Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (9): 181-186.DOI: 10.3778/j.issn.1002-8331.2011-0204

• Pattern Recognition and Artificial Intelligence • Previous Articles     Next Articles

Spatiotemporal Short-Term Traffic Flow Prediction Based on Broad Learning System

LUO Xianglong, GUO Huang, LIAO Cong, HAN Jing, WANG Lixin   

  1. 1.School of Information Engineering, Chang’an University, Xi’an 710064, China
    2.China Railway First Survey and Design Institute Group Co., Ltd., Xi’an 710043, China
  • Online:2022-05-01 Published:2022-05-01

时空相关的短时交通流宽度学习预测模型

罗向龙,郭凰,廖聪,韩静,王立新   

  1. 1.长安大学 信息工程学院,西安 710064
    2.中铁第一勘察设计院集团有限公司,西安 710043

Abstract: In order to solve the problem of long operation time in deep learning prediction model, the spatio-temporal correlation characteristic of traffic big data is mined sufficiently, a short-term traffic flow prediction model is proposed based on K-nearest neighbor(KNN) and broad learning system(BLS). KNN algorithm is used to choose K road sections, which has high spatio-temporal correlation with the predicted road section. Traffic flow data of selected road sections as the input of BLS model are used to predict respectively. The prediction results of selected sections are weighted, and the K prediction results with the minimum RMSE are taken as the final prediction values. The test results using dataset from the California PeMs show that the RMSE of the proposed model is reduced 46.56% compared with ARIMA, WNN, LSTM and KNN-LSTM model, the operation efficiency is improved obviously, thus KNN-BLS is an effective method for short-term traffic flow prediction.

Key words: traffic flow prediction, deep learning, broad learning, spatio-temporal correlation

摘要: 针对深度学习预测模型运算大的问题,在充分挖掘交通大数据的时空相关性的基础上,提出了一种基于K-最邻近(K-nearest neighbor,KNN)与宽度学习系统(broad learning system,BLS)相结合的短时交通流预测模型。利用KNN算法筛选与预测路段时空相关性高的K个路段,将选取路段的交通流数据作为BLS模型的输入分别进行预测,对选取不同路段的预测结果进行加权,以均方根误差(root mean square error,RMSE)为最小时对应K值的结果作为最终的预测值。美国加利福尼亚州交通局PeMs交通数据库实测的交通流量数据的测试结果表明,提出的模型相比于ARIMA、WNN、LSTM、KNN-LSTM模型均方根误差平均降低46.56%,运算效率明显提高,是一种有效的短时交通流预测方法。

关键词: 交通流预测, 深度学习, 宽度学习, 时空相关