集中与分散决策模式下的随机动态双边匹配策略研究

With the development of the mobile Internet technology, the matching decision between supply and demand is the core decision for many Internet platform enterprises (e.g., Didi, cloud computing platform) to create profits and enhance their competitiveness. This project studies the problem of bilateral matching in the random dynamic arrival environment between multi-type supply agents and multi-type demand agents. In this project, we first use the dynamic programming method to model the stochastic dynamic matching problem under the centralized decision mode, and get the optimal dynamic matching policy, including the optimal matching priorities and the quantities of matches. Secondly, aiming at the problem of random dynamic matching in decentralized decision making mode, the queueing matching models under different waiting mechanisms and information scenarios is established. We use game theory to obtain the participant matching decision equilibrium in decentralized decision making mode. We find the key parameters to improve the matching profit, and design the profit distribution mechanism, information disclosure mechanism, charging or subsidy mechanism that can match the optimal matching strategy. Finally, we simplify the calculation of value function by using the means of supply and demand instead of the random variables, and give a fast and efficient online heuristic algorithm for dynamic matching. The research of this project can not only obtain theoretical innovation, but also guide the practice of matching decision of existing enterprises.

随着移动互联网技术的发展，供给与需求的匹配决策是很多互联网匹配平台企业（如，滴滴、云计算平台）创造收益与提升竞争力的核心决策。本项目研究多类型供给参与者与多类型需求参与者随机动态到达环境下的双边匹配决策问题。本项目首先使用动态规划方法，对集中决策模式下的多阶段随机动态匹配问题进行建模，获得最优动态匹配策略，包括最优匹配顺序与最优匹配数量。其次，针对分散决策模式下的随机动态匹配问题，建立不同服务规则、信息情景等情况下的排队匹配模型，用博弈论的方法，获得分散决策模式下的参与者匹配决策均衡。找到提升匹配收益的关键参数，设计能达到与最优匹配策略下收益协同的收益分配机制、信息披露机制、收费或补贴机制。最后，采用将供给与需求随机变量用其均值近似代替的办法，简化对价值函数的计算，给出快速高效的动态匹配在线启发式算法。本项目的研究既能获得理论上的创新，又能指导现有企业匹配决策的实践。

本项目研究多类型供给参与者与多类型需求参与者随机动态到达环境下的双边匹配决策问题。本项目首先分析两种类型供给参与者和两种类型需求参与者随机动态到达的环境下，获得集中决策模式下的多阶段最优动态匹配策略，包括最优匹配顺序与最优匹配数量。其次，针对分散决策模式下的随机动态匹配问题，建立FCFS的服务规则下，分散决策模式下的参与者匹配决策均衡。同时找到提升分散决策模式下匹配收益的关键参数，设计能达到与最优匹配策略下收益协同的收益分配机制、信息披露机制、收费或补贴机制。最后，采用将供给与需求随机变量用其均值近似代替的办法，简化对价值函数的计算，给出快速高效的动态匹配在线启发式算法。本项目的研究既能获得理论上的创新，又能指导现有企业匹配决策的实践。在研究成果的基础上，已在国内外学术刊物上发表期刊论文15篇，其中：国际重要刊物12篇国内核心刊物3篇；另有一些论文正在审稿中，还有一批论文即将投稿。