摘要
为更实时准确地评估和预测交通运行态势,以交通检测器提供的大量数据作为研究对象,通过分析交通流时空特征及变化规律,建立考虑谱聚类的K近邻算法,并对城市交通的运行态势进行等级分层并评估。在时间序列分析的基础上,建立考虑时间特性的LSTM和GRU神经网络模型,预测城市交通运行态势并分析对比各模型预测效果。研究结果表明:与传统ARIMA模型相比,改进的GRU神经网络模型综合效果更好。
In order to accurately assess and predict the traffic situation in real time, a large amount of data provided by the traffic detector is used as the research object.By analyzing the temporal and spatial characteristics and changing laws of traffic flow, a K-Nearest Neighbor model based on spectral clustering is established for traffic situation assessment.Based on time series analysis, LSTM and GRU neural network models considering time characteristics are established to predict urban traffic situation and analyze and compare the prediction effects of each model.The research results show that compared with the traditional ARIMA model, the improved GRU neural network model works better.
作者
刘晶鑫
唐婕
曹瑾鑫
LIU Jingxin;TANG Jie;CAO Jinxin(Institute of Transportation,Inner Mongolia University,Hohhot 010070,China)
出处
《内蒙古大学学报（自然科学版）》
CAS
北大核心
2021年第2期198-205,共8页
Journal of Inner Mongolia University：Natural Science Edition
基金
国家自然科学基金(71961024)
内蒙古自治区关键技术攻关计划项目(2019GG287)。
关键词
交通工程
交通态势评估预测
循环神经网络模型
机器学习
traffic engineering
traffic situation assessment and forecast
recurrent neural network model
machine learning
