and fixed feature extraction methods, such as distance metrics, density estimation, and hierarchical clustering methods. Since flight trajectories in terminal areas often follow pre-planned routes, their operational status and spatial layout are relatively stable, allowing these traditional clustering methods to effectively extract the main flow patterns within the trajectories. However, these algorithms have significant limitations when dealing with flight trajectories within airspace sectors. Flight trajectories within sectors are affected by mixed operations across multiple phases, including takeoff, approach, and cruise, exhibiting high diversity and spatiotemporal continuity. Traditional methods tend to overlook subtle differences in flow patterns, resulting in overly coarse clustering results that fail to reveal the complex flow structures within local airspace, thus providing limited information for local airspace control and refined management.

[0003] Meanwhile, existing air traffic flow forecasting methods primarily focus on overall flow trends and macro-level statistics and predictions. These forecasting models typically perform data statistics and time series analysis based on overall phenomena, employing models such as Autoregressive Integral Moving Average (ARIMA), Support Vector Regression (SVR), and early deep learning models like Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) for overall flow modeling. Their main goal is to capture the overall flow change trend in the airspace. However, these deep learning-based forecasting models do not consider the impact of airspace structure on flow forecasting. Currently, the latest forecasting methods are mostly based on Graph Neural Networks (GNNs). By defining nodes in the airspace and the relationships between them, GNNs can effectively capture the complex topological structure and spatial dependencies in the airspace. This method can reflect the complexity of the actual airspace and improve forecast accuracy. However, in the graphing strategies of these GNN-based forecasting models, nodes are usually defined as sectors or large areas of airspace, ignoring the spatial distribution and flow direction of flow patterns within a sector. Therefore, they cannot accurately reveal the specific distribution and dynamic characteristics of each flow pattern at a fine-grained level.

[0004] In summary, while almost all existing traffic flow prediction methods, exemplified by the aforementioned schemes, can reflect overall traffic conditions well, their prediction data often lacks sufficient detailed information in local airspace management, especially when dynamically controlling traffic at the sector level. This hinders the implementation of precise air traffic management and route optimization. Furthermore, in existing technologies, flow pattern identification and traffic flow prediction are often treated as two independent research tasks. Flow pattern identification focuses on analyzing the internal structure of flight trajectories, extracting representative tracks through clustering methods, and revealing the movement patterns of aircraft within the airspace. Traffic flow prediction, on the other hand, emphasizes using statistical models or deep learning methods to predict the future number of flights and their flow trends within the airspace. Although there is some overlap in theory and methodology, the fact that flow pattern identification reveals diverse and directional subtle flow patterns in local airspace, while traffic flow prediction models tend to capture macroscopic statistical trends, makes it difficult to achieve effective synergy in data representation and feature fusion, thus limiting the application effectiveness of fine-grained airspace traffic flow prediction.

[0005] Therefore, how to further improve the deep analysis scheme of flight trajectory data within airspace sectors, thereby enhancing the precision and dynamic response capability of local area control, has become a topic that needs further research. Summary of the Invention

[0006] The embodiments of the present invention provide a traffic flow pattern recognition-driven refined traffic prediction method, which further improves the deep analysis scheme of flight trajectory data within airspace sectors, thereby enhancing the precision of local area control and dynamic response capability.

[0007] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0008] A traffic flow pattern recognition-driven refined traffic prediction method includes:

[0009] Flight trajectory data is acquired from the Automatic Dependent Surveillance-Broadcast (ADS-B) system and preprocessed, wherein data is separated according to the spatiotemporal characteristics of the flight trajectory data during the sub-preprocessing process; specific sector range data can be obtained from the Air Traffic Management System (ATM).

[0010] Low-dimensional features are extracted from the separated flight trajectory data, where low-dimensional features of temporal and spatial attributes are fused into joint features;

[0011] A graph structure model is constructed, in which representative trajectories reflecting the characteristics of each flow pattern are used as nodes, and the connection weights of the edges in the graph structure are related to the flight traffic between adjacent nodes. Specifically, using the obtained representative flow patterns, representative flight trajectories in each airspace sector are used as nodes, and a graph model is constructed based on historical trajectory data, in which the edge weights between nodes are determined by the flight traffic and flow direction between adjacent nodes, intuitively reflecting the distribution and direction of traffic flow within the airspace.

[0012] By establishing a parallel fusion prediction model, the air traffic flow prediction results within a predetermined time step are obtained.

[0013] In this embodiment, the preprocessing includes: summarizing the trajectories corresponding to the flight trajectory data into various sectors; wherein, the flight trajectory data within a specified airspace range is extracted, for example, the flight trajectory data of all sectors in the Shanghai airspace for one month is extracted, and the trajectories are summarized into various sectors;

[0014] The flight trajectory data is segmented according to the sampling interval and then normalized. Specifically, the raw trajectory data can be segmented according to the sampling interval (e.g., 10 seconds), and min-max normalization can be applied to each data attribute, mapping each attribute value to the [0,1] interval. For example, cubic spline interpolation can be used to uniformly sample different trajectories, ensuring that each trajectory has a fixed number of sampling points to guarantee consistent data length for the input deep learning framework. Raw flight trajectory data is acquired from the Automatic Dependent Surveillance-Broadcast (ADS-B) system. This data includes timestamps, call signs, ground speed, heading angle, vertical speed, location information (latitude, longitude, and altitude), flight number, and departure and arrival airports. The data is sampled and normalized, and the flight trajectory data is separated into temporal attribute data and spatial attribute data according to its spatiotemporal characteristics. The collected flight trajectory data is preprocessed according to predetermined rules, and cubic spline interpolation is used to uniformly sample each trajectory, ensuring that all trajectories have the same number of sampling points. The trajectory is further separated into temporal and spatial attributes, and min-max normalization is used to map each attribute value to a unified interval.

[0015] In this embodiment, the low-dimensional feature extraction process includes: using a deep autoencoder based on a temporal convolutional network (TCN) and a deep autoencoder based on a convolutional neural network (CNN) to extract low-dimensional features from the separated temporal attribute data and spatial attribute data, respectively, to obtain the low-dimensional features of the temporal attribute and the low-dimensional features of the spatial attribute. In the latent space of the joint feature representation, a clustering algorithm is employed, and a clustering allocation enhancement strategy is introduced to self-optimize the clustering results, thereby achieving accurate classification of different airflow patterns. Specifically, for the temporal attribute data, a deep autoencoder (DAE) based on a temporal convolutional network (TCN) is used to efficiently capture long-term temporal dependence features in the flight trajectory; for the spatial attribute data, a deep autoencoder based on a CNN is used to fully extract the local spatial layout and geometric features of the flight trajectory; the low-dimensional feature representations with temporal and spatial features output by the two autoencoders are denoted as follows: and They are then fused to generate a comprehensive spatiotemporal feature vector. In the latent representation zz output by the encoder, a clustering algorithm is used to cluster the low-dimensional representation of the flight trajectory. At the same time, a clustering assignment enhancement strategy based on the Student t-distribution is introduced. The clustering results are optimized by the KL divergence between the clustering algorithm and the auxiliary target distribution, so as to achieve detailed division of different flow patterns in the airspace. Based on the clustering results, the representative flight trajectory of each cluster is extracted as the key information to characterize each local flow pattern.

[0016] Low-dimensional feature extraction is performed on the separated temporal attribute data, including: modeling the dynamic changes of the trajectory at consecutive moments and constructing a deep autoencoder model based on TCN, followed by training the autoencoder model. Specifically, the dynamic changes of the trajectory at consecutive moments include: ground speed, heading angle, vertical speed, and position information. Dilated convolution and residual connections are used in the TCN structure to fully capture the long-term dependencies in the flight trajectory.

[0017] In this embodiment, low-dimensional feature extraction is performed on the separated spatial attribute data, including: establishing a CNN-based deep autoencoder model for extracting low-dimensional features of the spatial attribute data. The deep autoencoder model is used to extract low-dimensional features of spatial attribute data such as trajectory length, azimuth, flight status, and airport information hidden in the flight trajectory. For non-numerical airport information, embedding technology is used to convert it into a fixed-dimensional vector representation. By minimizing the reconstruction error, the low-dimensional representation output by the spatial autoencoder contains sufficient regional features, which is beneficial for subsequent flow pattern differentiation. A Gaussian mixture model is used to perform soft clustering on the fused joint features; a probability distribution based on the Student-t distribution is used to fit the cluster assignment, where the loss function in the cluster assignment process is optimized using KL divergence to obtain more discriminative clustering results. The cluster assignment enhancement improves the compactness of trajectories of similar flow patterns in the cluster, while simultaneously increasing the separation between different flow patterns.

[0018] In this embodiment, the process of constructing the graph structure model includes: using the representative trajectory of each flow pattern in the clustering results as nodes of the graph, where each node reflects the main flow direction within the corresponding airspace sector; counting the number of flight associations between different nodes based on historical trajectory data; constructing an adjacency matrix based on the statistical results; and obtaining the edge weights based on the flight traffic between nodes. The graph structure model can accurately reflect the spatial distribution and flow direction of each flow pattern within the airspace, facilitating subsequent refined traffic prediction.

[0019] In the parallel fusion prediction model, a graph convolutional network (GCN) is used to extract spatial topological dependencies in the graph structure, and a long short-term memory (LSTM) network is used to capture the temporal dynamic changes of node traffic data in the graph structure. For example, by using a parallel architecture and employing both GCN and LSTM to model the constructed graph model for multimodal features, spatial topological features and temporal dynamic features are extracted respectively. Then, feature fusion is performed, and finally, the parallel fusion prediction model outputs air traffic flow prediction results for multiple predetermined time steps (e.g., 30 minutes to 150 minutes), thereby supporting fine-grained control of air traffic management.

[0020] The parallel fusion prediction model includes: one branch uses a graph convolutional network (GCN) to model the spatial features of the constructed flight trajectory graph structure to extract the topological information between nodes in the airspace; another branch uses a long short-term memory network (LSTM) to model the node traffic data in time series to capture the dynamic trend of air traffic flow; the outputs of the two branches are combined through splicing or other fusion strategies to integrate features, and the traffic prediction results for the future preset time steps are output through a fully connected layer.

[0021] The training process of the parallel fusion prediction model includes: a) Data reconstruction stage: TCN and CNN autoencoders are trained separately to obtain low-dimensional representations in the temporal and spatial domains by minimizing reconstruction errors, and the pre-trained model is optimized using a joint loss function; b) Clustering stage: Flight trajectory data is finely clustered in the fused latent space using a joint clustering assignment enhancement strategy and Gaussian mixture model, and the clustering results are transformed into an analytical representation of air traffic flow using the constructed graph model; c) Traffic flow prediction stage: The constructed graph structure is trained as a whole using parallel GCN and LSTM models, parameters are optimized by defining theoretical loss functions such as mean squared error, and air traffic flow prediction results for multiple future time steps are output, thereby theoretically achieving fine prediction of airspace flow status and supporting air traffic management and route planning decisions. The outputs of the GCN and LSTM branches are fully fused in the feature layer, and the fused features are processed through a set of fully connected layers. Finally, the air traffic flow prediction results for multiple time periods in the future (such as 30 minutes, 60 minutes, 90 minutes, 120 minutes, and 150 minutes) are output, thereby achieving high-precision prediction of fine-grained traffic flow within the sector.

[0022] The present invention provides a refined air traffic flow prediction method based on flow pattern recognition. This method efficiently extracts spatiotemporal features from flight trajectories, accurately identifies local flow patterns, and constructs a graphical model based on representative flow patterns to achieve fine-grained prediction of airspace traffic flow, thereby significantly improving the precision and dynamic response capability of local area control. It is particularly suitable for in-depth analysis of flight trajectory data within airspace sectors, providing accurate auxiliary decision support for air traffic control, route planning, and airspace design. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of flight trajectory data preprocessing provided in an embodiment of the present invention;

[0025] Figure 2 A flow pattern recognition framework diagram provided in an embodiment of the present invention;

[0026] Figure 3 A traffic prediction framework diagram provided for embodiments of the present invention;

[0027] Figure 4This is a schematic diagram of the system framework provided for an embodiment of the present invention. Detailed Implementation

[0028] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Embodiments of the present invention will be described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in the specification of the present invention means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0029] The design concept of this embodiment lies in optimizing and developing a novel theoretical and technical framework. This framework enables deep identification of various flow patterns in flight trajectories at the airspace sector level and directly incorporates this detailed flow pattern information into the traffic flow prediction process. By employing advanced feature extraction methods (such as a deep spatiotemporal fusion model based on dual autoencoders), joint clustering allocation enhancement strategies, and traffic flow prediction methods based on graph structure modeling, the spatiotemporal characteristics of various flow actions within a sector can be captured more accurately. This achieves refined prediction of traffic flow within a local airspace, providing a theoretical basis and technical support for dynamic control and refined airspace management. This solution is primarily used to provide more accurate decision-making basis for functional subsystems of air traffic control systems, such as airspace control, route planning, and flight safety management.

[0030] To better illustrate the practical application of this embodiment, real flight data can be used, such as the trajectory data and sector range data of all aircraft in all adjacent sectors of an eastern municipality's airspace from October 1st to October 31st, 2021. The refined air traffic flow prediction method based on flow pattern recognition provided in this embodiment of the invention specifically includes the following main steps: 1. Data preprocessing; 2. Intra-sector flow pattern recognition; 3. Refined sector traffic flow prediction.

[0031] Step 1: Data preprocessing. The data preprocessing process is as follows: Figure 1 As shown, the specific implementation steps are divided into S1-S6.

[0032] S1. Summarize the ADS-B data of all aircraft in the airspace above a municipality in the east within a certain time period into various sectors, and obtain all the trajectories in each sector.

[0033] S2. Resample the ADS-B data of each trajectory within the sector into information points with uniform time distribution (e.g., one sampling point every 10 seconds).

[0034] S3. All information points (such as latitude, longitude, altitude, heading angle, vertical velocity, and ground velocity) of each trajectory within a sector are horizontally merged into a single row. The trajectory data is represented by features with equal time intervals between each column, and each row represents all features of one trajectory. The trajectory data obtained through this processing method is time series data. .

[0035] S4. Separate features with spatial attributes from ADS-B. Its features include:

[0036] Track length: The number of location points in each original track; Azimuth: Calculate the heading from the starting point to the ending point of the track to obtain the path azimuth; Climb or descent: 1 if the track is climbing within the sector, and -1 if it is descending; Airport information: Departure and arrival airport information for each track.

[0037] S5. The airport information attribute values ​​in the database are strings and need to be converted into numerical data. To avoid dimensionality explosion and sparse matrix phenomena, word embedding technology is used to encode the airport information.

[0038] S6. Separately and The min-max normalization method is used to map all attribute values ​​to the range [0, 1]. The formula is as follows:

[0039]

[0040] in, The maximum value in each attribute column, The minimum value in each attribute column.

[0041] Step 2: Intra-sector Flow Pattern Recognition

[0042] The intra-sector flow pattern recognition framework proposed in this invention consists of two deep autoencoders, one of which, based on a TCN, is used to process the temporal attribute data of the intra-sector trajectory. For reconstruction, another CNN-based deep autoencoder is used to process the spatial attribute data. Refactoring is required; the specific framework is as follows: Figure 2 As shown. During training, the two autoencoders are trained independently. After training, the latent representations output by the two encoders are fused, and a clustering algorithm is further used for clustering. The entire process is divided into three parts: temporal attribute data reconstruction, spatial attribute data reconstruction, and clustering assignment reinforcement.

[0043] (1) Reconstruction of time attribute data

[0044] Given that autoencoders based on fully connected layers reconstruct more time-sensitive... It is difficult to capture the temporal dependencies in flight trajectories, resulting in poor low-dimensional clustering performance. To better capture the temporal changes and dynamic characteristics of flight trajectories, this invention employs a Transient Network (TCN) to construct an autoencoder. It is well known that TCNs can better capture long-term dependencies in time series compared to fully connected layers. TCNs use dilated convolutions to give the network a larger receptive field and can obtain… The long-term historical information is calculated using the following formula:

[0045]

[0046] Where k is the size of the convolution kernel, and d is the dilation coefficient. It's a filter, and T is the transpose operation. This indicates the previous time-frequency information index. Indicates that in sequence x T The input vector corresponding to the nth time step, where i represents the index of the convolution kernel and j represents the index of the time step. Let represent the weight vector of the i-th filter in the convolution kernel. This represents the data at the j-th time step in the input sequence. The output obtained after performing temporal convolution.

[0047] The entire TCN network consists of multiple residual blocks. Residual connections are used for skip connections to mitigate the vanishing gradient problem and improve the model's stability and performance. For trajectory input... and dilated convolution operation The output of the residual block is defined as: Where o represents the output of the trajectory after processing by the residual block. This represents the activation function. Furthermore, to prevent overfitting, an average pooling layer is added after the TCN layer.

[0048] Depend on Figure 1 As can be seen, this invention employs multiple TCN layers to design a symmetrical encoder and decoder, aiming to achieve... The two stages of data mapping to the latent space and reconstruction are represented by the following formula:

[0049]

[0050] in, These are the description matrices for the input trajectory and the reconstructed trajectory, respectively. N is the number of trajectories, and K is the dimension of the trajectory. and These are the parameters for the encoder and decoder, respectively. , These represent the decoder and encoder, respectively, used to process temporal attribute data. During training, the objective function is minimized to achieve... and To achieve the greatest possible similarity, the objective function is defined as follows:

[0051]

[0052] Among them, model parameters , It uses the L2 norm, and the model parameters can be continuously optimized and updated through backpropagation during training, ultimately obtaining a low-dimensional representation from the encoder. In the formula, the parameter before the semicolon is used as the variable of the function, and the parameter after the semicolon is used as the parameter of the autoencoder.

[0053] (2) Spatial attribute data reconstruction

[0054] Spatial attribute data It consists of four parts: trajectory length, azimuth, ascent or descent, and airport information. The airport information attribute values ​​are strings and need to be converted to numerical data. A common method is one-hot encoding; however, there are 117 possible values ​​for airport information attributes, and one-hot encoding can lead to an overly sparse matrix and dimensionality explosion. Therefore, embedding technology is used to encode the airport information. For example: due to... There is no relationship between the data in each column, thus exhibiting high spatial characteristics. Therefore, a CNN is used to build an autoencoder. Reconstruct the data and obtain its low-dimensional feature representation. CNNs have a significant advantage in processing spatial attribute features due to their local connectivity and weight sharing characteristics.

[0055] The autoencoder model parameters for reconstructing spatial attribute data can be obtained using the following two formulas. and objective function .

[0056]

[0057]

[0058] in, This represents the reconstructed spatial attribute data. This represents the decoder parameters used to process spatial attribute data. These represent the encoder parameters used to process spatial attribute data. This represents the decoder used to process spatial attribute data. This indicates an encoder used to process spatial attribute data.

[0059] (3) Clustering assignment reinforcement

[0060] Based on the above, we can conclude that... and Low-dimensional representation of latent space and They are further merged to obtain the fused features. While the fused features can be directly clustered using clustering algorithms such as k-means or DBSCAN, this invention introduces cluster assignment enhancement to further improve clustering performance. Specifically, two distributions, Q and P, are introduced to enhance the clustering ability of z. Q is the student-t distribution, used to calculate the probability that trajectory i is clustered into cluster j. P is the target distribution, used to guide the model to optimize cluster assignment, thereby better fitting the actual distribution of trajectory data and continuously improving the clustering effect during the iteration process. The calculation formula is:

[0061]

[0062]

[0063] in, It is the centroid of j, the initial cluster centroid. The z-axis is estimated using the k-means algorithm. This represents the target probability that i is assigned to j. Let be the embedded representation of the i-th trajectory after passing through the autoencoder. This represents the cluster center number corresponding to the current data point. The clustering loss function can be defined by calculating the distance between P and Q using KL divergence:

[0064]

[0065] By minimizing the clustering loss, trajectories can be assigned to appropriate cluster centers more accurately, improving clustering results. This is a self-training strategy.

[0066] Given that this invention employs two autoencoders to process spatial and temporal features separately, then combines their results for clustering assignment enhancement, and simultaneously optimizes the parameters of each part through a fine-tuning process, the entire traffic pattern recognition model can be viewed as an end-to-end structure. Considering the reconstruction loss of the autoencoder used to process temporal attribute data... Autoencoder reconstruction loss for processing spatial attribute data Clustering loss The objective function of the traffic pattern recognition model is defined as follows:

[0067]

[0068] Where β is the weighting coefficient. After fine-tuning, trajectory i will obtain cluster labels, such as... , where k represents the number of clusters. It is worth noting that this invention can employ various clustering algorithms during the cluster assignment enhancement process, such as K-means, DBSCAN, and GMM.

[0069] Step 3: Refined Sector Flow Prediction

[0070] The refined intra-sector traffic prediction framework proposed in this invention consists of GCN and LSTM. GCN is used to learn the spatial features of the constructed air traffic topology, while LSTM is used to learn the time-series traffic features of each node. The specific framework is as follows: Figure 3 As shown, the entire process is divided into two parts: a refined traffic prediction mapping strategy and a parallel fusion prediction model.

[0071] (1) Traffic Refinement Prediction Mapping Strategy

[0072] As the background technology shows, existing GNN-based air traffic flow prediction methods are mainly limited to the macro-level prediction. However, ATC currently manages flights on a sector-by-sector basis, and previous prediction methods provide ATC with limited knowledge, restricting the diversity and accuracy of ATFM decision-making. Therefore, this section proposes a mapping strategy based on sector traffic pattern recognition, aiming to predict the traffic flow of various patterns within a sector. Based on the sector flow pattern recognition in Section 2, trajectory clustering results within each sector can be obtained. Furthermore, representative trajectories for each cluster are calculated and used as nodes in the graph. Notably, because representative trajectories are directional, not only can the traffic flow of each pattern within a sector be predicted, but the direction of that traffic flow pattern can also be determined, achieving refined sector traffic flow prediction. By providing ATC with richer intra-sector traffic flow prediction information, it helps improve the accuracy and diversity of its decision-making.

[0073] Since the trajectories between adjacent sector nodes are connected, this invention constructs an adjacency graph based on historical trajectories. Specifically, let A and B be two nodes (representing trajectories) in two adjacent sectors, and extract all complete trajectories contained in those adjacent sectors, assuming a total of... Trajectory The i-th trajectory Represented as , and Let A and B represent the start and end points of the complete trajectory within the adjacent sectors, respectively. Now we need to obtain the weight of the connection from A to B, which is calculated as follows:

[0074]

[0075]

[0076] in, Represents the set of all trajectories belonging to cluster A. , Represents the set of all trajectories in the cluster to which B belongs. , express arrive The number of flights, This represents the connection weight from A to B, and K represents the representative trajectory cluster from all adjacent sectors to B. The number of flights, This represents the last trajectory point of each trajectory within the sector.

[0077] (2) Parallel fusion prediction model

[0078] Given that the air traffic flow prediction model using a concatenated GCN and LSTM approach can lead to the weakening of some spatial and temporal features during information transmission, this invention employs a parallel approach to avoid this problem. The input to the GCN includes the flow characteristics of each node and the adjacency matrix A. Let... It is GCN's first The layer's input information (traffic samples), The input information for layer l+1 is calculated as follows:

[0079]

[0080] in, It is an activation function. It is a weight matrix. . It is a degree matrix. , where I represents the identity matrix.

[0081] LSTM is used to fully learn the time dependence of air traffic flow; its computational process can be found in the references, and will not be repeated here. and These are the outputs of GCN and LSTM, respectively, and their features are further fused by concatenation. To further predict traffic, three fully connected layers are used to process the fused features sequentially. By setting the dimension of the last layer to 5, traffic prediction at 5 time steps is achieved, such as [30min, 60min, 90min, 120min, 150min].

[0082] During the training process of this parallel fusion prediction model, the parameters of GCN and LSTM... Both are achieved by minimizing the mean squared error (AMSE) loss function. Updated, the calculation is as follows:

[0083]

[0084] Where Y is the actual traffic value matrix, This is a traffic prediction matrix, where L is the total number of samples and T is the output time step, which is 5 in this case. and They represent the first The true and predicted values ​​of a traffic sample at time step t.

[0085] The refined air traffic flow prediction method based on flow pattern recognition provided in this invention acquires raw flight trajectory data from the Automatic Dependent Surveillance-Broadcast (ADS-B) system and uses advanced data preprocessing techniques to split the acquired trajectory data into temporal attribute data and spatial attribute data with obvious spatiotemporal characteristics. Secondly, a deep autoencoder based on a temporal convolutional network (TCN) is used to effectively model the temporal features in the flight trajectory data, while an autoencoder constructed using a convolutional neural network (CNN) is used to deeply extract spatial features, thereby obtaining a low-dimensional feature representation of the flight trajectory. Thirdly, based on the above low-dimensional features, this invention introduces a joint clustering allocation enhancement strategy, which uses a clustering algorithm to accurately identify various flow patterns within the airspace, especially capturing representative flow patterns with obvious directions. Finally, a graph structure model is constructed based on the identified flow pattern representative trajectories, and a multimodal prediction architecture combining a graph convolutional network (GCN) and a long short-term memory network (LSTM) is used to perform directional fine-grained prediction of each flow pattern, thus theoretically achieving high-precision prediction of traffic flow distribution and its dynamic direction within airspace sectors. This invention proposes a theoretical method based on flow pattern recognition. It achieves accurate flow pattern mapping by extracting spatiotemporal features through dual-channel autoencoders and enhancing the joint clustering allocation. Combined with graph representation technology and parallel network prediction strategy, it provides an efficient and advanced technical path for airspace design optimization, air traffic control, and flight safety management.

[0086] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The above descriptions are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for refined traffic flow prediction driven by air traffic flow pattern recognition, characterized in that, include: Flight trajectory data is acquired from the Automatic Dependent Surveillance-Broadcast (ADS-B) system and preprocessed, wherein data is separated according to the spatiotemporal characteristics of the flight trajectory data during the sub-preprocessing process; Low-dimensional features are extracted from the separated flight trajectory data, where low-dimensional features of temporal and spatial attributes are fused into joint features; Construct a graph structure model, in which representative trajectories reflecting the characteristics of each flow pattern are used as nodes, and the connection weights of the edges in the graph structure are related to the flight traffic between adjacent nodes; By establishing a parallel fusion prediction model, the air traffic flow prediction results within a predetermined time step are obtained. The process of low-dimensional feature extraction includes: using a deep autoencoder based on a temporal convolutional network (TCN) and a deep autoencoder based on a convolutional neural network (CNN) to extract low-dimensional features from the separated temporal attribute data and spatial attribute data, respectively, to obtain the low-dimensional features of the temporal attribute and the low-dimensional features of the spatial attribute. Low-dimensional feature extraction is performed on the separated temporal attribute data, including: modeling the dynamic change features of continuous time points in the track, constructing a deep autoencoder model based on TCN, and then training the autoencoder model, which is used to implement... The two stages of data mapping to the latent space and reconstruction are represented as follows: ； in, These are the description matrices of the input trajectory and the reconstructed trajectory, respectively. It is the number of trajectories. It is the dimension of the trajectory. and These are the parameters for the encoder and decoder, respectively. , These represent the decoder and encoder, respectively, used to process temporal attribute data. During training, the efficiency is improved by minimizing the objective function. and Similarity; The minimization objective function is: ； Among them, model parameters , It is the L2 norm, and the low-dimensional representation obtained from the encoder is: ; Low-dimensional feature extraction is performed on the separated spatial attribute data, including: establishing a CNN-based deep autoencoder model for extracting low-dimensional features from the spatial attribute data. The low-dimensional features of the spatial attribute data include: trajectory length, azimuth, flight status, and airport information hidden in the flight trajectory; for non-numerical airport information, embedding technology is used to transform it into a fixed-dimensional vector representation; the deep autoencoder model is used to extract low-dimensional features from the spatial attribute data. Reconstruct the data and obtain its low-dimensional feature representation. ,include: ； in, This represents the reconstructed spatial attribute data. This represents the decoder parameters used to process spatial attribute data. These represent the encoder parameters used to process spatial attribute data. These are the parameters of the autoencoder model that reconstructs spatial attribute data. This represents the decoder used to process spatial attribute data. This represents the encoder used to process spatial attribute data; the objective function used is: .

2. The method according to claim 1, characterized in that, The preprocessing includes: The flight trajectory data is categorized into various sectors; The flight trajectory data is segmented according to the sampling interval and then normalized.

3. The method according to claim 1, characterized in that, Also includes: Soft clustering is performed on the fused joint features using a Gaussian mixture model; Cluster assignment is fitted using a probability distribution constructed based on the Student-t distribution, where the loss function in the cluster assignment process is optimized using KL divergence.

4. The method according to claim 1, characterized in that, The process of constructing a graph structure model includes: The representative trajectory of each flow pattern in the clustering results is used as a node in the graph, where each node is used to reflect the main flow direction in the corresponding spatial sector. The number of flight associations between different nodes is counted based on historical trajectory data. An adjacency matrix is ​​constructed based on the statistical results, and the weight of the edges is obtained based on the flight traffic between nodes.

5. The method according to claim 1, characterized in that, In the parallel fusion prediction model, a graph convolutional network (GCN) is used to extract the spatial topological dependencies in the graph structure, and a long short-term memory network (LSTM) is used to capture the temporal dynamic changes of node traffic data in the graph structure.

6. The method according to claim 1 or 5, characterized in that, The parallel fusion prediction model includes: One branch uses a graph convolutional network (GCN) to model the spatial features of the constructed flight trajectory graph structure in order to extract the topological information between nodes in the airspace; Another branch uses Long Short-Term Memory (LSTM) networks to perform time-series modeling of node traffic data in order to capture the dynamic trends of air traffic flow. The outputs of the two branches are integrated, and the traffic prediction results for the next preset time step are output through a fully connected layer.

Images