摘要
传统的轨迹相似度计算方法大多是针对欧氏空间中的轨迹而设计的,忽略了真实的轨迹被交通路网限制的这一事实,无法全面反映实际状况.本文提出了一种基于图神经网络和引入注意力机制的长短期记忆网络的深度学习模型来解决传统方法存在的问题.首先,将轨迹由路网上的路口来表示.然后为路网构建三个独立的视图,其中融合视图由路网知识图嵌入学习路口之间的关系而构建,使用图神经网络和多层感知机学习三个视图的信息,设计视图融合层生成路口表示向量,最后使用多层引入注意力机制的长短期记忆网络来学习轨迹的表示向量,同时还设计了一个判定轨迹相似度的损失函数.在两个真实的城市数据集上的实验结果表明,本文提出的计算方法相较于基线方法具有更高的有效性和可行性.
Traditional trajectory similarity computation methods are predominantly designed for trajectories in Euclidean space,neglecting the fact that real trajectories are constrained by road networks,which fails to fully reflect the actual conditions.In this paper,we propose a deep learning model based on Graph Neural Networks and an attention-enhanced Long Short-Term Memory network to address the issues inherent in conventional methods.Firstly,trajectories are represented by intersections on the road network.Subsequently,three independent views are constructed for the road network,where the fusion view is built by learning the relationships between intersections using road network knowledge graph embedding.Graph Neural Networks and Multilayer Perceptron are utilized to learn information from these three views,and a view fusion layer is designed to generate intersection representation vectors.Finally,a multilayer attention mechanism with LSTM network is employed to learn the representation vectors of trajectories,and a loss function for determining trajectory similarity is also devised.The experimental results demonstrate that the computational method proposed in this paper exhibits superior effectiveness and feasibility compared to the baseline methods when applied to two real urban datasets.
作者
王语涵
孙未未
卢霖统
WANG Yuhan;SUN Weiwei;LU Lintong(School of Computer Science and Technology,Fudan University,Shanghai 200433,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2024年第11期2561-2568,共8页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(62172107)资助
国家重点研发计划项目(2018YFB0505000)资助.
关键词
轨迹相似度计算
图神经网络
长短期记忆网络
注意力机制
知识图谱
trajectory similarity computation
graph neural networks
long short-term memory networks
attention mechanism
knowledge graph embedding
