基于GRU自编码器的船舶航线提取

船舶自动识别系统(automatic identification system, AIS)数据中蕴含着大量的船舶行为相关信息，从中提取出有效的航线，在海事监管、船只勘查等方面具有广泛应用。本文提出一种基于GRU自编码器(gate recuurent unit auto-encoder，GRU-AE)的船舶航线提取方法，首先采用GRU编码器将原始轨迹数据编码为统一格式的深度特征信息，其次利用DBSCAN (density-based spatial clustering of applications with noise)算法对深度特征信息进行聚类，最后将深度特征类簇中心通过解码器反演生成相应的船舶航线，从而实现在海量AIS数据中挖掘船舶轨迹规律。以波士顿港口为例，分析一年时间内10万多条AIS的船舶航行数据，实验表明本方法可对不同长度轨迹数据进行聚类及其航线提取，并可支撑船舶轨迹异常检测、路径规划、位置预测等研究，具有较好的应用适应性。

Ship route extraction based on GRU auto-encoder

The automatic identification system (AIS) data consists of a large amount of information associated with shipping behavior. A wide range of applications in maritime supervision and ship surveys can be made use of by extracting useful ship routes from AIS data. The gate recurrent unit auto-encoder (GRU-AE) is the foundation of a method for ship route extraction that is suggested. First, the GRU encoder is employed to encode the original trajectory data into deep feature information in a unified form; then, deep feature information is clustered using the DBSCAN algorithm; and finally, the deep feature cluster center is inverted through the decoder to produce the corresponding ship route, to achieve the mining of the ship trajectory pattern in the massive AIS data. Taking the Port of Boston as an example, more than 100,000 AIS ship navigation data in one year are examined. Experiments reveal that this technique can cluster and extract route data of various lengths and can support the research of ship trajectory abnormality identification, path planning, position prediction, etc., revealing good application adaptability.