讲座题目 | Optimal Order Assignment Policy for a Hybrid On-demand Delivery Platform with heterogenous Customers | ||
主讲人 （单位） | 王明征 (浙江大学) | 主持人 （单位） | 何勇 （东南大学） |
讲座时间 | 2024年12月19日 10:00 | 讲座地点 | 经管楼B201会议室 |
主讲人简介 |
个人简介：王明征，现为浙江大学管理学院教授。兼任浙江大学数据科学与管理工程学系副系主任、中国系统工程学会物流系统工程专业委员会副理事长、中国现代化研究会电子商务与网络空间管理分会副理事长、中国双法研究会网络科学分会副理事长。主要的研究兴趣是新零售运营与供应链管理、平台数智化运营管理、数据驱动决策、数据经济学、大数据分析与统计优化、在线学习优化算法、深度强化学习算法、智能物流优化模型与算法、商业分析与决策智能。取得的研究成果已经部分发表在Operations Research、Manufacturing and Service Operations Management、INFORMS Journal on Computing、IEEE Transactions on Automatics Control、EJOR、IJPE、TRE、IEEE Transactions SMC-S等国内外著名期刊上70余篇，同时多篇研究成果在国际顶级学术会议INFORMS- DS、ICIS和POMS ICC等上获得最佳论文奖。近几年来作为项目负责人主持1项国家自然科学基金委重点项目和1项国家自然科学基金委重点国际合作项目；作为骨干成员参加国家科技重大专项、国家自然科学基金委科研创新群体、国家自然科学基金委重点项目、国际合作交流项目重大项目和国家高科技研究计划（863项目）等十多项。目前主持国家自然科学基金委重点项目《新零售模式的运营管理理论与方法》和《即时零售平台的数智化运营协同管理方法》，面向国家和企业的重大需求，与京东物流、美团、阿里巴巴盒马鲜生以及饿了么开展产学研合作，结合企业的业务和数据，利用数据和模型双轮驱动决策优化方法，解决企业运营管理难题。 | ||
讲座内容摘要 | 报告摘要： We study an order assignment problem with heterogeneous orders in an on-demand delivery platform. The heterogeneity of orders and drivers, along with the uncertainty of demand and available supply, poses a challenge for the platform in optimally assigning orders to hybrid drivers or rejecting them. We develop an infinite horizon discrete-time event-based Markov Decision Process (MDP) to maximize the expected total discounted revenue of the platform. We prove this MDP has favorable properties. Leveraging these properties, we propose an optimal order assignment policy using rationing and priority thresholds which are dependent on the number of busy drivers. We also extend our policy to settings that consider order cancellation behaviors and endogenous delivery fees to increase the welfare of consumers. Numerical studies show significant improvements of our policy using synthetic and real-world data. Orders with high delivery fees and large rejection penalties should always be accepted. However, orders with low delivery fees and small rejection penalties could be rejected under certain circumstances. Although crowdsourced drivers deliver more slowly, they could still be an optimal assignment for both types of orders. Moreover, the substitution between full-time and crowdsourced drivers always benefits the platform. | ||
