Highway Truck Parking Prediction System and Statistical Modeling Underlying its Development

0442563 - ÚI 2016 RIV RS eng C - Konferenční příspěvek (zahraniční konf.)
Brabec, Marek - Konár, Ondřej - Kasanický, Ivan - Pelikán, Emil - Malý, Marek
Highway Truck Parking Prediction System and Statistical Modeling Underlying its Development.
Proceedings of the Second International Conference on Traffic and Transport Engineering. Belgrade: City Net Scientific Research Center, 2014 - (Čokorilo, O.), s. 164-170. ISBN 978-86-916153-2-1.
[ICTTE 2014. International Conference on Traffic and Transport Engineering /2./. Belgrade (RS), 27.11.2014-28.11.2014]
Grant CEP: GA TA ČR TA02031411
Institucionální podpora: RVO:67985807
Klíčová slova: highway truck parking * prediction system * dynamical statistical modeling * generalized additive model
Kód oboru RIV: BB - Aplikovaná statistika, operační výzkum

In this paper, we will describe a system for on-line prediction of truck parking demand along highway system in the Czech Republic. We will describe structure of the system developed during the TACR TA02031411 project and mention some of its specific functionalities. Further, we will explain in detail statistical modeling methodology which underlies the forecasting model in the core of the prediction procedure. Whole system relies on the use of indirect but very precise and relatively cheap to obtain toll transaction data (accessible through a cooperation with Kapsch Telematic Services, Inc.). Our statistical modeling starts with a recognition of the fact that the number of trucks parking at a given lot and given time is a latent variable to be estimated from the observable toll transaction data (which are available in the form of times when individual truck pass toll gates). After constructing appropriate proxy variable, we formulate a flexible class of statistical semi-parametric models constructed in a Markovian fashion. In fact, our model can be viewed as a non-homogeneous Markov chain, whose Poissonian transition probabilities change with several external covariates (describing e.g. weekly and daily periodicity of parking intensities) as well as spatially. Once the model is estimated (its parametric and nonparametric parts are estimated simultaneously), it is used for real time prediction for several short to medium horizons, using Monte Carlo simulations to obtain efficient and robust software implementation. We will demonstrate practical performance of the prediction system under routine conditions, based on evaluation against manual parking lot counting.
Trvalý link: http://hdl.handle.net/11104/0245366