横向相关性及参数影响下的车道级交通预测  

Lane Level Traffic Prediction under Influence of Lateral Correlation and Parameters

作　　者：侯越[1] 崔菡珂 邓志远 HOU Yue;CUI Han-ke;DENG Zhi-yuan(School of Electronics and Information Engineering,Lanzhou Jiaotong University,Lanzhou Gansu 730000,China)

机构地区：[1]兰州交通大学电子与信息工程学院,甘肃兰州730000

出　　处：《公路交通科技》2022年第5期122-130,共9页Journal of Highway and Transportation Research and Development

基　　金：国家自然科学基金项目(62063014)。

摘　　要：城市交通拥堵已成为阻碍城市发展的主要矛盾,鉴于道路交通受参数和时空因素影响,使得交通流预测模型精度不高,且在现实场景中易失效,同时以路段为研究对象的传统预测方法已无法满足智能网联技术发展的需求。为了解决传统预测模型在车道级横向空间相关性及参数影响方面考虑不足、且具有记忆局限性的问题,提高车道级短时交通流预测精度,提出PCA-stacked-GRU的车道级组合深度学习模型,采用主成分分析法对强相关车道的交通参数进行特征级融合,建立横向相关性及宏观参数相关性的数据组织,通过stacked-GRU模型实现横向相关性和参数影响的车道级交通流预测。在预处理后的数据集上验证可知,考虑车道级横向空间相关性和交通参数相关性的模型能够提高预测精度,微波测试集的流量和速度预测MAE分别为4.946960,3.109925,视频测试集的流量和速度预测MAE分别为4.503461,3.863718,相较于单一车道单一变量的预测模型精度均有较大的提高,与单一车道多变量的预测模型相比有更好的预测效果。所提出的PCA-stacked-GRU模型相较于其他常见的基准线模型,具有更高的预测精度和鲁棒性,能够应用于实际场景中的决策分析,提出交通拥堵问题的解决依据和更加精准、全面的交通信息,并为车道级地图导航和自动驾驶等提供数据支撑。Urban traffic congestion has become the main contradiction hindering urban development. Due to the influence of parameters and space-time factors on road traffic, the accuracy of traffic flow prediction model is not high, and it is easy to fail in the real scene. At the same time, the traditional prediction method based on road section cannot meet the demand of intelligent network technology development. To solve the problem of insufficient consideration and memory limitation of traditional prediction models in the influence of lane level horizontal spatial correlation and parameters, and improve the prediction accuracy of lane level short-term traffic flow, a PCA-stacked-GRU lane level combination depth learning model is proposed. The model adopts PCA method to feature level fuse the traffic parameters of strongly correlated lanes, establishes the data organization of horizontal correlation and macro parameter correlation, and realize the lane level traffic flow prediction with lateral correlations and parameter influence based on the stacked gate recurrent unit(stacked-GRU) model. The result of verification on the preprocessed dataset shows that the model which considering lane level lateral spatial correlation and traffic parameter correlation can improve the prediction accuracy, the predicted MAEs of flow and speed in microwave test set is 4.946 960 and 3.109 925, and those in video test set is 4.503 461 and 3.863 718. Compared with the prediction model of single lane and single variable, the accuracy of the model is greatly improved, and the prediction effect is better than that of the single lane multi-variable prediction model. Compared with other common baseline models, the proposed PCA-stacked-GRU model has higher prediction accuracy and robustness, it can be applied to decision analysis in actual scenarios, propose the basis for solving traffic congestion problems and more accurate and comprehensive traffic information, and provide data support for lane level map navigation and automatic driving.

关 键 词：城市交通 交通流预测 主成分分析 GRU 横向空间相关性 参数相关性 

分 类 号：U491[交通运输工程—交通运输规划与管理] TP183[交通运输工程—道路与铁道工程]