| This research exposes a statistical analysis of travel time observations on critical portions of the Montreal road network. Whereas we have access to odometric data collected by the Ministere des Transports du Quebec (MTQ) between 1998 and 2004 on several highway segments of the Montreal network, the required tasks are: review of existing traffic indicators and the technologies used to gather data, estimate statistically the state of the traffic on the highway network, assess the evolution of traffic on the network, propose a data collection methodology for future sampling of travel times on the road network and, finally, evaluate the reliability of travel times on the Montreal highway network.;For the current research project, floating car data collected by MTQ are processed in an innovative way. Sampled segments have been cut in equally-sized portions. Then, we had to make our data more accurate, by deleting abnormal observations, and observations with missing values. These values are spatial location, timestamp and date of the observation and travel time value.;Thanks to a reliable data collection, we applied a statistic analysis on the data of the first segment (highway 13th, direction South). The conclusions confirm the influence of several factors: month of the year, road conditions, period (AM or PM). This analysis was executed using analysis of variance, and proposes a graphical interface assisting the quality control of observed data.;Then, it appears that the frequency distribution of travel times seem similar for several network segments. That point leads us to consider groups of segments whose frequency distribution of travel times are similar. With a correlation analysis and a clustering algorithm, several groups have been exposed. Furthermore, we chose to evaluate the variability, reliability indicator, for each segment.;All over the world, the different managers of the traffic have been seeing a dramatic evolution of the highway traffic, thus for all kinds of cities. Faced with increasing pressure on the network that make congestion reduction almost impossible, managers now aim at better assessing this evolution and the impacts of various types of incident to provide more reliable network with respect to travel times. In fact, managers are not trying to eradicate congestion anymore, but try to measure congestion and anticipate the causes and consequences of an incident on the traffic. To allow this new concept, several technologies are used and very often used at the same time: static technologies (induction loops, radar, camera), and mobile technologies (floating cars, and more recently, Bluetooth and Cellular).;In order to complete the sampling requirement, and thanks to the previous analysis, we could compute a sampling frame based on the variability of travel times. Seeing the complexity of making such a sample, we used the results of the clustering algorithm and improve the future data collecting method by reducing the number of required samples. We applied this algorithm on every segment considering the differences of the number of actual samples collected per segment between 1998 and 2004.;Then, whereas the idea of clustering has been studied, we also noticed the singular form of the frequency distribution of travel times. Consequently, we applied a modelling to that distribution. From this modelling, we have been able to simulate the mean and the variability of travel times and develop several new indicators: risk of incident indicator, and Mean-Variability indicator. This last indicator can be used mathematically by the manager of the traffic or by the road user in his categorical form.;Finally, thanks to these last results, we have chosen to study the year 2004 in front of the results for all of the sampled years, in order to expose a critical evolution of the traffic on the first segment. The conclusions seem to tell us that no dramatic evolution of travel times has occurred on this segment.;The conclusions are the followings: we proved the influence of several factors, propose a sampling method based on the variability and the clustering algorithm, model the distribution of travel times, simulate mean and variability and develop new indicators. This leads to various applications. Since the modelling of the distribution is currently poor for some circuits, we will have to try a new method: add several factors to our data, like the type of way, the presence of crossing ways, etc, and apply the clustering algorithm on the entire collection of portions. Followed by a modelling of the frequency distribution of each newly created group, we will be able to simulate mean and variability, and give new values to the previously created indicators. Consequently, we will be able to propose a new sampling, with a new definition of segments. |
