How likely am I to find parking? – A practical model-based framework for predicting parking availability

Jun Xiao, Yingyan Lou, Joshua Frisby

Research output: Contribution to journalArticle

5 Citations (Scopus)

Abstract

Parking availability information (or occupancy of parking facility) is highly valued by travelers, and is one of the most important inputs to many parking models. This paper proposes a model-based practical framework to predict future occupancy from historical occupancy data alone. The framework consists of two modules: estimation of model parameters, and occupancy prediction. At the core of the predictive framework, a queuing model is employed to describe the stochastic occupancy change of a parking facility. While the underlying queuing model can be any reasonable model, we demonstrate the framework with the well-established continuous-time Markov M\M\C\C queue in this paper. The possibility of adopting a different queuing model that can potentially incorporate the parking-searching process is also discussed. The parameter estimation module and the occupancy prediction module are both built on the underlying queuing model. To apply the estimation and prediction methods in real world, a few practical considerations are accounted for in the framework with methods to handle variations of arrival and departure patterns from day to day and within a day, including special events. The proposed framework and models are validated using both simulated and real data. Our San Francisco case studies demonstrate that the parameters estimated offline can lead to accurate predictions of parking facility occupancy both with and without real-time update. We also performed extensive numerical experiments to compare the proposed framework and methods with several pure machine-learning methods in recent literature. It is found that our approach delivers equal or better performance, but requires a computation time that is orders of magnitude less to tune and train the model. Additionally, our approach can predict for any time in the future with one training process, while pure machine-learning methods have to train a specific model for a different prediction interval to achieve the same level of accuracy.

Original languageEnglish (US)
Pages (from-to)19-39
Number of pages21
JournalTransportation Research Part B: Methodological
Volume112
DOIs
StatePublished - Jun 1 2018

Fingerprint

Parking
Availability
learning method
Learning systems
Parameter estimation

ASJC Scopus subject areas

  • Civil and Structural Engineering
  • Transportation

Cite this

How likely am I to find parking? – A practical model-based framework for predicting parking availability. / Xiao, Jun; Lou, Yingyan; Frisby, Joshua.

In: Transportation Research Part B: Methodological, Vol. 112, 01.06.2018, p. 19-39.

Research output: Contribution to journalArticle

@article{8081f52ce15d4cae9c5040fd8f4c8e24,
title = "How likely am I to find parking? – A practical model-based framework for predicting parking availability",
abstract = "Parking availability information (or occupancy of parking facility) is highly valued by travelers, and is one of the most important inputs to many parking models. This paper proposes a model-based practical framework to predict future occupancy from historical occupancy data alone. The framework consists of two modules: estimation of model parameters, and occupancy prediction. At the core of the predictive framework, a queuing model is employed to describe the stochastic occupancy change of a parking facility. While the underlying queuing model can be any reasonable model, we demonstrate the framework with the well-established continuous-time Markov M\M\C\C queue in this paper. The possibility of adopting a different queuing model that can potentially incorporate the parking-searching process is also discussed. The parameter estimation module and the occupancy prediction module are both built on the underlying queuing model. To apply the estimation and prediction methods in real world, a few practical considerations are accounted for in the framework with methods to handle variations of arrival and departure patterns from day to day and within a day, including special events. The proposed framework and models are validated using both simulated and real data. Our San Francisco case studies demonstrate that the parameters estimated offline can lead to accurate predictions of parking facility occupancy both with and without real-time update. We also performed extensive numerical experiments to compare the proposed framework and methods with several pure machine-learning methods in recent literature. It is found that our approach delivers equal or better performance, but requires a computation time that is orders of magnitude less to tune and train the model. Additionally, our approach can predict for any time in the future with one training process, while pure machine-learning methods have to train a specific model for a different prediction interval to achieve the same level of accuracy.",
author = "Jun Xiao and Yingyan Lou and Joshua Frisby",
year = "2018",
month = "6",
day = "1",
doi = "10.1016/j.trb.2018.04.001",
language = "English (US)",
volume = "112",
pages = "19--39",
journal = "Transportation Research, Series B: Methodological",
issn = "0191-2615",
publisher = "Elsevier Limited",

}

TY - JOUR

T1 - How likely am I to find parking? – A practical model-based framework for predicting parking availability

AU - Xiao, Jun

AU - Lou, Yingyan

AU - Frisby, Joshua

PY - 2018/6/1

Y1 - 2018/6/1

N2 - Parking availability information (or occupancy of parking facility) is highly valued by travelers, and is one of the most important inputs to many parking models. This paper proposes a model-based practical framework to predict future occupancy from historical occupancy data alone. The framework consists of two modules: estimation of model parameters, and occupancy prediction. At the core of the predictive framework, a queuing model is employed to describe the stochastic occupancy change of a parking facility. While the underlying queuing model can be any reasonable model, we demonstrate the framework with the well-established continuous-time Markov M\M\C\C queue in this paper. The possibility of adopting a different queuing model that can potentially incorporate the parking-searching process is also discussed. The parameter estimation module and the occupancy prediction module are both built on the underlying queuing model. To apply the estimation and prediction methods in real world, a few practical considerations are accounted for in the framework with methods to handle variations of arrival and departure patterns from day to day and within a day, including special events. The proposed framework and models are validated using both simulated and real data. Our San Francisco case studies demonstrate that the parameters estimated offline can lead to accurate predictions of parking facility occupancy both with and without real-time update. We also performed extensive numerical experiments to compare the proposed framework and methods with several pure machine-learning methods in recent literature. It is found that our approach delivers equal or better performance, but requires a computation time that is orders of magnitude less to tune and train the model. Additionally, our approach can predict for any time in the future with one training process, while pure machine-learning methods have to train a specific model for a different prediction interval to achieve the same level of accuracy.

AB - Parking availability information (or occupancy of parking facility) is highly valued by travelers, and is one of the most important inputs to many parking models. This paper proposes a model-based practical framework to predict future occupancy from historical occupancy data alone. The framework consists of two modules: estimation of model parameters, and occupancy prediction. At the core of the predictive framework, a queuing model is employed to describe the stochastic occupancy change of a parking facility. While the underlying queuing model can be any reasonable model, we demonstrate the framework with the well-established continuous-time Markov M\M\C\C queue in this paper. The possibility of adopting a different queuing model that can potentially incorporate the parking-searching process is also discussed. The parameter estimation module and the occupancy prediction module are both built on the underlying queuing model. To apply the estimation and prediction methods in real world, a few practical considerations are accounted for in the framework with methods to handle variations of arrival and departure patterns from day to day and within a day, including special events. The proposed framework and models are validated using both simulated and real data. Our San Francisco case studies demonstrate that the parameters estimated offline can lead to accurate predictions of parking facility occupancy both with and without real-time update. We also performed extensive numerical experiments to compare the proposed framework and methods with several pure machine-learning methods in recent literature. It is found that our approach delivers equal or better performance, but requires a computation time that is orders of magnitude less to tune and train the model. Additionally, our approach can predict for any time in the future with one training process, while pure machine-learning methods have to train a specific model for a different prediction interval to achieve the same level of accuracy.

UR - http://www.scopus.com/inward/record.url?scp=85052980641&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85052980641&partnerID=8YFLogxK

U2 - 10.1016/j.trb.2018.04.001

DO - 10.1016/j.trb.2018.04.001

M3 - Article

AN - SCOPUS:85052980641

VL - 112

SP - 19

EP - 39

JO - Transportation Research, Series B: Methodological

JF - Transportation Research, Series B: Methodological

SN - 0191-2615

ER -