引用本文:吴跃高,俞万能,曾广淼,商逸帆,廖卫强.融合拼接注意力机制的船舶轨迹预测方法[J].控制理论与应用,2025,42(9):1798~1806.[点击复制]
WU Yue-gao,YV Wan-neng,ZENG Guang-miao,SHANG Yi-fan,LIAO Wei-qiang.Ship trajectory prediction method incorporating concatenated attention mechanism[J].Control Theory & Applications,2025,42(9):1798~1806.[点击复制]
融合拼接注意力机制的船舶轨迹预测方法
Ship trajectory prediction method incorporating concatenated attention mechanism
摘要点击 3347  全文点击 160  投稿时间：2023-08-16  修订日期：2025-01-09
查看全文  查看/发表评论  下载PDF阅读器   HTML
DOI编号  10.7641/CTA.2024.30557
  2025,42(9):1798-1806
中文关键词  注意力机制  AIS数据  深度学习  轨迹预测  模型预测控制
英文关键词  attention mechanism  AIS data  deep learning  trajectory prediction  model predictive control
基金项目  国家自然科学基金项目(52171308),福建省自然科学基金项目(2022J01333)资助.
作者单位E-mail
吴跃高 集美大学轮机工程学院 wyg729@foxmail.com 
俞万能 集美大学轮机工程学院  
曾广淼 集美大学轮机工程学院  
商逸帆 集美大学轮机工程学院  
廖卫强* 集美大学轮机工程学院 wq_liao@jmu.edu.cn 
中文摘要
      船舶轨迹预测对船舶的航行安全有着重要的价值,为提高对船舶未来轨迹的预测精度,本文将基于拼接注 意力的序列到序列模型(Seq2Seq-CA)与模型预测控制(MPC)相结合,提出了一种新的船舶轨迹预测方法.通过在 Seq2Seq中引入拼接注意力机制,增强了模型对序列特征的理解.为进一步提高模型预测轨迹的准确性与运动合理 性, 采用MPC对预测轨迹概率分布进行纠正,得到最终的输出轨迹.在训练和测试的过程中,为提高对序列数据的 利用率,随机初始化序列的起始位置并使用滑动窗口法对序列进行读取.在自动识别系统(AIS)数据集上测试轨迹 预测方法性能,根据定量分析,Seq2Seq-CA相较于原Seq2Seq提高17.2%的预测准确性. 通过结合MPC进行轨迹纠正 后, Seq2Seq-CA的预测准确性、鲁棒性分别提升39.9%和9.2%. 根据定性分析,本文提出的预测方法在不同船舶运 动模式下均能更准确合理地预测船舶的未来轨迹.
英文摘要
      Ship trajectory prediction plays a crucial role in ensuring the navigation safety of ships. In order to enhance the accuracy of predicting future ship trajectories, this paper introduces a novel ship trajectory prediction method that combines the concatenation attention mechanism in Seq2Seq-CA with model predictive control (MPC). The incorporation of the concatenation attention mechanism within Seq2Seq enhances the model’s understanding of sequence features. To further improve the accuracy and motion coherence of trajectory predictions, MPC is employed to correct the probability distribution of predicted trajectories, yielding the final output trajectories. During the training and testing processes, random initialization of sequence starting positions and the utilization of a sliding window approach are employed to enhance the utilization of sequence data. Performance testing on the AIS dataset demonstrates that Seq2Seq-CA improves prediction accuracy by 17.2%comparedtotheoriginalSeq2Seq. Aftertrajectory correction using MPC, Seq2Seq-CAexhibits a 39.9% increase in prediction accuracy and a 9.2% improvement in robustness. Qualitative analysis confirms that the proposed prediction method accurately and reasonably predicts future ship trajectories under various ship motion patterns.