Shan Liu , Ya Zhang , Zhengli Wang , Xiang Liu , Hai Yang
{"title":"利用主动对抗逆向强化学习和变形器估算个性化的出发地-目的地旅行时间","authors":"Shan Liu , Ya Zhang , Zhengli Wang , Xiang Liu , Hai Yang","doi":"10.1016/j.tre.2024.103839","DOIUrl":null,"url":null,"abstract":"<div><div>Travel time estimation is important for instant delivery, vehicle routing, and ride-hailing. Most studies estimate the travel time of specified routes, and only a few studies pay attention to origin–destination travel time estimation (OD-TTE) without a specified route. Moreover, most of these studies on OD-TTE ignore the personalized route preference and the cost of data annotation. To fill this research gap, we analyze the individual route preference and propose a personalized origin–destination travel time estimation method based on active adversarial inverse reinforcement learning (AA-IRL) and Transformer. To analyze the personalized route preference, we integrate adversarial inverse reinforcement learning with active learning, which effectively reduces the cost of sample annotation. After inferring the possible routes, we propose AdaBoost multi-fusion graph convolutional Transformer network (AMGC-Transformer) for travel time estimation. Numerical experiments conducted on ride-hailing and online food delivery trajectories in China validate the advantage of our method. Compared to relevant studies, our approach can improve F1-score of route inference by 2.50–3.35% and reduce the mean absolute error of OD-TTE by 7.44–11.66%.</div></div>","PeriodicalId":49418,"journal":{"name":"Transportation Research Part E-Logistics and Transportation Review","volume":"193 ","pages":"Article 103839"},"PeriodicalIF":8.3000,"publicationDate":"2024-11-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Personalized origin–destination travel time estimation with active adversarial inverse reinforcement learning and Transformer\",\"authors\":\"Shan Liu , Ya Zhang , Zhengli Wang , Xiang Liu , Hai Yang\",\"doi\":\"10.1016/j.tre.2024.103839\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>Travel time estimation is important for instant delivery, vehicle routing, and ride-hailing. Most studies estimate the travel time of specified routes, and only a few studies pay attention to origin–destination travel time estimation (OD-TTE) without a specified route. Moreover, most of these studies on OD-TTE ignore the personalized route preference and the cost of data annotation. To fill this research gap, we analyze the individual route preference and propose a personalized origin–destination travel time estimation method based on active adversarial inverse reinforcement learning (AA-IRL) and Transformer. To analyze the personalized route preference, we integrate adversarial inverse reinforcement learning with active learning, which effectively reduces the cost of sample annotation. After inferring the possible routes, we propose AdaBoost multi-fusion graph convolutional Transformer network (AMGC-Transformer) for travel time estimation. Numerical experiments conducted on ride-hailing and online food delivery trajectories in China validate the advantage of our method. Compared to relevant studies, our approach can improve F1-score of route inference by 2.50–3.35% and reduce the mean absolute error of OD-TTE by 7.44–11.66%.</div></div>\",\"PeriodicalId\":49418,\"journal\":{\"name\":\"Transportation Research Part E-Logistics and Transportation Review\",\"volume\":\"193 \",\"pages\":\"Article 103839\"},\"PeriodicalIF\":8.3000,\"publicationDate\":\"2024-11-02\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Transportation Research Part E-Logistics and Transportation Review\",\"FirstCategoryId\":\"5\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S1366554524004307\",\"RegionNum\":1,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"ECONOMICS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Transportation Research Part E-Logistics and Transportation Review","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1366554524004307","RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"ECONOMICS","Score":null,"Total":0}
Personalized origin–destination travel time estimation with active adversarial inverse reinforcement learning and Transformer
Travel time estimation is important for instant delivery, vehicle routing, and ride-hailing. Most studies estimate the travel time of specified routes, and only a few studies pay attention to origin–destination travel time estimation (OD-TTE) without a specified route. Moreover, most of these studies on OD-TTE ignore the personalized route preference and the cost of data annotation. To fill this research gap, we analyze the individual route preference and propose a personalized origin–destination travel time estimation method based on active adversarial inverse reinforcement learning (AA-IRL) and Transformer. To analyze the personalized route preference, we integrate adversarial inverse reinforcement learning with active learning, which effectively reduces the cost of sample annotation. After inferring the possible routes, we propose AdaBoost multi-fusion graph convolutional Transformer network (AMGC-Transformer) for travel time estimation. Numerical experiments conducted on ride-hailing and online food delivery trajectories in China validate the advantage of our method. Compared to relevant studies, our approach can improve F1-score of route inference by 2.50–3.35% and reduce the mean absolute error of OD-TTE by 7.44–11.66%.
期刊介绍:
Transportation Research Part E: Logistics and Transportation Review is a reputable journal that publishes high-quality articles covering a wide range of topics in the field of logistics and transportation research. The journal welcomes submissions on various subjects, including transport economics, transport infrastructure and investment appraisal, evaluation of public policies related to transportation, empirical and analytical studies of logistics management practices and performance, logistics and operations models, and logistics and supply chain management.
Part E aims to provide informative and well-researched articles that contribute to the understanding and advancement of the field. The content of the journal is complementary to other prestigious journals in transportation research, such as Transportation Research Part A: Policy and Practice, Part B: Methodological, Part C: Emerging Technologies, Part D: Transport and Environment, and Part F: Traffic Psychology and Behaviour. Together, these journals form a comprehensive and cohesive reference for current research in transportation science.