基于时空特征融合的短时交通流预测

Short-term Traffic Flow Prediction Based on Spatial-temporal Feature Fusion

城市道路交通拥堵加大碳排放和空气环境污染问题,短时交通流量预测能够有效缓解交通堵塞。本文提出一种基于时空特征融合的短时交通流预测模型。该模型通过点互信息(PMI)算法对监测站点做相关性分析,确定相关性较高站点,并将其交通数据处理成周期性序列和邻近序列;引入长短时记忆(LSTM)网络提取时间特征构建相关模型,完成时间和空间特征的融合;引进绝对误差序列分析优化模型,得到最终预测结果。本研究以长沙橘子洲大桥作为目标站点,大桥两端各个主要交通路口作为监测站点,利用各个站点的交通流数据集对模型进行验证。研究结果表明:该预测模型优于传统反向传播神经(BP)网络模型和LSTM模型,在平均绝对百分误差(MAPE)指标上,该模型相较于BP和LSTM分别降低3.12%和1.58%,在均方根误差(RMSE)指标上,该模型分别降低了8.45和3.34,为解决交通拥堵和减少碳排放问题提供了一定的参考。

Urban road traffic congestion increases carbon emissions and causes air pollution.Short-term traffic flow prediction can effectively alleviate the traffic congestion problem,so this paper proposes a short-term traffic flow prediction model which based on spatio-temporal feature fusion.The core idea of the model is the correlation analysis of monitoring stations is by the using Point Mutual Information(PMI)algorithm,the stations with high correlation are determined and their traffic data are processed into periodic sequence and adjacent sequence.The long-short-term memory(LSTM)network is introduced to extract temporal features and construct correlation models and the fusion of temporal and spatial features is completed.The research takes the Orange Island Bridge as the target site and the major traffic intersections at both ends of the bridge as the monitoring sites,and the model is verified by using the traffic flow data sets of the various stations.The research results show that the prediction model outperforms the traditional back propagation neural(BP)network model and LSTM model,and in the mean absolute percentage error(MAPE)index,the model is 3.12%and 1.58%lower than BP and LSTM,respectively,and in the root mean square error(RMSE)index,the model is 8.45 and 3.34 lower,respectively.It provides some reference for solving traffic congestion and reducing carbon emission problems.