Data-Driven Vehicle-Cargo Matching Problem Study

Logistics information platform as a driver and cargo owners to achieve information sharing infrastructure,matching drivers and cargo owners is the integration of logistics resources on the road freight market,to improve the logistics and transport efficiency of the foundation to ensure.The logistics information platform can reduce the empty load caused by the confusion of vehicle and cargo information through vehicle and cargo matching.At the same time,the current integration of Internet technology and the retail industry has further deepened,resulting in a series of new business models,such as: new retail,fresh food e-commerce,etc.,which has increased the sensitivity of consumers to the delivery time of goods.Many e-commerce platforms are cooperating with logistics information platforms in order to minimise delivery times.Therefore,how to efficiently match and accurately predict the complicated vehicle and cargo information is of great practical significance to the development of matching strategies of logistics information platforms and to improve the efficiency of logistics resources utilization.At present,scholars’ research on vehicle and cargo matching is mainly summarised as a single bilateral matching or path planning problem,with little consideration given to the existence of different ways of vehicle and cargo matching in different business scenarios.From a data-driven perspective,this paper proposes vehicle-cargo matching models based on offline and online conditions for different platform businesses such as long-distance freight and same-city freight,respectively,in order to provide a more comprehensive matching decision reference for realistic logistics information platforms and to extend the existing vehicle-cargo matching theory.Specifically,the research in this paper is concerned with the following aspects.Firstly,based on offline conditions,the platform carries out vehicle-cargo matching based on orders issued by cargo owners in advance for scheduled deliveries.In this paper,the XGBoost+Easy Ensemble vehicle-cargo matching model is proposed considering the potential association of drivers’ historical behaviour data with drivers,cargo owners and cargo orders,as well as the situation that drivers are only interested in a small number of cargo orders.In addition,this paper analyses different attribute features and designs comparison experiments to verify the effectiveness and scientific validity of the model.Commonly used classification models such as Logistic Regression,Parsimonious Bayesian Model,Random Forest and GBDT are used as reference benchmarks to compare with the XGBoost+Easy Ensemble vehicle and cargo matching model.At the same time,experiments were conducted on a real driver history dataset,and XGBoost+Easy Ensemble achieved better prediction results compared to the benchmark algorithm.Second,based on the online condition,the platform performs vehicle-cargo matching based on the orders issued by shippers for immediate delivery.In this paper,the vehicle-goods matching problem is modelled based on the online matching condition considering the existence of driver attrition,combined with two-part graph theory.The optimization objective is to maximize the overall order turnover of the platform by maximizing the future matching capacity of the drivers.In this paper,the online car and cargo matching problem is transformed into a Beta OM application scenario and solved.Experiments are then conducted to verify the feasibility of applying the Beta OM algorithm to the online vehicle and cargo matching problem using simulation data with Greedy as a comparison.