基于Encoder-Decoder LSTM的船舶轨迹预测方法

A prediction method of vessel trajectory based on Encoder-Decoder LSTM

船舶轨迹预测的准确性是预警事故和确保安全航行的关键,但精度和稳定性是目前亟待解决的问题。为了应对这一挑战,提出一种基于Encoder-Decoder LSTM的船舶轨迹预测方法。首先对船舶AIS轨迹数据进行去噪、分段、插值、停留点检测和归一化等预处理,提取船舶的航行轨迹。然后构建了一种基于Encoder-Decoder LSTM架构的船舶轨迹预测模型,并对模型参数进行初始化。最后使用长江江苏段天生港水域渡轮的真实AIS数据对提出的模型进行训练和验证,并与其他广泛使用的轨迹预测方法进行比较。定量分析表明,该方法可实现对船舶轨迹较为准确的预测,且所预测轨迹具有一定的参考价值。

Accurately predicting vessel trajectory is crucial for early warning and safe navigation,yet accuracy and stability remain major need to be solved at present.To remedy this,a vessel trajectory prediction method based on an Encoder-Decoder LSTM neural network is proposed.Firstly,vessel AIS trajectory data is preprocessed using methods such as denoising,segmentation,interpolation,stay point detection,and normalization to extract vessel sailing trajectories.Next,a vessel trajectory prediction model based on the Encoder-Decoder LSTM architecture is constructed,and the model parameters are initialized.Finally,the proposed model is trained and validated using real AIS data of ferries in the Tianshenggang waters in the Jiangsu section of the Yangtze River and compared with other widely-used trajectory prediction models.The results shows that this method can achieve accurate prediction of trajectories,and the predicted trajectories have a significant reference value.