|
Optimization of road traffic counts location in a network
Authors: Gilliéron, F. i J. Barceló
Congress: European ITS Conference
Place and date: Graz (Àustria), del 30 de juny al 2 de juliol de 2008
Language: anglès
Summary:
Simulation is a technique that can be seen as a sampling experiment on a dynamic real system through a computer model formally representing it. Simulation assumes that the evolution over time of the system’s model imitates properly the evolution over time of the modeled system and thus samples of the observational variables of interest are collected from which, using statistical analysis techniques, conclusions on the system behavior can be drawn. The reliability of this decision making process depends on the ability to produce a simulation model representing the system’s behavior closely enough for the purpose of using the model as a substitute of the actual system for experimental purposes.
The question on whether a model is valid or not can be formulated in terms of whether model results faithfully represent reality, a question for which statistical techniques provide a quantitative answer. In this framework the analyst perception of the reality relays on the information gathered through the data collection and the subsequent data processing to account for uncertainties. In almost all the discussions and methodological approaches for calibration and validation of traffic simulation models the attention has been drawn by the processes to accurately estimate model parameters and the statistical methods to assess the model validity, implicitly assuming that the available data to compare with were reliable enough. Data inputs to traffic models can be classified in two categories:
- Data directly observable, i.e. measurements of traffic variables (flows, speeds, occupancies, travel times…) based on available technologies
- Data not directly observable, as the transport demand modeled in terms of time sliced Origin-Destination matrices.
The common current practice usually considers data in the first category as error free in spite of the wide evidence against ad frequently are the basis for the methods to estimate the second ones, i.e. adjustment of OD matrices from link flow counts. However, the use of the usual statistical methods in a straightforward way can lead to serious mistakes, consequence of the uncertainties affecting the available data, observable as well as unobservable. Therefore a structured component checking quality of real world traffic data integrated in between the data collection system and the simulation model is required. This paper presents and discuses some procedures to assess the traffic data quality based on statistical and structural analysis, the classification of the data and its automatic integration in a learning process: completed by the discussion on their implementation into the ALMO system. The paper also presents some real life examples of the application of the proposed methods.
|
|
