摘要
为解决道路运行速度预测实际应用中数据缺失和预测精度低的问题,提出了一种基于时空图卷积的速度预测模型(STGCN)。通过引入时空图卷积机制来建模路网速度的时间依赖和空间依赖关系,可有效提取动态和多模态的路网速度时空依赖。针对工程应用中数据缺失问题,在深圳道路运行速度数据集上,基于均值、小波滤波和cart树分别对缺失数据进行补全,并在模型训练时采取掩码机制,增加可用的训练数据量,降低数据缺失对模型精度影响。模型预测结果表明,对于相同的数据集,与常见的FC-LSTM模型相比,STGCN模型可提升2.32%的预测精度,引入cart树进行缺失数据补全后速度预测误差降低3.45%,按照不同道路等级分类后分别对模型训练时模型预测精度MAPE降低4.35%。STGCN模型对于国内真实数据下的道路运行速度预测分析和提升速度精度具有重要意义。
In order to solve the problem of missing data and low prediction accuracy in the practical application of road speed prediction,a speed prediction model based on spatiotemporal graph convolution(STGCN)is proposed.By introducing the spatiotemporal graph convolution mechanism to model the temporal and spatial dependencies of road network speed,it can effectively extract dynamic and multimodal road speeds.Aiming at the problem of missing data in engineering applications,on the Shenzhen road speed data set,the missing data is completed based on the mean value,wavelet filtering and cart tree,and a mask mechanism is adopted during model training to increase the amount of available training data and reduce the impact of missing data on model accuracy.The model prediction results show that the prediction accuracy of STGCN is 2.32%higher than that of the FC-LSTM model in the same data set,and the model MPAE is relatively reduced by 3.45%after introducing the cart tree to complete the missing data.When the models are trained separately by road type,the model MPAE is relatively reduced by 4.35%.The STGCN model proposed in this paper is of great significance for the prediction and analysis of the speed of the road network under the quality of domestic real data and to improve the speed accuracy.
作者
胡铮
戴东生
林杨
曾秋霖
严伟
吕楷超
HU Zheng;DAI Dongsheng;LIN Yang;ZENG Qiulin;YAN Wei;LV Kaichao(Ningbo Transportation Development Research Center,Ningbo 315000,China;Shenzhen Urban Transport Planning Center Co.,Ltd,Shenzhen 518021,China)
出处
《交通与运输》
2023年第2期30-36,共7页
Traffic & Transportation
关键词
速度预测
图卷积
数据补全
时空预测
智能交通
Speed prediction
Graph convolution
Data complete
Spatiotemporal forecasting
Intelligent transportation
