Multi-aircraft ground taxiing trajectory interactive prediction model construction method

Abstract

The invention provides a method for constructing an interactive prediction model for ground sliding tracks of multiple aircrafts. According to the model, a dynamic interaction diagram is constructed with aircrafts as nodes and spatial proximity relations as edges, historical trajectory and planned path features are extracted by adopting a Transform encoder, and dynamic influences of adjacent aircrafts on target individual trajectory evolution are aggregated through a residual multi-head GAT module. And finally, generating a final prediction track by adopting a gating fusion mechanism and a time sequence decoder, emphatically optimizing the long-term prediction precision through a weighted loss function, and performing simulation on a real airport multi-point positioning system data set. The result shows that the average displacement error and the final displacement error of the model in a multi-aircraft scene are 3.517 m and 4.082 m respectively, and the performance is obviously superior to that of a baseline model such as STGAT. According to the method, the interaction among multiple subjects can be effectively captured, and reliable support is provided for improving the scene situation real-time perception and conflict early warning capability.

Description

Technical Field

[0001] This invention belongs to the field of multi-aircraft taxiing technology, and in particular relates to a method for constructing an interactive prediction model of multi-aircraft ground taxiing trajectories. Background Technology

[0002] With the rapid development of the global civil aviation industry, the operation of large hub airports is becoming increasingly complex, and the widespread adoption of multi-runway and multi-taxiway systems has led to a surge in surface traffic density. As a crucial link connecting parking positions and runways, aircraft ground taxiing directly impacts the overall operational capacity and stability of an airport. Against this backdrop, high-precision aircraft ground taxiing trajectory prediction has become a core technology for enhancing situational awareness, enabling intelligent conflict early warning, and optimizing scheduling decisions. It holds significant theoretical and practical importance for ensuring operational safety, reducing flight delays, and lowering fuel consumption.

[0003] Trajectory prediction technology has evolved from traditional physical models and Kalman filtering to data-driven methods. In recent years, deep learning models, such as Long Short-Term Memory (LSTM) networks and Gated Recurrent Units (GRUs), have achieved significant success in individual aircraft trajectory prediction tasks due to their powerful ability to capture temporal features. Some advanced research has further incorporated prior information, such as taxiway instructions issued by air traffic controllers, into the models, significantly improving the prediction accuracy for single aircraft. However, although these methods are becoming increasingly mature in individual aircraft modeling, they generally treat each aircraft as an isolated system for analysis, which deviates from the actual multi-aircraft operation at busy airports.

[0004] The main limitation of existing research lies in neglecting the complex interactive behaviors of multiple aircraft taxiing concurrently. In busy airport environments, aircraft movement is significantly influenced by the dynamics of surrounding individuals, such as proactive avoidance and deceleration to follow. These interactive behaviors are dynamic, localized, and unstructured, making them difficult to capture by traditional time-series models. Current prediction methods struggle to accurately predict and handle trajectory abrupt changes caused by group interactions, limiting their application in scenarios such as conflict early warning. Summary of the Invention

[0005] In view of this, the present invention aims to overcome the shortcomings of the above-mentioned problems in the prior art and proposes a method for constructing an interactive prediction model of ground taxiing trajectories of multiple aircraft, which shows the spatiotemporal influence relationship between multiple aircraft, thereby achieving more accurate and realistic trajectory prediction.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows:

[0007] In a first aspect, the present invention provides a method for constructing an interactive prediction model of ground taxiing trajectories for multiple aircraft, wherein the prediction model adopts an encoding-interaction-decoding structure;

[0008] During the encoding stage, the Transformer encoder is used to extract features of historical trajectories and planned paths;

[0009] During the interaction phase, a dynamic interaction graph of aircraft is constructed with aircraft as nodes and spatial proximity relationships as edges. This transforms the unstructured situation into structured graph data. Interaction modeling is performed using a graph attention network. The residual multi-head GAT module is used to aggregate and dynamically learn the influence weights of different neighboring aircraft on the trajectory evolution of the target individual, thereby distinguishing and aggregating key interaction information.

[0010] During the decoding stage, a gated fusion mechanism and a time-series decoder are used to generate the final predicted trajectory, and a weighted loss function is used to optimize the long-term prediction accuracy.

[0011] Furthermore, the mathematical formula for the prediction model is expressed as follows:

[0012] set up Let i represent the plane coordinates of aircraft i at time t, and let be the total number of aircraft taxiing on the field. , This represents the historical taxiing trajectory of aircraft i, and similarly, This represents the sequence of predicted trajectory coordinates for aircraft i over the next L seconds. Let r be the coordinate of the planned taxiing path of aircraft i. To represent the sequence of discrete points along the planned taxiing path of aircraft i, a model is established. Predict the ground taxiing trajectories of multiple aircraft.

[0013] Furthermore, the Transformer encoder includes a trajectory encoder and a path encoder, and employs a dual-layer masking strategy to ensure that the trajectory encoder captures the motion state characteristics of the aircraft only from the historical trajectories of the real aircraft.

[0014] Furthermore, the residual multi-head GAT module achieves deep feature learning of graph data structures by combining residual connections and multi-head attention mechanisms. This module first performs multiple rounds of iterative updates on each aircraft node and its neighbors: in each round of iteration, dynamic weights between nodes are calculated in parallel by multiple independent attention heads, where each head uses linear transformation and LeakyReLU activation to generate attention coefficients, and then the attention coefficients are multiplied with the adjacency matrix and masked to consider only nodes with effective connections, and then normalized to obtain the final attention weights; then the neighbor node information is aggregated using these weights, the weighted features of each head are averaged and added to the feature residuals of the previous round, and then Tanh activation is used to enhance the nonlinear expressive power.

[0015] Furthermore, in the decoding stage, the trajectory decoder consists of a gating mechanism and a temporal convolutional network-gated recurrent unit. The gating mechanism dynamically fuses trajectory features and path features, adjusting the representation of trajectory features to incorporate the information of path features.

[0016] Furthermore, in the decoding stage, the loss function is as follows:

[0017] ;

[0018] In the formula, L represents the sequence length (the length of the predicted trajectory sequence), and f represents the coordinate dimension. It is the weight of the predicted point at time t. These are the actual coordinates. It is the predicted coordinates (position parameters). It is a scale parameter that represents uncertainty.

[0019] Furthermore, it also includes using the average displacement error and final displacement error, as well as the minimum average displacement error and minimum final displacement error indices to evaluate the model prediction results.

[0020] Secondly, the present invention provides an electronic device, including a processor and a memory communicatively connected to the processor and used to store executable instructions of the processor, wherein the processor is used to execute the above-described method for constructing an interactive prediction model of a multi-aircraft ground taxiing trajectory.

[0021] Thirdly, the present invention provides a computer-readable storage medium storing a computer program, characterized in that: when the computer program is executed by a processor, it implements the above-mentioned method for constructing an interactive prediction model of ground taxiing trajectories for multiple aircraft.

[0022] Compared with existing technologies, the interactive prediction model construction method for multi-aircraft ground taxiing trajectories described in this invention has the following advantages:

[0023] The model constructed in this invention effectively improves the accuracy of multi-aircraft trajectory prediction and can effectively capture the interaction between multiple entities, providing reliable support for improving the real-time situational awareness and conflict early warning capabilities. Attached Figure Description

[0024] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0025] Figure 1 This is a schematic diagram of the prediction model structure in Embodiment 1 of the present invention;

[0026] Figure 2This is a schematic diagram of the encoder network structure in Embodiment 1 of the present invention;

[0027] Figure 3 This is a schematic diagram showing the dimensional changes of trajectory features and path features in a training batch in Embodiment 1 of the present invention;

[0028] Figure 4 This is a schematic diagram showing the changes in each feature dimension in the gating fusion module of Embodiment 1 of the present invention;

[0029] Figure 5 This is a schematic diagram illustrating the prediction effect of the model in Embodiment 1 of the present invention. Detailed Implementation

[0030] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0031] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0032] Example 1

[0033] This invention proposes an interactive prediction model for aircraft ground taxiing trajectories based on graph attention networks, taking a multi-agent dynamic system perspective. This model demonstrates the spatiotemporal influence relationships among multiple aircraft, thereby achieving more accurate and realistic trajectory predictions. First, a dynamic interactive graph model of aircraft is constructed, abstracting multiple aircraft on the taxiway network at a specific moment as nodes in the graph. Edges are constructed based on their spatial proximity relationships, transforming the unstructured situational scenario into structured graph data. Then, a graph attention network is introduced into this scenario for interactive modeling. An attention mechanism is used to dynamically learn the influence weights of different neighboring aircraft on the trajectory evolution of the target individual, effectively distinguishing and aggregating key interactive information. Finally, the above ideas are integrated into an end-to-end prediction framework that combines feature encoding, graph interactive learning, and temporal decoding, and has been fully validated on a large airport real-world operational dataset.

[0034] The prediction model of this invention adopts an "encoding-interaction-decoding" framework. The overall framework of the model is as follows: Figure 1 As shown below. The specific implementation process of the model is explained in detail below:

[0035] The mathematical formulas used in model building are expressed as follows:

[0036] set up This represents the plane coordinates of aircraft i at time t. Let represent the set of two-dimensional real numbers, and let represent the total number of all taxiing aircraft on the field. , This represents the historical taxiing trajectory of aircraft i. Similarly, This represents the sequence of predicted trajectory coordinates for aircraft i over the next L seconds. Let r be the coordinate of the planned taxiing path of aircraft i. This represents the sequence of discrete points along the planned taxiing path of aircraft i. A model is established. Predict the ground taxiing trajectories of multiple aircraft.

[0037] In the encoding stage, to adapt to the unequal length data sequences input in multi-aircraft scenarios and to provide a high-quality information foundation for the multi-subject feature fusion module built based on GAT, this invention constructs a multi-aircraft trajectory encoder and a path encoder, and designs a two-layer masking strategy (aircraft mask and trajectory mask) to accurately distinguish between real data and padding data, ensuring that the trajectory encoder captures the motion state features of aircraft only from the historical trajectories of real aircraft. The two-layer masking strategy includes: assuming the batch size is B, and the maximum number of subject aircraft contained in different samples of this batch is mn, constructing an aircraft mask M of shape [B, mn]. plane In the mask, True indicates that the location is a real aircraft, while False indicates that the location is filled with fake data. Since the length of the trajectory sequence input for each aircraft varies, a mask of shape [B, mn, l] is constructed. traj-max The trajectory mask M traj , where l traj-max This represents the maximum length of the trajectory sequence in this batch. In the mask, False indicates that the position is a real trajectory coordinate pair, while True indicates that the position is a spurious coordinate pair generated by padding. Finally, the trajectory data of all samples are integrated into a shape [B, mn, l] through a padding operation. traj-max The tensor F of [3] traj Similarly, after extracting taxiway features, the planned taxiway path data of aircraft from different samples can be organized into a shape [B, mn, l] through a padding operation. road-max The tensor F of [ , 2] road , where l road-max This represents the length of the longest position information sequence among all planned taxi paths for all aircraft in the current batch; the corresponding taxi path mask is of shape [B, mn, l]. traj-max The tensor M roadIn the mask, False represents the actual gliding path coordinates, while True is used to identify the filled portion. The mask function of this invention is to ensure that when the model generates future trajectory points in an autoregressive manner, each step of the prediction strictly depends on the known historical trajectory and the generated future points, without seeing the true information of the future. This prevents the model from focusing on the trajectory coordinates of subsequent moments when calculating the current moment, thereby forcing the model to adhere to temporal causality to simulate the process of gradually predicting future trajectories in real-world scenarios.

[0038] The trajectory encoder and path encoder of this invention both employ a Transformer encoder, consisting of N identical encoder layers. Each encoder layer contains two main sub-layers: a multi-head self-attention mechanism and a feedforward neural network. Each sub-layer is equipped with residual connections and layer normalization, which, combined with position encoding, injects positional information into the model, helping to understand the sequence's order information. The complete architecture is as follows: Figure 2 As shown.

[0039] This invention introduces a Transformer encoder to extract the overall features of the path sequence, providing a data foundation for the new feature fusion method below. The processing of the path sequence corresponding to a single aircraft is shown in formula (1-6).

[0040] (1)

[0041] (2)

[0042] (3)

[0043] (4)

[0044] (5)

[0045] (6)

[0046] In the formula, E road It is the sequence of path coordinates for processing the current constraints on aircraft taxiing. The embedding layer function is defined by Norm (layer normalization), Linear (linear layer), ReLU (activation function), and concat() (function to connect two features). Q is also mentioned. road K refers to a query issued regarding the predicted target trajectory (the state at a certain future moment). road The key provided by the historical trajectory environment, V road The value F refers to the historical trajectory environment. cls This represents the sequence of aircraft taxiing coordinates after encoder processing. MSA stands for Multi-Head Attention, P stands for Position Encoding, indicating the current trajectory's position within the entire trajectory, and is the final output. This refers to the global characteristics of the current constrained aircraft taxiing path coordinate sequence.

[0047] In the interaction phase, this invention designs a residual multi-head GAT module to fuse the potential influence between different subject features. Graph Attention Network (GAT) dynamically learns the attention weights between nodes, assigning different importance to different neighbors of each node, realizing adaptive feature aggregation of graph structure data, and solving the limitation of traditional GCN's fixed weight allocation, which cannot flexibly allocate information flow according to the correlation between node pairs. Its core idea is to assign learnable attention coefficients to each node's neighbors, thereby capturing complex local structural patterns. This invention introduces an improved attention mechanism based on GAT, swapping the application order of the weight matrix and attention vector, making the attention function more flexible, as shown in formula (7):

[0048] (7)

[0049] To capture more complex relationships between nodes and address the issue of insufficient expressive power in attention mechanisms, among which... Indicates feature splicing, This reflects the importance of node j to node i. It is a trainable attention vector. It is a trainable projection matrix. LeakyReLU is an improved ReLU activation function that no longer outputs zero when the input is negative, but allows a very small negative value.

[0050] The residual multi-head GAT module achieves deep feature learning of graph data structures by combining residual connections with a multi-head attention mechanism. This module first performs multiple rounds of iterative updates on each aircraft node and its neighbors: in each iteration, multiple independent attention heads compute dynamic weights between nodes in parallel. Each head generates attention coefficients using a linear transformation and LeakyReLU activation. These attention coefficients are then multiplied by the adjacency matrix and masked to consider only nodes with valid connections, followed by normalization to obtain the final attention weights. Subsequently, these weights are used to aggregate neighbor node information, the weighted features of each head are averaged and added to the feature residuals from the previous round, and then Tanh activation is applied to enhance nonlinear expressive power.

[0051] For a single training sample, the real aircraft node The formula for calculating the attention of the residual multi-head graph is shown in (8-9):

[0052] (8)

[0053] (9)

[0054] In the formula, D represents the number of attention heads, and K represents the number of residual multi-head graph attention iterations. This represents the characteristics of aircraft i at the k-th iteration. This represents the set of surrounding aircraft nodes of aircraft node i. This represents the weight matrix of the d-th attention head. These are the parameters used to calculate the attention coefficient for the d-th head. Let represent the normalized attention weight of the d-th head node j relative to i.

[0055] Figure 3 The middle part shows the dimensionality changes of trajectory features and path features in a training batch. and This represents the raw data for all aircraft (including populated dummy aircraft) in a batch. This represents the connectivity between aircraft in each sample of a training batch. After the residual multi-head graph attention network fuses the interaction relationships between other subjects and the target subject, and This refers to the trajectory features and path features after integrating the features of each subject.

[0056] In the decoding phase, the trajectory decoder consists of a gating mechanism and a Temporal Convolutional Network-Gated Recurrent Unit (TCN-GRU). The gating mechanism dynamically fuses trajectory features and path features, adjusting the representation of trajectory features to incorporate information from path features. Compared to static fusion, dynamic gating can more flexibly handle complex relationships between features, enhancing the model's ability to model these relationships. The dimensionality changes of trajectory and path features in the gating mechanism-based fusion module are as follows: Figure 4 As shown.

[0057] Compared to the single-aircraft trajectory prediction model, the TCN-GRU decoder removes the output of multiple trajectories and their probability components, outputting only one prediction result. Figure 4 The input data includes F traj-gat and F road-gat The output is the trajectory feature after fusing path features. The following will be F traj-road Input the TCN-GRU decoder to predict the future trajectory coordinate sequence.

[0058] To suppress the accumulation of error over time steps and optimize the stability of the model's long-term predictions, this invention uses a monotonically increasing weight sequence, where time steps further from the current time have greater weights. Furthermore, to optimize the final displacement error, the weight of the last time step is multiplied by a hyperparameter, significantly amplifying the contribution of the final displacement error to the loss. Detailed definitions are shown in Equation (10).

[0059] (10)

[0060] In the formula, L represents the sequence length (the length of the predicted trajectory sequence), and f represents the coordinate dimension. It is the weight of the predicted point at time t. These are the actual coordinates. It is the predicted coordinates (position parameters). It is a scale parameter that represents uncertainty.

[0061] For trajectory prediction tasks, this invention uses relevant indicators of displacement error to evaluate the prediction results.

[0062] (1) Average displacement error and final displacement error

[0063] In tasks such as trajectory prediction or motion estimation, it is necessary to measure the degree of deviation between the predicted trajectory and the actual trajectory. Commonly used evaluation metrics include Average Displacement Error (ADE) and Final Displacement Error (FDE). ADE reflects the average error level throughout the prediction process and can evaluate the overall accuracy of the predicted trajectory at various time points. FDE focuses on the error at the final moment of the prediction sequence, highlighting the model's ability to predict the final state. Using both together, we can assess the model's ability to address accumulated errors in the medium to long term and its ability to pinpoint the final moment.

[0064] Given the true trajectory and model predicted trajectory Each position is Then ADE is located as the arithmetic mean of the Euclidean distances at each time point, as shown in formula (11-12):

[0065] (11)

[0066] FDE focuses on the accuracy of predicting the location of the trajectory endpoint, measuring the model's ability to predict long-term goals.

[0067] (12)

[0068] (2) Minimum average displacement error and minimum final displacement error

[0069] In the field of multimodal trajectory prediction, prediction models typically generate multiple possible future trajectories to address uncertainties in the real world. Minimum Average Displacement Error (minADE) and minimum Final Displacement Error (minFDE), as specifically designed evaluation metrics, provide a more reasonable measure of the performance of such multimodal prediction methods. minADE and minFDE reflect the model's ability to at least fit the true trajectory among multiple possibilities. As the number of predicted trajectories increases, minADE and minFDE generally tend to decrease, but the computational cost also increases. Therefore, a trade-off must be struck between performance improvement and computational cost to select an appropriate number of outputs.

[0070] Assuming the model outputs The predicted trajectories are as follows: The actual trajectory The ADE predicted for the k-th trajectory is shown in formula (13):

[0071] (13)

[0072] The minimum average displacement error is taken as the minimum value among all predictions, as shown in formula (14):

[0073] (14)

[0074] The FDE of the k-th prediction is the distance between the predicted endpoint and the true endpoint, as shown in formula (15):

[0075] (15)

[0076] The minimum final displacement error is taken as the minimum endpoint error among multiple predictions, as shown in formula (16):

[0077] (16)

[0078] To verify the effectiveness of the proposed model, a case study was conducted based on real-world operational data from a multi-point positioning system at a large hub airport. The experiment aims to systematically evaluate the model's predictive performance in multi-aircraft concurrent taxiing scenarios through comparison with mainstream methods and ablation studies of the model itself.

[0079] The experimental data comes from a multi-point positioning system at a major international airport, recording real-world flight trajectories. To focus on multi-agent interaction, the dataset selects scenarios with 6 to 12 concurrently taxiing aircraft as samples. The dataset contains 80,856 training samples and 16,176 test samples, with a training / test ratio of 5:1. The model predicts the trajectory for the next 8 seconds based on historical trajectories from 10-20 seconds ago.

[0080] We use ADE and FDE, the standard metrics in trajectory prediction, as evaluation indicators. ADE measures the average deviation of the overall trajectory, while FDE focuses on the prediction accuracy of the final position. We use the Spatial-Temporal Graph Attention Network (STGAT) and the What-If Motion Predictor (WIMP) as comparative models in our analysis.

[0081] To demonstrate the role of the residual multi-head GAT module proposed in this invention for taxiing trajectory prediction, the following experiments were conducted. The model was divided into four parts: the trajectory encoder (Traj-Decoder), the path encoder (Road-Decoder), the gated fusion mechanism in the trajectory decoder, and the weighted negative log-likelihood loss (Loss) in the loss function. The prediction results are shown in Table 1.

[0082] Table 1

[0083]

[0084] The absence of either the trajectory encoder or the path encoder significantly degrades model performance. Specifically, in the experimental setting with only the path encoder, ADE and FDE reached 4.892 and 5.874, respectively; while with only the trajectory encoder configured, the two metrics were 4.735 and 5.621, respectively. Compared to the best performance achieved by the complete model (ADE 3.517, FDE 4.082), the performance drop of the two defective configurations exceeded 30%, fully demonstrating that both the trajectory encoder and the path encoder are indispensable for building a complete trajectory prediction model; with both encoders intact, other components exhibit clear auxiliary optimization effects. Removing the gating fusion mechanism caused ADE and FDE to increase to 3.641 and 4.155, respectively, indicating that this mechanism can effectively promote the fusion of dual-path features; while removing the weighted loss function mainly caused the FDE metric to increase to 4.233, indicating that this loss function has a positive effect on improving long-term prediction accuracy.

[0085] This invention selected the STGAT and WIMP models for performance comparison on the same test set, and the results are shown in Table 2.

[0086] Table 2

[0087]

[0088] As shown in Table 2, the model proposed in this invention significantly outperforms all comparable models in both ADE and FDE metrics. Compared to STGAT and WIMP, the proposed model's advantage lies in its encoder employing a more powerful Transformer structure, and its residual multi-head GAT module and gating fusion mechanism are better suited to the motion characteristics and constraints of aircraft taxiing. Two typical multi-aircraft concurrent taxiing scenarios are randomly selected for visualization, and the model's prediction performance is shown below. Figure 5 As shown, the left figure is a schematic diagram of the results of test sample 1, and the right figure is a schematic diagram of the results of test sample 2.

[0089] Depend on Figure 5 It can be seen that regardless of whether the aircraft is in a straight-line following state or performing complex avoidance interactions near intersections, the model of this invention (red trajectory) can highly match the actual future trajectory (blue trajectory). Especially in Figure 5 In Test Sample 2, two closely spaced aircraft taxiing in the same direction in the lower right corner were accurately predicted by the model, demonstrating their cooperative motion while maintaining a safe distance. This intuitively proves that the model can not only make accurate individual predictions but also effectively capture and reflect the dynamic interactions between multiple entities. Furthermore, the model's average inference time on a single sample is only 30.42 ms, fully meeting real-time requirements.

[0090] The model proposed in this invention outperforms existing mainstream methods in all aspects of prediction performance. On a real airport surface dataset, compared with the classic STGAT model, the ADE of the proposed model is reduced from 6.288m to 3.517m, a reduction of 43.27%; the FDE is reduced from 11.406m to 4.082m, a reduction of 64.12%. This result shows that the Transformer encoder and the interactive framework designed for aircraft motion characteristics used in the model of this invention can more effectively learn the complex patterns and constraints of aircraft taxiing compared to the general spatiotemporal graph model.

[0091] The model proposed in this invention combines high accuracy and real-time performance, demonstrating potential for engineering applications. Experimental results show that when predicting trajectories 8 seconds in the future, the average inference time on the test set is only 30.42 milliseconds, meeting the requirements for real-time HASP monitoring of surface conditions. Visualization analysis further confirms that the model can accurately reproduce the cooperative motion process of aircraft in typical scenarios such as intersections and following aircraft, providing a solid data and algorithmic foundation for developing a new generation of intelligent conflict early warning and collaborative decision-making systems.

[0092] Example 2

[0093] An electronic device includes a processor and a memory communicatively connected to the processor and used to store processor-executable instructions, the processor being used to execute the above-described method for constructing an interactive prediction model of a multi-aircraft ground taxiing trajectory.

[0094] Example 3

[0095] A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, is the aforementioned method for constructing an interactive prediction model of ground taxiing trajectories for multiple aircraft.

[0096] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in this disclosure all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0097] It should be noted that personal information collected from users should be used for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. Furthermore, such collection / sharing should only be conducted after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign an agreement / authorization that includes authorization of relevant user information before the user uses the function. In addition, any necessary steps must be taken to protect and safeguard access to such personal information data and ensure that others with access to personal information data comply with their privacy and procedures.

[0098] This disclosure is intended to provide implementation schemes for users to selectively prevent the use or access to their personal information data. Specifically, this disclosure is intended to provide hardware and / or software to prevent or block access to such personal information data. Once personal information data is no longer needed, risks can be minimized by restricting data collection and deleting data. Furthermore, where applicable, such personal information is de-identified to protect user privacy.

[0099] The acquisition, transmission, storage, use, and processing of data in this disclosed technical solution all comply with the relevant provisions of national laws and regulations.

[0100] It should be noted that in the embodiments disclosed herein, certain software, components, models, and other existing solutions in the industry may be mentioned. These should be considered as exemplary and are intended only to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used such solutions.

[0101] In the foregoing descriptions of the embodiments, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0102] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0103] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of this disclosure includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of this disclosure pertain.

[0104] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0105] It should be understood that various parts of this disclosure can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0106] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0107] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0108] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present disclosure have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present disclosure. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present disclosure.

Claims

1. A method for constructing an interactive prediction model of ground taxiing trajectories for multiple aircraft, characterized in that: The prediction model employs an encoder-interaction-decoder structure; During the encoding stage, the Transformer encoder is used to extract features of historical trajectories and planned paths; During the interaction phase, a dynamic interaction graph of aircraft is constructed with aircraft as nodes and spatial proximity relationships as edges. This transforms the unstructured situation into structured graph data. Interaction modeling is performed using a graph attention network. The residual multi-head GAT module is used to aggregate and dynamically learn the influence weights of different neighboring aircraft on the trajectory evolution of the target individual, thereby distinguishing and aggregating key interaction information. During the decoding stage, a gated fusion mechanism and a time-series decoder are used to generate the final predicted trajectory, and a weighted loss function is used to optimize the long-term prediction accuracy.

2. The method for constructing an interactive prediction model of ground taxiing trajectories for multiple aircraft according to claim 1, characterized in that: The mathematical formula for the prediction model is expressed as follows: set up Let i represent the plane coordinates of aircraft i at time t, and let be the total number of aircraft taxiing on the field. , This represents the historical taxiing trajectory of aircraft i. This represents the sequence of predicted trajectory coordinates for aircraft i over the next L seconds. Let r be the coordinate of the planned taxiing path of aircraft i. To represent the sequence of discrete points along the planned taxiing path of aircraft i, a model is established. Predict the ground taxiing trajectories of multiple aircraft.

3. The method for constructing an interactive prediction model of ground taxiing trajectories for multiple aircraft according to claim 1, characterized in that: The Transformer encoder includes a trajectory encoder and a path encoder. It employs a dual-layer masking strategy to ensure that the trajectory encoder captures the motion state characteristics of the aircraft only from the historical trajectories of the real aircraft.

4. The method for constructing an interactive prediction model of ground taxiing trajectories for multiple aircraft according to claim 1, characterized in that: The residual multi-head GAT module combines residual connectivity with multi-head attention mechanism to achieve deep feature learning of graph data structure. The module first performs multiple rounds of iterative updates on each aircraft node and its neighbors: in each round of iteration, the dynamic weights between nodes are calculated in parallel by multiple independent attention heads, where each head uses linear transformation and LeakyReLU activation to generate attention coefficients, and then the attention coefficients are multiplied with the adjacency matrix to perform a masking operation to consider only nodes with effective connections, and then normalized to obtain the final attention weights; Then, these weights are used to aggregate neighbor node information, the weighted features of each head are averaged and added to the feature residuals of the previous round, and then Tanh activation is used to enhance the nonlinear expressive power.

5. The method for constructing an interactive prediction model of ground taxiing trajectories for multiple aircraft according to claim 1, characterized in that: In the decoding stage, the trajectory decoder consists of a gating mechanism and a temporal convolutional network-gated recurrent unit. The gating mechanism dynamically fuses trajectory features and path features, adjusting the representation of trajectory features to incorporate information from path features.

6. The method for constructing an interactive prediction model of ground taxiing trajectories for multiple aircraft according to claim 1, characterized in that: During the decoding phase, the loss function is as follows: ； In the formula, L represents the sequence length, and f represents the coordinate dimension. It is the weight of the predicted point at time t. These are the actual coordinates. It is a predicted coordinate. It is a scale parameter that represents uncertainty.

7. The method for constructing an interactive prediction model of ground taxiing trajectories for multiple aircraft according to claim 1, characterized in that: It also includes using the average displacement error and final displacement error, and the minimum average displacement error and minimum final displacement error as indices to evaluate the model prediction results.

8. An electronic device comprising a processor and a memory communicatively connected to the processor and used for storing processor-executable instructions, characterized in that: The processor is used to execute the method described in any one of claims 1-7.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.

Images