基于时空特征融合的舰船航迹预测方法

针对海上舰船日益增多、海情日益复杂的严峻形势,改进舰船航迹预测方法,实现对海域态势的有效管控成为亟待解决的问题。结合舰船航迹获取简单、数量较大的显著优势,提出利用舰船航迹数据驱动的基于时空特征融合的舰船航迹预测方法。首先,联合卷积神经网络(CNN)和长短时记忆网络(LSTM)构造时空特征融合网络,充分提取舰船航迹的时空特征；然后,利用海量 AIS(Automatic Identification System)数据进行网络训练；最后,利用网络输出的航速和航向对舰船航迹进行预测。仿真结果表明,提出的网络具有准确的舰船航迹预测能力,能够适应舰船机动运动场景。与传统预测方法相比,该方法能够使预测 MSE减少 0.2~1.4,预测性能大大提高。

A Ship Track Prediction Method Based on Spatial-temporal Feature Fusion

In view of the severe situation of increasing number of ships and increasingly complex sea conditions, it is ur.gent to improve the method of ship track prediction to achieve effective control of sea situations. Combined with the obvi. ous advantages of simple-getting and large number of ship track acquisition, this paper proposes a ship track prediction method based on spatial-temporal feature fusion driven by ship track data. Firstly, a spatial-temporal feature fusion net.work is constructed by combining CNN and LSTM to fully extract the spatiotemporal features of ship tracks. Then, massive AIS (Automatic Identification System) data are used for network training. Finally, the ship track is predicted by using the speed and course output from the network. Simulation results show that the network proposed in this paper has the ability to predict the ship track accurately and can be adapted to ship maneuvering scenarios. Compared with the traditional pre. diction method, the proposed method can reduce the predicted mean-square error by 0.2~1.4, which greatly improves the prediction effect.