基于EEMD-Att-GCN-LSTM的高速公路交通流量预测
Prediction of Expressway Traffic Flow Based on EEMD-Att-GCN-LSTM
孙金鑫 1钟剑 2姚磊 1韩明君 2戴涛 1苏婧2
作者信息
- 1. 湖北随岳南高速公路有限公司,湖北 监利 433300
- 2. 北京诚达交通科技有限公司,北京 100088
- 折叠
摘要
由于天气、突发事件、时间等因素影响,导致交通流具有非线性和不平稳性,这对现有的交通流预测十分有挑战.目前高速公路流量预测大多是对时间序列进行建模分析,缺乏对数据结合动态空间相关性的建模.为满足快速、准确预测高速公路流量,提出一种基于注意力机制、经验模态分解、图卷积神经网络和长短期记忆网络的组合预测模型.首先通过集合经验模态分解对交通流数据进行去噪,降低高速公路流量的非平稳性;然后用时空注意力机制动态捕获时空相关性.将时间特征和空间特征的输出加权融合并作为全连接层的输入,最后以均方根误差、平均绝对误差和平均绝对百分比误差来对此模型进行性能评价,相较于其他模型,所提出的模型在结合EEMD进行数据去噪处理后能够显著提升预测精度,同时在中长期流量预测方面也具有良好的实用性.
Abstract
The current predictions of expressway traffic flow primarily rely on time series modeling,without incorporating dynamic spatial correlation into the modeling process.For the purpose of efficiently and accurately predicting expressway traffic flow,a combined prediction model based on attention mechanism(ATT),ensemble empirical mode decomposition(EEMD),graph convolutional neural network(GCN),and long short-term memory(LSTM)is proposed.Firstly,EEMD is em-ployed to denoise the traffic flow data,thereby reducing the non-stationarity inherent in expressway traffic flow.Subsequently,a spatio-temporal attention mechanism is utilized to dynamically capture the spatio-temporal correlations.Finally,the outputs of temporal and spatial features are weighted and fused as inputs into a fully connected layer for comprehensive analysis.The performance evaluation metrics of this model includes mean absolute error(MAE),mean absolute percentage error(MAPE),and root mean square error(RMSE).It exhibits practicality in medium and long-term traffic flow prediction.
关键词
智能交通/交通流量预测/注意力机制/经验模态分解/图卷积网络/长短时记忆网络/高速公路Key words
ITS/traffic-flow prediction/ATT/EEMD/GCN/LSTM/expressway引用本文复制引用
出版年
2024
NETL
NSTL
万方数据