Article Details

Abstract

Existing trajectory processing methods typically construct task-specific models, which struggle to capture the underlying correlations and intrinsic patterns across different trajectory tasks, thereby limiting the generalization capability of trajectory intelligence. To address this issue, this paper investigates a Transformer-based multi-task modeling approach for pure trajectory modalities and proposes a unified multi-task model, termed MultiTaskModel. In the encoding stage, the model introduces a control domain, where control tokens are concatenated with the embedded representations of the trajectory data domain to form a unified feature vector for each trajectory point. In the decoding stage, control tokens are employed to identify different tasks and interpret the corresponding model outputs. Extensive experiments are conducted on a publicly available vessel navigation dataset. Compared with conventional single-task baselines, the proposed method achieves a 12.20% improvement in prediction error for a 3-hour forecasting horizon. Furthermore, it yields a 6.81% enhancement in downstream trajectory prediction performance and attains a matching accuracy of 94.15% for the downstream trajectory matching task. Experimental results demonstrate that the proposed approach enables unified modeling of diverse downstream tasks, including trajectory matching, while effectively improving trajectory prediction performance.

Keywords

trajectory foundation model; multi-task learning; unified modeling; time series prediction

References

1. [1] Forti N，Millefiori L M，Braca P，et al. Prediction of Vessel Trajectories From AIS Data via Sequence-To-Sequence Recurrent Neural Networks [C]//ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing，2020：8936-8940.
2. [2] Nguyen D，Vadaine R，Hajduch G，et al. GeoTrackNet—A Maritime Anomaly Detector Using Probabilistic Neural Network Representation of AIS Tracks and A Contrario Detection [J].IEEE Transactions on Intelligent Transportation Systems，2021，23（6）：5655-5667.
3. [3] Capobianco S，Millefiori L M，Forti N，et al. Deep Learning Methods for Vessel Trajectory Prediction Based on Recurrent Neural Networks [J].IEEE Transactions on Aerospace and Electronic Systems，2021，57（6）：4329-4346.
4. [4] Murray B，Perera L P. An AIS-based deep learning framework for regional ship behavior prediction [J/OL].Reliability Engineering and System Safety，2021，215：107819.https://doi.org/10.1016/j.ress.2021.107819.
5. [5]Li M Z，Li B，Qi Z G，et al. Enhancing Maritime Navigational Safety: Ship Trajectory Prediction Using ACoAtt-LSTM and AIS Data [J/OL].ISPRS International Journal of Geo-Information，2024，13（3）：85.https://doi.org/10.3390/ijgi13030085.
6. [6] Huang F L，Liu Z R，Li X H，et al. Vessel Trajectory Prediction Based on AIS Data: Dual-Path Spatial–Temporal Attention Network with Multi-Attribute Information [J/OL].Journal of Marine Science and Engineering，2024，12（11）：2031. https://doi.org/10.3390/jmse12112031.
7. [7] Wang Yihe, Zhang Xiufeng, Xiao Fangbing A ship trajectory interpolation prediction method based on AIS data and STGCN [J]. Water Transport Management, 2025, 47 (3): 29-32
8. [8] Gehring J，Auli M，Grangier D，et al.Convolutional Sequence to Sequence Learning [C]// Proceedings of the 34th International Conference on Machine Learning，2017：1243-1252.
9. [9] Wu X Y，Chen J F，Xiong C Q，et al. Vessel Trajectory Prediction Method Based on the Time Series Data Fusion Model [J].Promet-Traffic & Transportation，2024，36（6）：1160-1175.
10. [10] Spadon G，Kumar J，Eden D，et al. Multi-path Long-term Vessel Trajectories Forecasting with Probabilistic Feature Fusion for Problem Shifting [J/OL].Ocean Engineering，2024，312：119138.https://doi.org/10.1016/j.oceaneng.2024.119138.
11. [11] Nguyen D，Fablet R. A Transformer Network With Sparse Augmented Data Representation and Cross Entropy Loss for AIS-Based Vessel Trajectory Prediction [J].IEEE Access，2024，12：21596-21609.
12. [12] Gao Long, Wu Junfeng, Yang Zhutian, etc Long term prediction of TCNformer ship trajectory based on spatiotemporal feature fusion [J]. Journal of Naval Aviation University, 2024, 39 (4): 437-4444491
13. [13] Qiang H M，Guo Z Y，Chu Z，et al. Motion-inspired Spatial-temporal Transformer for Accurate Vessel Trajectory Prediction [J/OL].Engineering Applications of Artificial Intelligence，2025，148：110391.https://doi.org/10.1016/j.engappai.2025.110391.
14. [14] Wang W T，Xiong W，Ouyang X，et al. TPTrans: Vessel Trajectory Prediction Model Based on Transformer Using AIS Data [J/OL].ISPRS International Journal of Geo-Information，2024，13（11）：400.https://doi.org/10.3390/ijgi13110400.
15. [15] Gong J C，Li H H，Jiao H，et al. Uncertainty-aware ship Trajectory Prediction via Spatio-Temporal Graph Transformer[J/OL].Transportation Research Part E: Logistics and Transportation Review，2025，203：104315.https://doi.org/10.1016/j.tre.2025.104315.
16. [16] Radford A，Kim J W，Xu T，et al. Robust Speech Recognition via Large-scale Weak Supervision [C]//ICML'23: Proceedings of the 40th International Conference on Machine Learning，2023：28492-28518.

About the authors

Zhang Songcheng (1988-), male, Han ethnicity, from Huanggang, Hubei, is an algorithm engineer and a master's student. His research interests include natural language processing and the application of large-scale models in the industry; Corresponding author: Wu Fei (1987-), male, Han ethnicity, born in Jiujiang, Jiangxi Province, senior engineer, vice dean, master's student, research direction: natural language processing, industry applications of large models.