Assessing On-Demand Mobility Services and Policy Impacts: A Case Study from Chengdu, China

TL;DR

The paper develops a graph-theory-based, rolling-window ride-hailing simulation framework fed by real Chengdu taxi data to compare ride-hailing with street-hailing and to evaluate supply- and demand-side policies. By formulating batch trip-vehicle matching to minimize pickup delay and solving it with the Jonker–Volgenant algorithm, the authors quantify performance across $APWT$, $ADM$, and $ADEC$. They report that ride-hailing substantially outperforms street-hailing under identical conditions, with the largest gains during midnight and near airports, and they show that policy effects include diminishing returns from fleet expansion, generally worse overall performance from geofencing, and potential improvements from spatially targeted demand management. The study provides a practical AMoD benchmark on a real urban network and offers policy guidance for spatial coordination and demand modulation to enhance efficiency and passenger experience.

Abstract

The rapid expansion of ride-hailing services has significantly reshaped urban on-demand mobility patterns, but it still remains unclear how they perform relative to traditional street-hailing services and how effective are related policy interventions. This study presents a simulation framework integrating a graph theory-based trip-vehicle matching mechanism with real cruising taxi operations data to simulate ride-hailing services in Chengdu, China. The performances of the two on-demand mobility service modes (i.e., ride-hailing and street-hailing) are evaluated in terms of three key performance indicators: average passenger waiting time (APWT), average deadheading miles (ADM), and average deadheading energy consumption (ADEC). We further examine the impacts of spatiotemporal characteristics and three types of policies: fleet size management, geofencing, and demand management, on the performance of ride-hailing services. Results show that under the same fleet size and trip demand as street-hailing taxis, ride-hailing services without cruising achieve substantial improvements, reducing APWT, ADM, and ADEC by 81\%, 75\%, and 72.1\%, respectively. These improvements are most pronounced during midnight low-demand hours and in remote areas such as airports. Our analysis also reveals that for ride-hailing service, (1) expanding fleet size yields diminishing marginal benefits; (2) geofencing worsens overall performance while it improves the performance of serving all trips within the city center; and (3) demand-side management targeting trips to high-attraction and low-demand areas can effectively reduce passenger waiting time without increasing deadheading costs.

Figures (19)

• Figure 1: Process of the ride-hailing simulation framework.
• Figure 2: Illustration of the batch matching process. Trips and vehicles are collected within a period. The feasible pairs make up a bipartite graph representing an assignment problem.
• Figure 3: Service rate and average passenger queuing time of ride-hailing services influenced by $t_{\text{queue}}^{\max}$ and $t_{\text{pu}}^{\max}$
• Figure 4: Changes of ride-hailing performances over the maximum pick-up time $t^{\max}_{pu}$. Average passenger waiting time (APWT), average deadheading mileage (ADM), and average deadheading energy consumption (ADEC) all decrease as $t^{\max}_{pu}$ increases, while these improvements gradually diminish. The light-colored bands represent the range of results from all 30 random experiments. Each data point represents the mean value derived from the corresponding experiments, with vertical error bars indicating the 95% confidence intervals.
• Figure 5: Comparison between ride-hailing (at $t^{\max}_{pu} \in [8,20]$ minutes) and street-hailing services. The performances of ride-hailing service with $t^{\max}_{pu} \in[8,20]$ have no noticeable difference.