基于多任务学习的多线路公交车程时间预测  

Multiple bus lines trip time prediction based on multi task learning

作　　者：谢文亮 沈吟东[1] XIE Wenliang;SHEN Yindong(Artificial Intelligence Research Institute,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]华中科技大学人工智能研究院,湖北武汉430074

出　　处：《华中科技大学学报（自然科学版）》2022年第11期90-95,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(71571076);国家重点研发计划资助项目(2018YFF030301,201YFF0300300)。

摘　　要：针对现有公交车程时间预测方法不适用于公交有限数据集,且难以对区域内多条公交线路进行预测的不足,研究设计了一种基于多任务学习的多线路公交车程时间预测模型.首先,对多线路公交车程时间预测问题进行分析,并根据问题的特点建立多任务学习预测模型,实现同时对多条公交线路的车程时间的预测.然后,采用湖北省X市五条不同公交线路的车程时间数据集进行预测实验.实验结果表明:所设计的多任务学习预测模型能够有效实现对区域内多条公交线路的车程时间的预测,且建模资源消耗少于对每条公交线路单独建立预测模型的总和,对于每条公交线路车程时间的预测精度也优于对每条线路建立的单任务预测模型.To solve the shortcomings that the existing bus trip time prediction methods are not applicable for bus limited dataset and multiple bus lines in the region,a multiple bus lines trip time prediction model was studied and designed based on multi task learning. First,the multiple bus lines trip time prediction problem was discussed,and a multi task learning prediction model was established according to the characteristics of this problem,for predicting the multiple bus lines trip time at the same time.Then,the prediction experiment was carried out based on the trip time datasets of five different bus lines in X city,Hubei Province.Experimental results show that the designed multi task learning prediction model can forecast the trip time of multiple bus lines in the region effectively,and the modeling resource consumption of this model is less than the sum of that of the prediction models established separately for each bus line. Meanwhile,this model performs better trip time forecasting accuracy than the single task prediction model for each bus line.

关 键 词：区域调度 多条公交线路 多任务学习 公交车程时间 预测模型 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]