海洋测绘Issue(4)：43-47,5.DOI:10.3969/j.issn.1671-3044.2025.04.009
动态多头注意力与跨层优化的船舶轨迹预测算法
Ship trajectory prediction with dynamic multi-head attention and cross-layer optimization
孟菲 1耿晓晖 1刘卓然 2甄超 1徐素宁3
作者信息
- 1. 中国四维测绘技术有限公司,北京 100089
- 2. 北京建筑大学测绘与城市空间信息学院,北京 102616
- 3. 自然资源部国土卫星遥感应用中心,北京 100048
- 折叠
摘要
Abstract
Accurate ship trajectory prediction can lay the foundation for maritime traffic management,reduce the risk of ship collision and improve the efficiency of search and rescue.However,the existing methods have the problem of insufficient capture of data quality and spatio-temporal characteristics in the process of ship trajectory prediction.In order to improve the accuracy and stability of ship trajectory prediction,this paper proposes a TrAISformer-AB algorithm which combines dynamic multi head attention and cross layer optimization.Firstly,an adaptive multi head attention mechanism is designed to make the model dynamically select the number of attention heads according to the complexity and feature relationship of the input data,so as to improve the learning ability and generalization ability of the model.Secondly,relu6 is used as the activation function to enhance the expression ability and stability of the model.Finally,the cross layer residual connection is designed to reduce the gradient disappearance problem and improve the stability and convergence speed of training.Experimental results show that the TrAISformer-AB algorithm has the lowest prediction error compared with the comparison algorithm,and the prediction accuracy is improved by 70％compared with the benchmark model.The accuracy and stability of ship trajectory prediction are effectively improved.关键词
船舶轨迹预测/自适应多头注意力机制/ReLU6激活函数/跨层残差连接/船舶自动识别系统Key words
ship trajectory prediction/adaptive multi-head attention mechanism/ReLU6 activation function/cross-layer residual connection/ship automatic identification分类
天文与地球科学引用本文复制引用
孟菲,耿晓晖,刘卓然,甄超,徐素宁..动态多头注意力与跨层优化的船舶轨迹预测算法[J].海洋测绘,2025,(4):43-47,5.
