基于CNN-F-LSTM-Attention的船舶轨迹预测

Ship Trajectory Prediction Based on CNN-F-LSTM-Attention

随着经济全球化进程加快,国际贸易规模不断扩大,海上运输业快速发展,在交通密度大、条件复杂的港口,交通安全管理面临着巨大挑战。船舶碰撞是海上频发的事故类型之一,准确的船舶轨迹预测对于海上交通管理以及保障船舶行驶安全极为重要。目前常用的船舶轨迹预测方法为长短时记忆网络,但其具有大量门控权重参数,结构复杂,且对空间及时序特征挖掘不够充分。针对以上问题,提出一种结合卷积神经网络、改进后长短时记忆网络与注意力机制的船舶轨迹预测模型。该模型通过改进后的长短时记忆网络降低了结构复杂性,提高了训练速度和泛化性能;同时引入卷积神经网络充分挖掘轨迹数据的空间特征和时序特征,通过注意力机制为不同特征分配不同权重,以过滤无用的特征信息,提高模型精度。在真实数据集上的实验结果表明,所提模型无论是经纬度还是航向航速预测方面,准确率均较对照主流模型有所提升。

With the acceleration of economic globalization and the continuous expansion of international trade,the maritime transportation industry is developing rapidly.In ports with high traffic density and complex conditions,traffic safety management is facing enormous challenges.Ship collision is one of the frequent types of accidents at sea,and accurate ship prediction is extremely important for maritime traffic management and ensuring the safety of ship navigation.The commonly used method for predicting ship trajectories is the Long Short Term Memory Network,but it has a large number of gate control weight parameters,a complex structure,and insufficient exploration of spatial and temporal features.A ship trajectory prediction model combining convolutional neural network,improved long short-term memory network,and attention mechanism is proposed to address the above issues.This model reduces structural complexity,improves training speed and generalization performance through an improved long short-term memory network;At the same time,convolutional neural networks are introduced to fully explore the spatial and temporal features of trajectory data,and different weights are assigned to different features through attention mechanisms to filter out useless feature information and improve model accuracy.The experimental results on real datasets show that the proposed model has improved accuracy in predicting latitude,longitude,heading,and speed compared to mainstream control models.