车货供需匹配模型与算法研究综述

Survey of vehicle-cargo matching models and algorithms

随着货车保有量及货运需求的迅速增长,车货供需匹配问题成为了货运电子商务平台的核心和热点。本文主要针对车货供需匹配问题从模型和算法两个方面对现有文献进行梳理和总结。在车货供需匹配模型方面,考虑的优化目标主要包括满意度、公平性和稳定性三个方面的评价指标,根据应用场景,将车货供需匹配模型分为一对一、一对多和多对多三类,其中一对一车货供需匹配模型针对整车运输,其他两种则对应零担运输。随着三种应用场景的模型复杂程度越来越高,相应求解难度、求解时间也呈现递增趋势。在车货供需匹配算法方面,根据货运需求数据的结构与特点可以划分为最优化算法、人工智能算法、推荐算法以及其他算法四类:对于小规模、时效性要求不高的货运需求,可以根据模型的特点与特性设计最优化求解算法;对于大数据、交互性数据或实时性要求高的货运需求,人工智能算法和推荐算法则是车货供需匹配问题的有效途径,其中人工智能算法通过预测车主行为或匹配结果实现匹配任务,而推荐算法可以针对车货需求大数据实现有效的召回并推荐。最后,本文总结了现有研究的不足之处,并从中归纳出三个值得进一步研究的方向:一是结合实际业务场景和车货信息大数据背景,提高车货供需匹配方法的实际可行性;二是进一步挖掘更多的评价指标,如车主偏好以及订单目的地接单概率等指标;三是关注车货供需匹配的实时决策问题,重点考虑货运需求的动态随机性以及平台和车主的长期收益目标。

With the rapid increase in truck ownership and freight demand,vehicle-cargo matching problem has become the core and hotspot of freight e-commerce platforms.In this study,we mainly focused on the vehicle-cargo matching problem to sort out and summarize the existing literature from both model and algorithm aspects.In terms of the vehicle-cargo matching model,this study is based on the bilateral matching theory,which is designed in accordance with the satisfaction,fairness,and stability of the three aspects of the evaluation index.Based on the application scenarios,vehicle-cargo matching models can be divided into three categories:one-to-one vehicle-cargo matching model,one-to-many vehicle-cargo matching model,and many-to-many vehicle-cargo matching model.Here,the one-to-one vehicle-cargo matching model is for the entire truck transportation,and the other two correspond to the less-than-truckload(LTL)transportation.The complexity of the three application scenarios is increasing,and the corresponding difficulty and time taken to solve the problems also show an increasing trend.In terms of vehicle-cargo matching algorithms,the data can be divided into four categories based on the structure and characteristics of freight demand:optimization algorithms,artificial intelligence algorithms,recommendation algorithms,and other algorithms.For small-scale freight transportation needs that do not require high timeliness,optimization algorithms are often used,and the corresponding solution process and architecture are designed according to the characteristics and features of the model.For big data,interactive data or real-time matching requirements of freight transportation needs,artificial intelligence algorithms,and recommendation algorithms are effective methods for the vehicle-cargo matching problem,in which artificial intelligence algorithms can be classified according to the owner’s behavior or matching results,and recommendation algorithms can effectively recall and recommend the big data for vehicle and cargo information.In the end,the shortcomings of previous studies are summarized,and three directions for further research are highlighted.First,the practical feasibility of the method for trucks and cargo matching should be improved by combining the actual business scenarios and the background of big data of truck and cargo information.The second is the further investigation of evaluation indexes,such as the preference of vehicle owners,and the probability of receiving orders at the order destination.The third is the real-time decision making for vehicle and cargo matching,with a focus on the dynamic randomness of freight demand and the goal of long-term benefits for platforms and owners.