基于改进GRU模型的高速公路短时交通量预测  

Short-term traffic flow prediction method ofhighway based on improved GRU model

作　　者：温惠英[1] 元昱青 赵胜 WEN Huiying;YUAN Yuqing;ZHAO Sheng(School of Civil Engineering and Transportation South China University of Technology,Guangzhou 510641,China)

机构地区：[1]华南理工大学土木与交通学院,广东广州510641

出　　处：《广西大学学报（自然科学版）》2023年第2期459-468,共10页Journal of Guangxi University（Natural Science Edition）

基　　金：国家自然科学基金项目(52172345)

摘　　要：为了提高短时交通流的预测精度,采用灰狼算法(grey wolf optimizer,GWO)优化门控循环单元(gate recurrent unit,GRU)神经网络参数,构建基于超参数自适应寻优的高速公路短时交通流量预测模型,提取交通流的时变特征,准确预测短时交通流量。选取高速公路出口匝道交通数据作为实验数据输入,基于TensorFlow为后端的Keras完成GWO-GRU模型框架的搭建,并与支持向量回归算法(support vector regression,SVR)、K近邻算法(k-nearest neighbor,KNN)、长短期记忆神经网络(long-short term memory,LSTM)、门控循环单元(GRU)模型进行对比分析。实验结果表明,在3种不同时间间隔的高速公路匝道交通数据集的预测中,改进后的GRU模型具有较好的预测性能,其平均绝对误差(mean absolute error,MAE)比次优模型分别减小了9.22%、8.54%、8.03%。In order to improve the prediction accuracy of short-term traffic flow,this paper used grey wolf optimizer(GWO)to optimize neural network parameters of gate recurrent unit(GRU).A short-term highway traffic flow prediction model based on hyperparameter adaptation was constructed to extract time-varying characteristics of traffic flow and predict short-term traffic flow accurately.The highway exit ramp traffic data was selected as the experimental data input,and the GMO-GRU model framework was built based on TensorFlow as the back-end Keras.The prediction performance was compared with SVR,KNN,LSTM and GRU models.The experimental results show that the mean absolute error(MAE)of the improved GRU model is 9.22%,8.54%and 8.03%lower than that of the suboptimal model,respectively,in the prediction of the highway ramp traffic data sets at three different time intervals,indicating that the improved GRU model has better prediction performance.

关 键 词：高速公路 短时交通量预测 灰狼算法 超参数自适应优化 门控循环单元神经网络 

分 类 号：U491.31[交通运输工程—交通运输规划与管理]