Modeling Usage Frequencies and Vehicle Preferences in a Large-Scale Electric Vehicle Sharing System

Carsharing systems deploying electric vehicles (EVs) have grown significantly over recent years. This study aims to employ data-driven approaches to understand users’ decision-making in terms of usage frequencies and vehicle preferences. Massive data of over five million transactions from EVCARD – the biggest EV sharing company in Shanghai is collected and analyzed. Given the best predictive performance, gradient boosting decision trees (GBDT) are selected to depict partial dependence plots (PDP), which can visualize the relationships between predictors and users’ behaviors. Key findings are listed as follows:
(1)
a positive relationship between the number of coupons (the deduction users can directly use when paying the rental fee) and monthly usage frequency is observed with decreasing marginal effect;
(2)
users whose most-visited stations are located in areas with poor transit access would have higher usage frequencies;
(3)
state of charge (SoC) significantly affect users’ preferences on vehicles and users tend to be ‘greedy’ on the SoC; and
(4)
users are less likely to choose vehicles with greater ages, fewer seats, higher rental prices, and smaller battery ranges. Findings provide useful insights into the operation and management of carsharing systems. For example, stations can be located in areas with low accessibility of public transit to attract high-frequency users. Additionally, economic incentives can be provided for users of short-distance trips to encourage the use of low-SoC EVs.