报告题目:共享出行供需匹配系统的优化
Optimization for on-demand matching process
in ride-sourcing systems
演讲人: 柯锦涛，香港理工大学研究助理教授
时间:2021年7月5日10:00-12:00
地点: 浙江大学紫金港校区安中大楼A322
讲座摘要:
作为共享出行的代表，以滴滴、优步为代表网约车逐步成为人们日常出行的一种重要的交通模式。尽管网约车在商业上取得了巨大的成功，它的出现也给城市管理者带来了新问题和新挑战，其中一个重要的问题是如何高效地匹配实时订单和空闲司机。本研究提出了一个优化模型，可以根据不同的供需情况，动态地调整实时供需匹配中的两个重要决策变量，即匹配时间窗口和匹配半径，从而更好地匹配实时供需。此外，本研究提出了一个在线-离线学习框架，将强化学习与在线组合优化技术相结合，进一步提高匹配效率。实验表明，传统的组合优化模型往往是短视的，只能最优化当前时刻的系统效率，而这种决策往往不能达到长时间尺度上（如一整天）的最优效果。通过考虑到当前的决策对未来供需状态的影响，本研究提出的强化学习算法可以计算每个匹配决策的长期价值，从而最大化长期的系统效率。
As a symbolic icon for shared mobility in recent years, ride-sourcing service, provided by digital platforms like DiDi, Uber and Lyft, has been playing an increasingly important role in meeting mobility needs by efficiently connecting passengers and dedicated drivers through online platforms. Despite its great success in business, ride-sourcing service has also aroused many challenging issues in operations, management, and regulations, from the perspective of different stakeholders, including both private sectors (ride-sourcing platforms) and public sectors (governments). One critical issue in platform operations for ride-sourcing service is the optimizations for on-demand matching process between idle drivers (supply) and waiting passengers (demand). In particular, this talk will present a novel model to jointly optimize two key decision variables in on-demand ride matching process⸺matching time interval and matching radius⸺under different supply and demand levels. The model will enable ride-sourcing platforms to dynamically adjust their matching time interval and matching radius in response to changing supply and demand in a dynamic system. Additionally, this talk will discuss how to integrate advances of reinforcement learning approaches into traditional optimization models to further improve the efficiency of the driver-passenger matching procedure. It is shown that pure optimization models are generally myopic and only focus on immediate feedbacks during the current time interval without considering future system rewards. In contrast, the combination of reinforcement learning and optimization can be more far-sighted by considering the interactions between platform operations and system dynamics, and thus achieve better performance over a long horizon.
演讲嘉宾简介:
柯锦涛，博士，香港理工大学研究助理教授。分别于浙江大学和香港科技大学获得学士和博士学位。致力于共享出行系统运营与监管、大数据驱动的城市多模式交通系统的管理与优化、基于机器学习的短时交通需求量预测、交通定价与博弈等方面的研究。在Transportation Research Part A/B/C/E, IEEE Transactions on Knowledge and Data Engineering, IEEE Transactions on Intelligence Transportation System等期刊发表SCI/SSCI期刊论文20篇，论文总引用数超过1000次。现担任国际期刊Transportation Research Part C青年编委。
Dr. Jintao Ke is a Research Assistant Professor in Hong Kong Polytechnic University. He received the B.E. and Ph.D. degrees from the Department of Civil Engineering, Zhejiang University, and the Department of Civil and Environmental Engineering, Hong Kong University of Science and Technology (supervisor: Prof. Hai Yang), in 2016 and 2020, respectively. His research interests include shared mobility on demand, transportation big data analytics, multimodal intelligent transportation systems, transportation pricing, short-term travel demand forecasting, etc. The vision of his research is to develop novel models, algorithms, and conduct data-driven quantitative analyses to better manage, operate, and regulate shared mobility and other emerging mobility services. He has published around 20 SCI/SSCI indexed research papers in top-tier journals in the field of transportation research and data mining, such as Transportation Research Part A/B/C/E, IEEE Transactions on Intelligence Transportation System, IEEE Transactions on Knowledge and Data Engineering. He is in the Early Career Editorial Advisory Board (EAB) of Transportation Research Part C.