How Much Urban Traffic is Searching for Parking?

Chase Dowling, Tanner Fiez, Lillian J. Ratliff, Baosen ZhangarXiv 2017 · OpenReview

Abstract

With the increasing availability of transaction data collected by digital parking meters, paid curbside parking can be advantageously modeled as a network of interdependent queues. In this article, we introduce methods for analyzing a special class of networks of finite capacity queues, where tasks arrive from an exogenous source, join the queue if there is an available server, or are rejected and move to another queue in search of service according to the network topology. Such networks are useful for modeling curbside parking since queues in the network perform the same function and drivers searching for an available server are under combinatorial constraints. We provide a motivating example for such networks in the context of drivers searching for parking in the neighborhood of Belltown in Seattle, Washington, USA. Lastly, we introduce a simulation tool to test the assumptions made to estimate performance metrics. Our results suggest that a Markovian relaxation of the problem when solving for mean rate metrics is comparable to deterministic service times reflective of a driver's tendency to park for the maximum allowable time.

Introduction

In dense urban centers, "cruising" for parking—the act of circling blocks in search of an available space—is a significant yet often unquantified contributor to traffic congestion. Historical estimates suggest that searching for parking can account for up to 30% of through-traffic in busy districts. While modern cities collect vast amounts of transaction data from digital parking meters, translating these raw logs into an accurate picture of traffic flow remains a challenge.

Standard queueing models typically fail to capture the nuances of urban parking. Curbside parking is characterized by finite capacity (a block can only hold so many cars) and non-separable dynamics (congestion on one block spills over to adjacent streets). Crucially, drivers do not simply exit the system when a spot is unavailable; they continue through the street topology, adding to the very congestion they seek to escape. This paper bridges the gap between transaction data and traffic engineering by modeling parking as a network of interdependent, finite-capacity queues.

Modeling Curbside Parking as a Queueing Network

The authors represent a city's parking infrastructure as a Network of Finite Capacity Queues. In this framework, each block-face is treated as a node where the "servers" are the physical parking spaces. The edges of the network represent the streets, modeled as infinite-server queues with fixed travel times to account for the transit between blocks.

Schematic of the parking network model showing blockface queues (M/M/k/k) and road queues (G/D/∞).

When a driver arrives at a block-face that is already at capacity, they are "rejected" and must transit to an adjacent block-face based on the legal street maneuvers. This "searching" behavior continues until a spot is found or the driver leaves the area. A key technical contribution of this work is the Markovian Relaxation. While real-world parking durations are often deterministic—drivers frequently park for the maximum allowed time (e.g., 2 or 4 hours)—the authors test whether assuming an exponential distribution (a Markovian property) can simplify the mathematical analysis without sacrificing accuracy.

Empirical Validation and Simulation

To validate the model, the study utilizes a high-resolution dataset from the Seattle Department of Transportation (SDOT), covering the Belltown neighborhood from March to July 2016. This area includes 256 block-faces with capacities ranging from 1 to 20 spaces. The authors developed a custom Python-based simulator to perform Monte Carlo simulations of these networks, allowing for the analysis of irregular real-world street layouts that do not fit symmetric theoretical models.

Key Results and Discussion

The simulation results demonstrate high accuracy in predicting real-world conditions. The model achieved an average occupancy error of only -5.3% compared to observed data in Belltown. Furthermore, the estimated rejection rates—the frequency at which drivers are turned away from full blocks—showed a minimal average error of -0.19 vehicles per hour.

A significant finding is the robustness of the Markovian relaxation. As shown in the performance metrics, the steady-state occupancy and rejection rates for exponential service times closely track those of fixed (deterministic) service times across various arrival rates.

Comparison of performance metrics (occupancy percentage and rejection rates) for exponential vs. fixed service times in a 10-node network.

While exponential service times may take slightly longer to reach a steady state in simulation, they eventually converge to the same occupancy levels as deterministic models. This suggests that the easier-to-calculate Markovian models are a viable proxy for complex urban parking systems.

Conclusion and Future Work

This research provides a scalable framework for quantifying the impact of parking search on urban traffic. By proving that the system remains stable as long as arrival rates do not exceed total capacity, and demonstrating the effectiveness of Markovian relaxations, the authors offer tools for more efficient traffic management. Future work will address current limitations, such as the assumption of legal parking and uniform random search strategies, to further refine these estimates of "cruising" traffic.

Summary updated

Abstract

With the increasing availability of transaction data collected by digital parking meters, paid curbside parking can be advantageously modeled as a network of interdependent queues. In this article we introduce methods for analyzing a special class of networks of finite capacity queues, where tasks arrive from an exogenous source, join the queue if there is an available server or are rejected and move to another queue in search of service according to the network topology. Such networks can be useful for modeling curbside parking since queues in the network perform the same function and drivers searching for an available server are under combinatorial constraints and jockeying is not instantaneous. Further, we provide a motivating example for such networks of finite capacity queues in the context of drivers searching for parking in the neighborhood of Belltown in Seattle, Washington, USA. Lastly, since the stationary distribution of such networks used to model parking are difficult to satisfactorily characterize, we also introduce a simulation tool for the purpose of testing the assumptions made to estimate interesting performance metrics. Our results suggest that a Markovian relaxation of the problem when solving for the mean rate metrics is comparable to deterministic service times reflective of a driver's tendency to park for the maximum allowable time.
BibTeX
@misc{dowling2017how,
  author = {Chase Dowling and Tanner Fiez and Lillian J. Ratliff and Baosen Zhang},
  title = {{How Much Urban Traffic is Searching for Parking?}},
  year = {2017},
  eprint = {1702.06156},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1702.06156},
}

Loading related content...

Figure