Abstract:Accurate multi-step flight trajectory prediction plays an important role in Air Traffic Control, which can ensure the safety of air transportation. Two main issues limit the flight trajectory prediction performance of existing works. The first issue is the negative impact on prediction accuracy caused by the significant differences in data range. The second issue is that real-world flight trajectories involve underlying temporal dependencies, and existing methods fail to reveal the hidden complex temporal variations and only extract features from one single time scale. To address the above issues, we propose FlightPatchNet, a multi-scale patch network with differential coding for flight trajectory prediction. Specifically, FlightPatchNet first utilizes the differential coding to encode the original values of longitude and latitude into first-order differences and generates embeddings for all variables at each time step. Then, a global temporal attention is introduced to explore the dependencies between different time steps. To fully explore the diverse temporal patterns in flight trajectories, a multi-scale patch network is delicately designed to serve as the backbone. The multi-scale patch network exploits stacked patch mixer blocks to capture inter- and intra-patch dependencies under different time scales, and further integrates multi-scale temporal features across different scales and variables. Finally, FlightPatchNet ensembles multiple predictors to make direct multi-step prediction. Extensive experiments on ADS-B datasets demonstrate that our model outperforms the competitive baselines. Code is available at: <a class="link-external link-https" href="https://github.com/FlightTrajectoryResearch/FlightPatchNet" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

This paper attempts to solve two main problems in flight trajectory prediction: 1. **The impact of data range differences on prediction accuracy**: In flight trajectory prediction, longitude and latitude are usually represented in degrees, while altitude is represented in meters. Since 1 degree is approximately equal to 111 kilometers, there are significant differences between the data ranges of longitude and latitude and the data range of altitude. This difference may cause the prediction error to become unacceptable within the original data range. For example, after using the Min - Max normalization algorithm, the absolute prediction error of longitude is \(10^{-4}\), but after converting back to the original data range, the actual prediction error is approximately 0.036° (about 4,000 meters), which is unacceptable for the flight trajectory prediction task. 2. **Temporal dependence in real - world flight trajectories**: Existing methods, when extracting temporal features of flight trajectories, usually consider only a single time scale and are unable to reveal the hidden complex temporal changes. For example, Figure 1 shows that the original longitude and latitude sequences are too smooth, masking rich short - term temporal changes, which can be clearly observed in the difference sequences. In addition, the temporal change patterns of longitude and latitude are significantly different from those of altitude. Altitude has an obvious global trend but intense local fluctuations. A single - scale model has difficulty capturing local details and global trends simultaneously, which requires strong multi - scale temporal modeling capabilities. To address these problems, the authors propose **FlightPatchNet**, a multi - scale patch network combined with differential encoding techniques for flight trajectory prediction. Specifically: - **Differential encoding**: Convert the original values of longitude and latitude into first - order differences and keep the original values of other variables as input. In this way, the difference in data range is effectively reduced, and the potential temporal changes in the flight trajectory are revealed. - **Global temporal attention**: Introduce global temporal embeddings to explore the correlations between different time steps. - **Multi - scale patch network**: Design a multi - scale patch network to capture temporal features at different time scales through stacked patch - mixing blocks and integrate multi - scale temporal features. - **Multi - scale aggregator**: Further integrate multi - scale temporal patterns to capture the correlations between scales and the relationships between variables. - **Direct multi - step prediction**: Perform direct multi - step prediction through multiple predictors, using complementary multi - scale temporal features to improve the generalization ability of the model. Experimental results show that FlightPatchNet exhibits significant advantages under various prediction lengths, especially in the prediction of longitude and latitude. For the prediction of altitude, although it does not reach the best, it still provides more stable and robust results.

FlightPatchNet: Multi-Scale Patch Network with Differential Coding for Flight Trajectory Prediction

4-D Flight Trajectory Prediction With Constrained LSTM Network

An Aircraft Trajectory Prediction Method Based on Trajectory Clustering and a Spatiotemporal Feature Network

Global and Local Inter-Attribute Relationships Based Graph Convolutional Network for Flight Trajectory Prediction

Spatiotemporal Propagation Learning for Network-Wide Flight Delay Prediction

Air Traffic Flow Prediction in Aviation Networks Using a Multi-Dimensional Spatiotemporal Framework

PMM-Net: Single-stage Multi-agent Trajectory Prediction with Patching-based Embedding and Explicit Modal Modulation

Flight Trajectory Prediction Using an Enhanced CNN-LSTM Network

Flight Trajectory Prediction Based on Hybrid- Recurrent Networks

Multi-Aircraft Trajectory Collaborative Prediction Based on Social Long Short-Term Memory Network

CiPN-TP: a channel-independent pretrained network via tokenized patching for trajectory prediction

A Non-autoregressive Multi-Horizon Flight Trajectory Prediction Framework with Gray Code Representation

A Novel Trajectory Prediction Method Based on CNN, BiLSTM, and Multi-Head Attention Mechanism

FlightBERT: Binary Encoding Representation for Flight Trajectory Prediction

LSTM-based Flight Trajectory Prediction

A Flight Trajectory Prediction Method based on Trajectory Clustering

Flight Time Prediction for Fuel Loading Decisions with a Deep Learning Approach

A spatial–temporal model for network-wide flight delay prediction based on federated learning

WTTFNet: A Weather-Time-Trajectory Fusion Network for Pedestrian Trajectory Prediction in Urban Complex

Predicting Flight Delay with Spatio-Temporal Trajectory Convolutional Network and Airport Situational Awareness Map

Phased Flight Trajectory Prediction with Deep Learning