基于田口集成学习的最佳交通流量预测模型  

Optimized traffic flow forecasting model with Taguchi ensemble learning approach

作　　者：祁欣学 覃锡忠[1] 贾振红[1] 常春 樊树铭 QI Xin-xue;QIN Xi-zhong;JIA Zhen-hong;CHANG Chun;FAN Shu-ming(College of Information Science and Engineering,Xinjiang University,Urumqi 830046,China;Planning Department,China Mobile Communications Group Xinjiang Limited,Urumqi 830063,China)

机构地区：[1]新疆大学信息科学与工程学院,乌鲁木齐830046 [2]中国移动通信集团新疆有限公司计划部,乌鲁木齐830063

出　　处：《计算机工程与设计》2019年第12期3451-3456,共6页Computer Engineering and Design

基　　金：新疆维吾尔自治区自然科学基金项目(2018D01C047)

摘　　要：针对预测准确性对数据相关性和网络结构极为敏感的问题,采用田口方法选择集成学习的结构参数,在保证学习多样性的同时,提高选择过程的效率,保证预测的准确性。实验仿真计算结果表明,在多个路段的交通流量数据集上,该方法皆取得了满意的预测效果,模型具有较好的鲁棒性。将仿真结果与其它算法进行比较,比较结果表明,相比传统ARIMA、MLP模型,其预测精度分别提高了4.51%、11.23%。To solve the problem that accuracy is extremely sensitive to data correlation and network structure,the forecasting model using the Taguchi method to select the ensemble learning structure parameters was designed.While ensuring the diversity of learning,the efficiency of the selection process was improved,and the accuracy of the forecasting was also guaranteed.Experimental results show that the proposed method can get satisfactory forecasting on the traffic data sets of multiple road segments,and the model has good robustness.Comparing the simulation results with the traditional ARIMA and MLP models,it is verified that the accuracy is improved by 4.51% and 11.23%respectively.

关 键 词：数据相关长度 网络结构 田口方法 集成学习 鲁棒性 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]