基于数据编解码的时空交通流预测方法  

Spatio-temporal traffic flow prediction method based on data encoding and decoding

作　　者：龙佰超 关为生 肖建力[1] LONG Baichao;GUAN Weisheng;XIAO Jianli(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《上海理工大学学报》2023年第2期120-127,共8页Journal of University of Shanghai For Science and Technology

基　　金：国家自然科学基金资助项目(61603257,61906121)。

摘　　要：针对路网的拓扑信息不完整而无法实现时空结合交通流预测的情况,提出了一种基于时间序列预测模型联合数据编解码机制的预测方法。对路网内路段交通流数据进行编码得到路网信息的链状结构,以此获取路网结构中的拓扑信息;通过时序模型对链状结构进行交通流预测,完成对链状结构的时序特征提取;最终,通过解码方法得到路网的时空交通流预测结果。采用GPS数据,选取不同路网进行对比实验,引入数据编解码的时空交通流预测方法与时间序列模型进行比较,并且与基线模型HA和ARIMA展开了对比实验。实验结果表明:深度学习模型引入数据编解码机制后,模型性能明显提升;引入数据编解码机制的深度学习模型的性能比基线模型的性能更优越。该方法仅仅使用简单的时间序列深度网络再联合数据的编解码机制即可实现时空结合的交通流预测。A prediction method based on a time series prediction model combined with a data encoding and decoding mechanism was proposed for the situation that the topological information of the road network is incomplete,and it is not possible to realize the spatio-temporal traffic flow prediction.The chain structure of road network information was obtained by encoding the traffic flow data of road sections in the network,so as to obtain the topological information in the road network structure.The traffic flow prediction of the chain structure was carried out by the time series model.The extraction of the temporal features of the chain structure was completed.Finally,the spatio-temporal traffic flow prediction results of the road network were obtained by the decoding method.Different road networks and their GPS data were selected for comparison experiments.The spatio-temporal traffic flow prediction method of data encoding and decoding was introduced for comparison with the time series model and the baseline models of HA and ARIMA.The experimental results show that the performance of the deep learning model is significantly improved after the data encoding and decoding mechanism is introduced.The performance of the deep learning model with this mechanism is superior to that of the baseline model.The proposed method can achieve spatio-temporal traffic flow prediction only using a simple time series deep network combined with the data encoding and decoding mechanism.

关 键 词：智能交通系统 交通流预测 数据编解码 时序模型 深度学习 

分 类 号：U491.112[交通运输工程—交通运输规划与管理]