Modeling Urban Taxi Services with e-hailings: A Queueing Network Approach

Wenbo Zhang, Harsha Honnappa, Satish V. Ukkusuri

This paper, titled "Modeling Urban Taxi Services with e-hailings: A Queueing Network Approach," presents a sophisticated mathematical framework to understand the modern, competitive landscape of urban mobility. By combining queueing theory with real-world data from New York City, the authors bridge the gap between "virtual" matching platforms and "physical" road congestion.

TL;DR

The authors develop a novel queueing network model that simultaneously captures the competition between traditional street-hailing taxis and e-hailing services (like Uber) while accounting for urban road congestion. Using New York City data, they demonstrate that this stochastic approach can accurately estimate passenger wait times and vehicle delays at a macroscopic scale.

What Problem Does This Paper Address?

The urban taxi market has undergone a seismic shift. We no longer have a single, centralized taxi authority; instead, we have a "two-sided market" where traditional taxi services (TTS) compete with app-based taxi services (ATS).

Existing models often struggle with three main challenges:

  1. Spatial Heterogeneity: Demand is high in downtown cores but sparse in suburbs.
  2. Network Externalities: A pickup in one zone reduces vehicle supply there but increases it in a destination zone later.
  3. Stochasticity: Human behavior—when a passenger requests a ride or when a driver goes offline—is inherently random.

Most previous research focused on either equilibrium models (which assume a steady state that rarely exists in reality) or agent-based simulations (which are computationally expensive). This paper seeks a middle ground: a queueing network approach that is mathematically rigorous yet computationally efficient enough to model an entire city.

Methods

The authors represent the city as a network of spatial units (e.g., Community Districts). Each unit contains two distinct types of queues:

  • The Synchronized Queue ($SM/M/1$): This is the paper's most novel structural contribution. Unlike a standard queue where customers wait for a server, a taxi service requires two independent "arrivals"—a passenger and a vehicle—to be matched. The authors use an "assembly-like" queue to model this synchronization process.
  • The Road Transport Queue ($M/M/c$): Once matched, the passenger-vehicle pair enters the road system. This is modeled as a multi-server queue where the number of "servers" ($c$) represents the road capacity, derived from the Macroscopic Fundamental Diagram (MFD)—a concept relating traffic density to flow.

The authors prove that this system reaches a stationary state distribution, allowing them to derive performance metrics like average wait times and system-wide delays. They also use Linear Programming to solve for the "effective arrival rates" of vehicles across the network, accounting for empty cruising and rebalancing.

Datasets

The model was validated using an extensive multi-source dataset from New York City (April 2017):

  • NYC Taxi & Limousine Commission (TLC): Trip records for Yellow Cabs (TTS).
  • Uber Trajectory Data: High-frequency GPS traces of empty and occupied Uber vehicles.
  • NYC DOT Traffic Data: Link volumes and speeds for the road network.
  • Google Maps Directions API: Used to estimate shortest-path travel times and segment taxi searches.

Key Insights and Ideas

  • The Poisson Justification: One of the most significant "aha moments" is the authors' extensive statistical testing. They found that at a specific spatiotemporal scale—Community Districts with 1-minute intervals—passenger and vehicle arrivals actually follow a Poisson distribution. This provides the empirical "green light" to use Markovian queueing models.
  • Holistic Feedback Loops: The model captures the "feedback" between the taxi market and the road system. If roads are congested ($M/M/c$ queue slows down), vehicles take longer to reach passengers, which in turn degrades the matching efficiency ($SM/M/1$ queue).
  • ATS vs. TTS Dynamics: The model highlights that ATS (e-hailing) is generally more predictable and easier to model at scale because the centralized platform reduces some of the "random walk" searching behavior seen in traditional street-hailing.

Key Results

  • High Accuracy in "Hot" Zones: The model is exceptionally accurate in high-demand areas like Manhattan. For ATS, the absolute percentage error for estimated pickup rates was less than 5% in almost every spatial unit.
  • Wait Time Estimation: The model successfully estimated "sojourn times" (the total time from a passenger's request to the end of the match) with less than 10% relative error for ATS.
  • TTS Limitations: While the model performed well for the 90% of TTS activities concentrated in Manhattan, it was less accurate in outer boroughs where data is sparse and arrivals are less frequent, making the Poisson assumption weaker.

Limitations and Future Work

The authors acknowledge several areas for improvement:

  • Data Resolution: The model's reliance on high-resolution GPS data means it struggles in regions where such data is unavailable or where taxi activity is rare (e.g., outer Manhattan).
  • Behavioral Nuance: The current macroscopic view doesn't fully account for individual driver incentives or the immediate impact of dynamic pricing (surge) on driver supply.
  • Future Directions: The authors suggest generalizing the network to include more complex individual behaviors and performing sensitivity analyses to see how the system responds to major supply/demand shocks (like transit strikes or extreme weather).

Reference: Zhang, W., Honnappa, H., & Ukkusuri, S. V. (2018). Modeling Urban Taxi Services with e-hailings: A Queueing Network Approach. arXiv:1812.02021v1 [stat.AP].

Summary updated

Abstract

The rise of e-hailing taxis has significantly altered urban transportation and resulted in a competitive taxi market with both traditional street-hailing and e-hailing taxis. The new mobility services provide similar door-to-door rides as the traditional one and there is competition across these various services. In this study, we propose an innovative queueing network model for the competitive taxi market and capture the interactions not only within the taxi market but also between the taxi market and urban road system. An example is designed based on data from New York City. Numerical results show that the proposed modeling structure, together with the corresponding stationary limits, can capture dynamics within high demand areas using multiple data sources. Overall, this study shows how the queueing network approach can measure both the taxi and urban road system performance at an aggregate level. The model can be used to estimate not only the waiting/searching time during passenger-vehicle matching but also the delays in the urban road network. Furthermore, the model can be generalized to study the control and management of taxi markets.
BibTeX
@inproceedings{zhang2019modeling,
  author = {Wenbo Zhang and Harsha Honnappa and Satish V. Ukkusuri},
  title = {{Modeling Urban Taxi Services with e-hailings: A Queueing Network Approach}},
  booktitle = {Transportation},
  year = {2019},
  doi = {10.1016/j.trpro.2019.05.039},
}
Figure