传统人工监管方式难以适配大规模AIS（Automatic Identification System）轨迹数据的分析需求。本文构建了适配AIS数据质量特性的预处理体系，提出深度时空自编码网络（ST-AE）模型，通过CNN与Bi-LSTM协同融合架构及注意力机制捕捉轨迹时空依赖关系，结合轻量化设计实现检测精度与实时性的平衡，并配套设计模型训练策略与自适应异常判定机制。基于AIS数据集构建实验环境，选取滑动窗口+3σ阈值、随机森林、常规LSTM自编码等主流方法与本文模型进行对比，结果表明ST-AE模型在准确率、召回率及实时性上均优于其他3种方法，可有效适配多种船舶异常行为检测需求。研究结果可为海事智能监管提供了可靠的技术支撑。
Traditional manual supervision methods are difficult to adapt to the analysis requirements of large-scale Automatic Identification System (AIS) trajectory data. In this study, a preprocessing system adapted to the quality characteristics of AIS data was constructed, and a deep spatio-temporal autoencoder (ST-AE) model was proposed. The ST-AE model captures the spatio-temporal dependencies of trajectories through the synergistic fusion architecture of Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (Bi-LSTM) combined with an attention mechanism. Meanwhile, a lightweight design is integrated to achieve a balance between detection accuracy and real-time performance. Corresponding model training strategies and adaptive anomaly determination mechanisms were also developed to support the model. An experimental environment was established based on AIS datasets, and three mainstream methods, namely sliding window + 3σ threshold, random forest, and conventional LSTM autoencoder, were selected for comparative experiments with the proposed ST-AE model. The results indicate that the ST-AE model outperforms the other three methods in terms of accuracy, recall, and real-time performance, and can effectively meet the detection requirements of various ship abnormal behaviors. This study provides reliable technical support for maritime intelligent supervision.
2026,48(5): 146-150 收稿日期：2025-9-30
DOI：10.3404/j.issn.1672-7649.2026.05.023
分类号：U664
基金项目：2026年度广西自然科学基金（2025JJH160108）；2023年度钦州市科学研究与技术开发计划项目（20233141）；2025年度钦州市科学研究与技术开发计划项目（20251706）；2024年度广西高等教育本科教学改革工程项目（2024JGB275）
作者简介：董海亮(1985-),男,副教授,研究方向为船舶交通安全与管理及航海教育
参考文献：
[1] 李业, 任鸿翔, 张政. 基于Encoder-Decoder LSTM的船舶轨迹预测方法[J]. 海洋测绘, 2024, 44(1): 21-25
LI Y, REN H X, ZHANG Z. Vessel trajectory prediction method based on encoder-decoder LSTM[J]. Hydrographic Surveying and Charting, 2024, 44(1): 21-25
[2] 冀娜, 胡磊. 深度学习算法下船舶航行轨迹异常行为检测方法[J]. 舰船科学技术, 2025, 47(15): 164-167
JI N, HU L. Abnormal behavior detection method of vessel navigation trajectory based on deep learning algorithm[J]. Ship Science and Technology, 2025, 47(15): 164-167
[3] 吕国华, 胡学先, 杨明, 等. 基于联邦随机森林的船舶AIS轨迹分类算法[J]. 信息网络安全, 2022, 22(4): 67-76
LV G H, HU X X, YANG M, et al. Vessel AIS trajectory classification algorithm based on federated random forest[J]. Netinfo Security, 2022, 22(4): 67-76
[4] 王超, 刘润泽, 郭春燕, 等. 船舶AIS轨迹压缩算法应用发展研究[J]. 珠江水运, 2024(10): 131-134
WANG C, LIU R Z, GUO C Y, et al. Research on application and development of vessel AIS trajectory compression algorithm[J]. Pearl River Water Transport, 2024(10): 131-134
[5] 孙波, 黄永忠, 李峥. 基于改进DBSCAN算法的船舶异常行为检测[J]. 武汉船舶职业技术学院学报, 2025, 24(5): 105-109
SUN B, HUANG Y Z, LI Z. Vessel abnormal behavior detection based on improved DBSCAN algorithm[J]. Journal of Wuhan Institute of Shipbuilding Technology, 2025, 24(5): 105-109
[6] 杨阿华, 常鑫, 张强, 等. 基于AIS数据的敏感水域船舶异常行为检测[J]. 指挥信息系统与技术, 2023, 14(4): 32-37
YANG A H, CHANG X, ZHANG Q, et al. Vessel abnormal behavior detection in sensitive waters based on AIS data[J]. Command Information System and Technology, 2023, 14(4): 32-37
