Coordinating Guidance, Matching, and Charging Station Selection for Electric Vehicle Ride-Hailing Services through Data-Driven Stochastic Optimization

TL;DR

The paper tackles the inefficiencies in EV ride-hailing caused by rider demand uncertainty and CS selection by proposing a data-driven stochastic framework with two interconnected models: (i) proactive idle EV guidance under probabilistic demand forecasts and (ii) batched EV matching that accounts for CS waiting times. By integrating forecasting with a Monte Carlo-based sample average approximation and a CS-aware matching objective, the framework jointly optimizes idle relocation and downstream matching, reducing rider waiting times and charging delays. Empirical results from the NYC case study show notable improvements in matching rate and waiting times, with substantial reductions in charging waiting times for EVs with low state of charge and a tunable trade-off between rider experience and charging efficiency. The approach offers practical benefits for ride-hailing platforms and provides a foundation for further research into driver charging decisions and charging-station surge pricing under uncertainty.

Abstract

Electric vehicles (EVs) play a pivotal role in sustainable ride-hailing services primarily due to their potential in reducing carbon emissions and enhancing environmental protection. Despite their significance, current research in the realm of EV batched matching frequently overlooks critical aspects such as rider demand uncertainty and charging station (CS) selection, leading to inefficiencies like decreased matching rates and prolonged waiting times for both riders and EV drivers. To fill the research gap, we propose a data-driven optimization framework that incorporates two inter-connected stochastic optimization models to address the challenges. The first model aims to relocate the idle EVs under satisfied conditions to the designated regions based on the probabilistic rider demand forecasting result before the real rider demand is revealed. Taking the solutions of the first model as the input, the second model optimizes the batched matching results by minimizing the rider's average waiting time and EV charging waiting time at CS. This integrated framework not only elevates the matching rate through the incorporation of rider demand uncertainties in the guidance module but also substantially curtails both rider and EV charging waiting times by synergizing guidance with CS selection choices. Empirical validation of our framework was conducted through an extensive case study in New York City, utilizing real-world data sets. The validation results demonstrate that the proposed data-driven optimization framework outperforms the benchmark models in terms of the proposed evaluation metrics. Most importantly, when deploying our framework, the charging waiting time of the EVs with low SOC can be reduced up to 73.6% compared to the benchmark model without CS selection.

Figures (13)

• Figure 1: The diagram of the proposed approach for EV ride-hailing system
• Figure 2: An example of CS selection during the batched matching optimization, EV with low SoC has the priority to match the rider whose destination deploys CS with short expected waiting time
• Figure 3: Heatmap of Rider Pick-Up Density
• Figure 4: Heatmap of Rider Drop-Off Density
• Figure 5: Charging Station Distribution