| 106 | 0 | 117 |
| 下载次数 | 被引频次 | 阅读次数 |
为保障航行安全、避免碰撞事故，针对当前船舶轨迹预测模型忽略变量相关性、时间依赖性与变量相关性匹配不当、未充分考虑海洋复杂环境问题导致的预测效果不佳，提出了一种基于选择性状态空间的船舶轨迹预测模型。首先，设计基于BallTree空间索引和Haversine距离的时空匹配机制，将公共开源获取的海风、海浪和海流多源水文气象特征与轨迹数据相融合，构建领域特色数据集。其次，以变量为嵌入单位保留各航行参数的独立特性，将增强的特征序列转换为高低分辨率的多维特征表达。最后，利用三个并行选择性状态空间Mamba模块提取航行参数间的多元相关性，并通过前馈网络独立学习每个航行参数序列的时间关联实现轨迹的精准预测。试验结果表明，与基线模型相比，所提模型在地理空间上的预测精度更高，稳定性更强。研究结果可为智能航保建设提供技术支撑。
Abstract：To address the limitations of current ship trajectory prediction models—including the neglect of variable correlations， mismatches between temporal dependencies and variable relationships， and inadequate consideration of complex marine environments leading to poor predictive performance—this study proposes a ship trajectory prediction model based on selective state space named MDMamba (Multi-Dimension Mamba). First， a spatiotemporal matching mechanism utilizing BallTree spatial indexing and Haversine distance is designed to integrate multi-source hydro-meteorological features， such as sea wind， waves， and currents obtained from publicly available sources， with trajectory data， thereby constructing a domain-specific dataset. Second， variables are treated as embedding units to preserve the independent characteristics of each navigation parameter， allowing the enhanced feature sequence to be transformed into multi-dimensional feature representations at both high and low resolutions. The low-resolution forward Mamba performs forward modeling along the variable dimension in the low-dimensional feature space， extracting abstract correlations and sequential information among navigation parameters while learning long-term trends and macroscopic patterns between variables. In contrast， the low-resolution reverse Mamba conducts reverse modeling along the variable dimension， capturing reverse dependencies among navigation parameters and compensating for the unidirectional limitations of the Mamba model. The high-resolution Mamba focuses on capturing fine-grained correlations among navigation parameters in the high-dimensional feature space， learning short-term dynamics and local dependencies between variables. Finally， a feedforward network independently learns the temporal associations of each navigation parameter sequence to achieve precise trajectory predictions. Using ship trajectory data from New York Harbor as the experimental subject， comparative experiments are conducted against Bi–LSTM (Bidirectional Long Short-Term Memory) networks， Bi–GRU (Bidirectional Gated Recurrent Unit)， the Mamba model， and the Transformer model. The results show that the MDMamba model reduces the Mean Squared Error (MSE) by 73.0%， 69.7%， 91.1%， and 46.1%， respectively， reflecting the smallest deviation between predicted and actual values. It also reduces the Root Mean Squared Error (RMSE) by 48.0%， 44.9%， 70.2%， and 26.7%， demonstrating greater prediction stability. Additionally， compared to the Transformer model， it reduces the Mean Absolute Error (MAE) by 19.5% and the average Haversine distance error by 144 meters， achieving higher predictive accuracy in geospatial terms.
[1] 范中洲， 郭婷婷， 郑力铭. 基于改进集对分析法的船舶碰撞风险评价[J]. 安全与环境学报， 2021， 21(2): 470–474.
[2] 田延飞， 何琪， 滑林， 等. 海上船舶碰撞事故关联规则挖掘及致因分析[J/OL]. 安全与环境学报: 1–8[2025–10–27]. https://doi.org/10.13637/j.issn.1009-6094.2024.2230.
[3] 邵波， 刘巧， 柯善钢， 等. 融合LDA–BN的船舶碰撞事故致因分析[J]. 安全与环境学报， 2025， 25(1): 157–164.
[4] Scheepens R， Van De Wetering H， Van Wijk J J. Contour based visualization of vessel movement predictions[J]. International Journal of Geographical Information Science， 2014， 28(5): 891–909.
[5] Alizadeh D， Alesheikh A A， Sharif M. Prediction of vessels locations and maritime traffic using similarity measurement of trajectory[J]. Annals of GIS， 2020， 27(2): 151–162.
[6] Liu Jiao， Shi Guoyou， Zhu Kaige. Vessel trajectory prediction model based on AIS sensor data and adaptive chaos differential evolution support vector regression[J]. Applied Sciences， 2019， 9(15): 2983.
[7] 姜立超， 尚晓兵， 金豹， 等. 基于遗传算法–v支持向量回归的船舶轨迹预测[J]. 哈尔滨工程大学学报， 2024， 45(10): 2001–2006.
[8] Chen Gang， Wang Wei， Xue Yifan. Identification of ship dynamics model based on sparse Gaussian process regression with similarity[J]. Symmetry， 2021， 13(10): 1956.
[9] Rong H， Teixeira A P， Soares C G. Ship trajectory uncertainty prediction based on a Gaussian process model[J]. Ocean Engineering， 2019， 182: 499–511.
[10] Volkova T A， Balykina Y E， Bespalov A. Predicting ship trajectory based on neural networks using AIS data[J]. Journal of Marine Science and Engineering， 2021， 9(3): 254.
[11] Borkowski P. The ship movement trajectory prediction algorithm using navigational data fusion[J]. Sensors， 2017， 17(6): 1432.
[12] Valsamis A， Tserpes K， Zissis D， et al. Employing traditional machine learning algorithms for big data streams analysis: the case of object trajectory prediction[J]. Journal of Systems and Software， 2017， 127: 249–257.
[13] 王雨晨， 邱士远， 李培培， 等. 基于核极限学习机的多标签数据流半监督在线分类方法[J]. 模式识别与人工智能， 2024， 37(8): 741–754.
[14] Liu Xin， Yuan Haiwen， Xiao Changshi， et al. Hybrid-driven vessel trajectory prediction based on uncertainty fusion[J]. Ocean Engineering， 2022， 248(1): 110836.
[15] Yang Chenghong， Wu C H， Shao Jenchung， et al. AIS-based intelligent vessel trajectory prediction using Bi–LSTM[J]. IEEE Access， 2022， 10: 24302–24315.
[16] 马全党， 张丁泽， 王群朋， 等. 双向门控循环单元在船舶轨迹预测中的应用[J]. 安全与环境学报， 2024， 24(1): 83–91.
[17] Zhang Ling， Zhang Jingzhi， Niu Jiong， et al. Track prediction for HF radar vessels submerged in strong clutter based on MSCNN fusion with GRU–AM and AR model[J]. Remote Sensing， 2021， 13: 2164.
[18] 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.
[19] 石悦， 罗贺， 蒋儒浩， 等. 融合数据质量增强和时空信息编码网络的船舶海上轨迹预测方法[J]. 模式识别与人工智能， 2025， 38(1): 51–67.
[20] Qiang Huimin， Guo Zhiyuan， Xie Shiyuan， et al. MSTFormer: motion inspired spatial-temporal transformer with dynamic-aware attention for long-term vessel trajectory prediction[PP/OL]. (2023–03–21)[2025–10–27]. https://arxiv.org/abs/2303.11540.
[21] Gu A， Dao T. Mamba: linear-time sequence modeling with selective state spaces[PP/OL]. (2024–05–31)[2025–10–27]. https://arxiv.org/abs/2312.00752.
[22] Biderman S， Schoelkopf H， Anthony Q G， et al. Pythia: a suite for analyzing large language models across training and scaling[PP/OL]. (2023–05–31)[2025–10–27]. https://arxiv.org/abs/2304.01373.
[23] Wang Ben， Komatsuzaki A. GPT–J–6B: a 6 billion parameter autoregressive language model[PP/OL]. (2023–05–26)[2025–10–27]. https://arxiv.org/abs/2212.14815.
[24] Zhang Susan， Roller S， Goyal N， et al. OPT: open pre-trained transformer language models[PP/OL]. (2022–06–21)[2025–10–27]. https://arxiv.org/abs/2205.01068.
[25] Poli M， Massaroli S， Nguyen E， et al. Hyena hierarchy: towards larger convolutional language models[PP/OL]. (2023–04–19)[2025–10–27]. https://arxiv.org/abs/2302.10866.
[26] Zhu Lianghui， Liao Bencheng， Zhang Qian， et al. Vision Mamba: efficient visual representation learning with bidirectional state space model[PP/OL]. (2024–11–14)[2025–10–27]. https://arxiv.org/abs/2401.09417.
[27] Patro B， Agneeswaran V. SiMBA: simplified Mamba-based architecture for vision and multivariate time series[PP/OL]. (2024–04–24)[2025–10–27]. https://arxiv.org/abs/2403.15360.
[28] Mehrens A R， Hahmann A N， Larsén X G， et al. Correlation and coherence of mesoscale wind speeds over the sea[J]. Quarterly Journal of the Royal Meteorological Society， 2016， 142(701): 3186–3194.
[29] Gu A， Goel K， Ré C. Efficiently modeling long sequences with structured state spaces[PP/OL]. (2022–08–05)[2025–10–27]. https://arxiv.org/abs/2111.00396.
[30] Gu A， Dao T， Ermon S， et al. HiPPO: recurrent memory with optimal polynomial projections[PP/OL]. (2020–10–23)[2025–10–27]. https://arxiv.org/abs/2008.07669.
基本信息:
DOI：10.13637/j.issn.1009-6094.2025.1917
中图分类号:U675.79
引用信息:
[1]陈万志,敖衔洋,李鑫.基于选择性状态空间的船舶轨迹预测模型[J].安全与环境学报().DOI:10.13637/j.issn.1009-6094.2025.1917.
基金信息:
国家重点研发计划项目（2018YFB1403303）; 辽工程GPU资源支持项目（2026–01）
2026-08-21
2026-08-21
2026-08-21