基于时空信息融合学习的路段行程车速短时预测  

Short-Term Prediction of Road Travel Speed Based on Spatio-Temporal Information Fusion Learning

作　　者：杨顶 邓明君 徐丽萍 YANG Ding;DENG Mingjun;XU Liping(School of Transportation and Logistics,East China Jiaotong University,Nanchang 330013,China)

机构地区：[1]华东交通大学交通运输与物流学院,南昌330013

出　　处：《计算机工程》2021年第12期78-86,共9页Computer Engineering

基　　金：国家自然科学基金(51965021);江西省自然科学基金(20142BAB201015);江西省教育厅科研项目(CJJ160476)。

摘　　要：路段行程车速的变化受时间和空间维度信息的综合影响,多数神经网络模型仅从时间维度上预测路段行程车速的变化规律,未能全面考虑路网结构和上下游交通状态对路段行程车速的影响。结合图卷积网络和门控循环单元构建深度学习模型,挖掘路段行程车速的时空特性。通过在线地图平台获取路段实时行程车速,使用等维递补方法更新历史序列数据,提高预测实时性。在深圳市部分区域路网上的实验结果表明,该模型的多步预测精度均在90%以上,相比自回归积分滑动平均模型、支持向量机回归模型和门控循环单元模型最高提升了6.9%、1.3%和0.4%,具有更优的路段行程车速预测效果。The change of road travel speed in a certain road is affected by both temporal and spatial factors.Most of the existing neural network models predict the change of road travel speed by considering only the temporal factors,and ignore the impact of road network structure and upstream and downstream traffic.To address the problem,the spatiotemporal features of road travel speed are analyzed,and on this basis a deep learning model is constructed by combining the Graph Convolutional Network(GCN)and the Gated Recurrent Unit(GRU).The real-time road travel speed is obtained through the online map platform,and the historical sequence data is continuously updated by using the method of equal-dimensional recursive compensation,so the real-time performance of prediction is improved.The model is tested with the data of road networks in some regions of Shenzhen.The experimental results show that the proposed model exhibits excellent performance in predicting travel speed in a certain road section,and provides an accuracy of over 90%for multi-step prediction.Its prediction accuracy is up to 6.9%higher than that of the Auto-Regressive Integral Moving Average(ARIMA)model,1.3%higher than that of the Support Vector Regression(SVR)model,and 0.4%higher than that of the GRU model.

关 键 词：短时预测 图卷积网络 门控循环单元 时空相关性 在线地图 等维递补 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]