基于CBAM&ConvLSTM的短时交通拥塞预测
Short-term traffic congestion prediction based on CBAM&ConvLSTM
余文斌 1沈鑫禹 1钱铭 1冯昊 1王苏勋 1张成军1
作者信息
- 1. 南京信息工程大学,江苏大气环境和装备技术协同创新中心,南京 210044;南京信息工程大学计算机学院,南京 210044;南京信息工程大学数字取证教育部工程研究中心,南京 210044
- 折叠
摘要
短时交通拥塞预测是智能交通的重点问题,其难点在于时空序列的数据处理和特征提取.卷积长短期记忆网络(Convolutional Long Short-Term Memory,ConvLSTM)适合处理兼具时间和空间相关性的交通数据.而卷积注意力机制(Convolutional Block Attention Module,CB AM)在空间和时间维度引入注意力机制,使模型对于数据的变化更加敏感.文中结合ConvLSTM和CBAM,设计了 一种新模型,对短时交通拥塞进行预测.实验基于百度地图实时数据,并与其他主流模型进行了比较.结果表明,该模型在交通数据的适应性方面优于其他模型,为解决交通拥塞的预测问题提供了一种新的思路和方法.
Abstract
Short-term traffic congestion prediction is a crucial key issue in intelligent transportation,which is difficult to reflect in the data processing and feature extraction of spatiotemporal sequences.Convolutional Long Short-Term Memory(ConvLSTM)is suitable for processing data with temporal and spatial correlation.Convolutional Block Attention Module(CBAM)introduces an attention mechanism in the spatial and tempo-ral dimensions,making the model more sensitive to the data.Therefore,this paper combines ConvLSTM and CBAM to design a new model to predict traffic congestion.Experiments are based on real-time data of Baidu Maps and compared with mainstream models.The results show that the model outperforms other mod-els in the adaptability of traffic data,which provides a new idea and method for solving the problem of traf-fic congestion prediction.
关键词
深度学习/短时交通拥塞预测/卷积长短期记忆网络/卷积注意力机制/时空预测Key words
deep learning/short-term traffic congestion prediction/Convolutional Long Short-Term Memo-ry Network/Convolutional Block Attention Module/spatiotemporal prediction引用本文复制引用
基金项目
国家自然科学基金(61501247)
国家自然科学基金(61703212)
国家自然科学基金(61802197)
出版年
2024
NETL
NSTL
万方数据