针对目前船舶轨迹预测模型主要考虑轨迹的位置特征，缺乏对船舶操纵性能和环境因素的探究，导致预测精度不高的问题，本文提出一种基于用户画像和改进Seq2Seq的船舶轨迹预测新模型。首先该模型基于用户画像理论，采用频数统计和聚类方法创建了船舶用户画像；其次，创设多编码器实现对船舶用户画像、船舶轨迹和船舶标签等特征的分别编码；最后，基于Seq2Seq架构，增设画像特征检索模块和LSTM融合层，实现画像特征的快速提取和融合。实验结果表明，相较于LSTM、BiLSTM、A+LSTM、A+BiLSTM和Seq2Seq五种模型，新模型在长时间预测方面具备较高的精度和较好的稳定性。
The current ship trajectory prediction models mainly consider trajectory position features and lack exploration of ship maneuverability and environmental factors, leading to low prediction accuracy, this paper proposes a new ship trajectory prediction model based on user profiles and an improved Seq2Seq model. First, based on the theory of user profiles, the proposed model uses frequency statistics and clustering methods to create ship user profiles. Next, multiple encoders are created to encode features such as ship user profiles, ship trajectories, and ship tags separately. Finally, based on the Seq2Seq architecture, a feature retrieval module and LSTM fusion layer are added to achieve rapid extraction and fusion of profile features. Experimental results show that, compared to models such as LSTM, BiLSTM, A+LSTM, A+BiLSTM, and Seq2Seq, the new model has higher accuracy and stability in long-term predictions.
2025,47(13): 149-156 收稿日期：2024-7-27
DOI：10.3404/j.issn.1672-7649.2025.13.026
分类号：U612
基金项目：国家自然科学基金资助项目（52261160383）；浙江省自然科学基金资助项目（LY21E090005）；舟山市科技计划项目（2021C21010）
作者简介：杨欣奕(2000-),男,硕士研究生,研究方向为交通组织与资源优化配置。
参考文献：
[1] 马恺. 基于受限水域的无人船路径规划[D]. 哈尔滨: 哈尔滨工程大学, 2024.
[2] 姜佰辰, 关键, 周伟, 等. 基于多项式卡尔曼滤波的船舶轨迹预测算法[J]. 信号处理, 2019, 35(5): 741-746.
[3] 谢鸿伟, 张英俊, 邢胜伟, 等. 基于模型预测控制的船舶自主避碰方法[J]. 船舶工程, 2021, 43(8): 23-30.
[4] ZHANG X, LIU G, HU C, et al. Wavelet analysis based hidden markov model for large ship trajectory prediction [C]//proceedings of the the 38th chinese control conference, 2019.
[5] 任宇翔, 赵建森, 刘卫, 等. 基于AIS数据和LSTM网络的船舶航行动态预测[J]. 上海海事大学学报, 2019, 40(3): 32-37.
[6] 甄荣, 金永兴, 胡勤友, 等. 基于AIS信息和BP神经网络的船舶航行行为预测[J]. 中国航海, 2017, 40(2): 6-10.
[7] SUO Y, CHEN W, CLARAMUNT C, et al. A ship trajectory prediction framework based on a recurrent neural network [J]. Sensors (Basel), 2022, 20(18): 51-54.
[8] ZHANG L, ZHU Y, SU J, et al. A hybrid prediction model based on knn-lstm for vessel trajectory [J]. Mathematics, 2022, 10(23): 44-47.
[9] 张扬, 彭鹏菲, 卢锐. 基于Seq2Seq的船舶轨迹变步长预测方法研究[J]. 电光与控制, 2023, 30(4): 74-77.
[10] ZHANG Y, HAN Z, ZHOU X, et al. METO-S2S: A S2S based vessel trajectory prediction method with Multiple-semantic Encoder and Type-Oriented Decoder [J]. Ocean Engineering, 2023, 6(1): 1-12.
[11] NGUYEN D D, LE VAN C, ALI M I. Vessel Trajectory Prediction using Sequence-to-Sequence Models over Spatial Grid[C]//Proceeding of the 12th ACM Intemational Conference on Distributed Event, Association for Computing Machinery, 2018.
[12] YOU L, XIAO S, PENG Q, et al. ST-Seq2Seq: A spatio-temporal feature-optimized Seq2Seq model for short-term vessel trajectory prediction[J]. IEEE Access, 2020, 8: 218565-218574.
[13] CHEN P, YANG F, MOU J, et al. Regional ship behavior and trajectory prediction for maritime traffic management: A social generative adversarial network approach [J]. Ocean Engineering, 2024, 7(10): 54-59.
[14] 王森杰, 何正伟. 基于生成对抗网络的船舶航迹预测模型[J]. 中国航海, 2021, 44(2): 72-77.
[15] SUTSKEVER I, VINYALS O, LE Q V. Sequence to sequence learning with neural networks [J]. Advances in Neural Information Processing Systems, 2014, 27(9): 34-42.
[16] HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9(8): 36-41.
[17] 王聪. 基于船舶操纵性的避碰决策研究[D]. 大连: 大连海事大学, 2024.
[18] 万洪亮, 潘家财, 甄荣, 等. 基于CNN-GRU的船舶轨迹预测[J]. 广州航海学院学报, 2022, 30(2): 12-18.
