摘要
为了有效地实施智能交通管理系统,需要进一步提高交通流量预测的准确率。提出了一种基于门限递归单元循环神经网络的短时交通流量预测方法,该方法可以不依靠先验知识,有效地利用"序列信息"建模。通过使用该方法对加拿大大不列颠哥伦比亚省的真实交通流量数据进行建模分析,并对比了在不同滞后时间的输入数据下该方法的预测效果,然后将其与ARIMA和SVR的预测结果进行了对比,同时也展示了该方法在工作日和周末的实际预测效果。结果表明:该方法预测效果良好,其平均绝对百分误差比ARIMA与SVR分别平均降低了74.72%和12.15%,预测值和实际交通流量吻合度高,是一种预测精度高且有效的交通流量预测方法。
In order to effectively implement the intelligent traffic management system,it is necessary to improve the accuracy of traffic flow prediction.A short-term traffic flow prediction method was proposed based on the gated recurrent unit recurrent neural network.The new method can effectively model the“sequence information”without relying on the prior knowledge.By adopting this method,the real traffic flow data of British Columbia in Canada was modeled and analyzed,and the predicted effects by inputting data under different lag time were compared.In the meantime,the predicted results were compared with the ones of ARIMA and SVR.The study also demonstrated the predicted effect of new method in weekdays and weekends.The comparison results show that the new method performs well,and the mean absolute percentage errors reduced 74.72%and 12.15%respectively comparing with ARIMA and SVR.This new method achieves a high conformity between the predicted and actual traffic flow,which is effective and accurate for the future traffic flow prediction.
作者
王体迎
时鹏超
刘蒋琼
刘博艺
时天昊
WANG Tiying;SHI Pengchao;LIU Jiangqiong;LIU Boyi;SHI Tianhao(Mechanical and Electrical College,Hainan University,Haikou 570228,Hainan,P.R.China;Institute of Tropical Agriculture and Forestry,Hainan University,Haikou 570228,Hainan,P.R.China;College of Information Science&Technology,Hainan University,Haikou 570228,Hainan,P.R.China;College of Transportation,Shandong University of Science and Technology,Qingdao 266590,Shandong,P.R.China)
出处
《重庆交通大学学报（自然科学版）》
CAS
北大核心
2018年第11期76-82,共7页
Journal of Chongqing Jiaotong University(Natural Science)
基金
海南省自然科学基金项目(20155212)
关键词
交通运输工程
智能交通系统
交通流量预测
门限递归单元
递归神经网络
traffic and transportation engineering
intelligent transportation system
traffic flow prediction
gated recurrent unit
recurrent neural network
