利用出租车GPS轨迹数据进行短时交通流量预测:以重庆市解放碑街区为例  

Short-term Traffic Flow Forecasting Using GPS Track Data of Cabs:Take the Jiefangbei Neighborhood of Chongqing City as Example

作　　者：汪孝之 牟凤云 张用川[2] 王俊秀 WANG Xiao-zhi;MU Feng-yun;ZHANG Yong-chuan;WANG Jun-xiu(Key Laboratory of Urban Land Resources Monitoring and Simulation Ministry of Land and Rescources of China,Shenzhen 518038,China;Smart City College,Chongqing Jiaotong University,Chongqing 400074,China)

机构地区：[1]自然资源部城市国土资源监测与仿真重点实验室,深圳518038 [2]重庆交通大学智慧城市学院,重庆400074

出　　处：《科学技术与工程》2023年第28期12265-12274,共10页Science Technology and Engineering

基　　金：自然资源部城市国土资源监测与仿真重点实验室开放基金(KF-2021-06-102);研究生科研创新项目(2023S0130)。

摘　　要：交通流量预测是智能交通系统的重要组成部分。以重庆市解放碑街区为研究区域进行交通流量预测分析,基于研究区域内出租车GPS轨迹数据处理获取时间间隔为5、10、15 min的交通流量序列。同时为充分挖掘交通流量序列特征规律,减小序列非线性、非平稳性带来的影响,提出一种基于信号分解的预测模型LE-RL(linear regression model-empirical mode decomposition-random forest-long short-term memory network)。通过一般线性模型(linear regression model,LR)将原始序列分解成周期序列、趋势序列和残差,同时引入经验模态分解(empirical mode decomposition,EMD)方法对残差进一步分解以充分挖掘序列特征;模型预测方面,构建随机森林(random forest,RF)模型对周期序列和趋势序列进行预测,接着引入长短期记忆网络模型(long short-term memory network,LSTM)构建RF-LSTM残差模型对EMD分解的各分量进行预测,通过叠加各模型预测成果得到最终预测结果;为验证模型精度,设置对照模型进行比对。结果表明:所构建的LE-RL模型在预测精度上均高于对照模型,可以满足基于不同样本时间间隔的交通流量预测的需要。Traffic flow prediction is an important part of intelligent transportation system.Traffic flow prediction analysis was carried out in the Jiefangbei neighborhood of Chongqing,and traffic flow sequences with time intervals of 5,10,15 min were obtained based on the processing of GPS trajectory data of cabs in the study area.At the same time,in order to fully explore the characteristic law of traffic flow series and reduce the impact of series nonlinearity and non-smoothness,a prediction model was proposed based on signal decomposition linear regression model-empirical mode decomposition-random forest-long short-term memory network(LE-RL).The original series was decomposed into periodic series,trend series and residuals through the linear regression model(LR),and the empirical mode decomposition(EMD)method was introduced to further decompose the residuals to fully mine the characteristics of the series.In terms of model prediction,the random forest(RF)model was constructed to predict the periodic series and trend series,and then the long short-term memory network(LSTM)was introduced to construct the RF-LSTM residual model to predict the components of EMD decomposition,and the final prediction results were obtained by adding the prediction results of each model.A control model was set to verify the accuracy of the model.The results show that the constructed LE-RL model is higher than the control model in terms of prediction accuracy and can meet the needs of traffic flow prediction based on different sample time intervals.

关 键 词：交通流量预测 时间序列分解 长短期记忆网络(LSTM) 随机森林(RF) 机器学习 

分 类 号：U491.1[交通运输工程—交通运输规划与管理]