Aiming at the challenges in long-term ship trajectory prediction, including high complexity of navigation behavior, insufficient intention mining, and difficulties in modeling long-term temporal dependencies, a ship trajectory prediction method based on a causal long-sequence transformer-LSTM hybrid model(CLT-TLSTM)is proposed. Using origin-destination voyage trajectory data from loading and unloading operations as the core, trajectory purification is achieved through a three-level data processing mechanism, and feature representation is enhanced by integrating positional triangular features with sine-cosine encoding of heading. A cascaded architecture of“Transformer encoder + dual-layer LSTM”is adopted, in which a causal self-attention mechanism treats historical trajectory sequences as the“cause”and future trajectory sequences as the“effect”, enabling the prediction of 4-hour future trajectories based on historical data in 3 hours. Experimental results show that the proposed method achieves excellent performance. Compared to those of the TrAISformer model, the mean absolute error reduced by 10%, the root mean square error reduced by 6%, and the coefficient of determination reaches 0.996 226, demonstrating significant improvement in long-term prediction accuracy and robustness.
从每个航次单元中提取航次元数据,含起讫港名称、航行时间区间、船舶MMSI码(Maritime Mobile Service Identify)。以此为索引从海量历史AIS动态信息中匹配对应航次的定位数据:MMSI码用于锁定目标船舶,时间区间与起讫港则限定数据的时空范围,三者协同实现数据与航次的关联,为后续处理明确分析边界。
2.1.2 基于运动特征的无效数据筛选
针对原始AIS数据中存在的无效信息及异常值,本文基于船舶运动特征分析构建了一套系统的数据筛选机制。
1)设定航速与位移双重阈值。船舶对地航速(SOG)持续低于4节且位置变化不超过0.1 n mile的轨迹段被判定为非航行状态,予以剔除,避免非航行数据干扰模型对航行规律的学习[23]。
WANGM M, WANGY F, DINGJ, et al. Interaction aware and multi-modal distribution for ship trajectory prediction with spatio-temporal crisscross hybrid network[J]. Reliability Engineering & System Safety, 2024,252:No.110463.
RISTICB, SCALAB L, MORELANDEM, et al. Statistical analysis of motion patterns in AIS Data: anomaly detection and motion prediction[C]∥Proceedings of the 11th International Conference on Information Fusion. Piscataway, USA: IEEE, 2008:1-7.
[10]
PERERAL P, OLIVEIRAP, GUEDES SOARESC. Maritime traffic monitoring based on vessel detection, tracking, state estimation, and trajectory prediction[J]. IEEE Transactions on Intelligent Transportation Systems, 2012,13(3):1188-1200.
LUOW D, ZHANGG J. Ship motion trajectory and prediction based on vector analysis[J]. Journal of Coastal Research, 2020,95():1183-1188.
[13]
TONGX P, CHENX, SANGL Z, et al. Vessel trajectory prediction in curving channel of inland river[C]∥Proceedings of the International Conference on Transportation Information and Safety. Piscataway, USA: IEEE, 2015:706-714.
JURKUSR, VENSKUSJ, TREIGYSP. Application of coordinate systems for vessel trajectory prediction improvement using a recurrent neural networks[J]. Engineering Applications of Artificial Intelligence, 2023,123:No.106448.
CAPOBIANCOS, MILLEFIORIL M, FORTIN, 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.
JIANGD P, SHIG Y, LIN, et al. TRFM-LS: transformer-based deep learning method for vessel trajectory prediction[J]. Journal of Marine Science and Engineering, 2023,11(4):No.880.