@phdthesis{,
    author = {Auer, Markus},
    title = {Empirical Analysis of Driving and Route Choice Behavior in Traffic Networks Based on Connected Vehicle Data},
    editor = {},
    booktitle = {},
    series = {},
    journal = {},
    address = {},
    publisher = {},
    edition = {},
    year = {2019},
    isbn = {},
    volume = {},
    number =  {},
    pages = {},
    url = {},
    doi = {},
    keywords = {Empirische Verkehrsdaten; Routenwahl; Umfahrung; Mobilitätsauswertung},
    abstract = {The empirical analysis of driving and route choice behavior in traffic networks is based on position data from a large connected vehicle fleet. While the driving behavior is analyzed on a macroscopic level, the route choice behavior in congested networks is an analysis on microscopic level. The mobility parameters of individual motor car traffic for selected 16 European countries shows little differences in the car usage between the individual countries. There are roughly 4 trips per day with an average trip travel time of 20 to 25 minutes and distance of 15 to 20 km. Some countries like Luxembourg have considerable shorter trip travel times and distances than the majority, while both travel time and distance are higher in Denmark. The deviations from the majority can be explained by the size of the country which does not allow long straight trips without crossing the border. In case of Denmark the longer trips are probably caused by the special road network due to the many islands. The overall travel time ranges mainly between 80 and 100 minutes per day and the total distance between 60 and 80 km. The temporal analysis of the vehicle data yields traffic volume curves which are already well-known from detector data. Furthermore, velocity curves over the day are obtained for the whole network, which are qualitatively different to velocity measurements on single detectors. The weekly distribution of average number of trips, travel time, distance and velocity show the different vehicle usage patterns from Monday to Friday and on Saturday and Sunday. As the vehicle fleet data has a much higher accuracy than travel survey data and a higher coverage than detector data, detailed distributions of the mobility parameters can be generated. Similar distributions for the travel distance has been already established with mobile phone billing data. The general results of this study can be confirmed. However, there are little variations in the curve fit parameters. This can be explained by the fact that the analysis of mobile phone data comprises all forms of transport while the present study is restricted on individual motor car traffic. Further distributions are obtained for the number of trips, trip travel time and velocity. Some of the fitted statistical distributions are well suited like the exponential distributions for the trip travel time and total distance per day. The Poisson distribution for the number of trips seem to be qualitatively correct, but there are some deviations. Therefore, the number of trips are probably distributed according to similar distribution variant of the Poisson distribution. The average trip velocity is well approximated by the lognormal distribution in the interval between 5 and 100 km/h. The analysis of the road class dependency shows that the traffic volume is more or less equally distributed between the road classes for each country. Some effects in the traffic volume or velocity curves are observable only for distinct road classes. The velocity drop in the rush hour is only clearly formed for regional and long distance roads and very weak for local roads. The detailed analysis of routing decisions in congested freeway networks in the second part of this work is conducted on freeways located in Germany, United Kingdom and France. The general type of detour analyzed in this work is the bypassing of a congested freeway part between two interchanges in the secondary road network. This is a very general case and applies for long distance routes with long parts on freeways. The processing of the trajectories can be performed either topology-free or on a link basis. The topology-free approach requires no exact knowledge of the digital map, but in turn is also less accurate than the topology-based processing of the map-matched position data. With map-matched position data together with the digital map the exact reconstruction of the driven route is better feasible and yields better results in the classification of main and alternate routes. The empirical analysis of the detours is divided into several parts: - Spatial Variability: Permanently signed detour routes can be used between subsequent interchanges of freeways in order to circumvent congestion on the respective freeway part. From the empirical trajectory data it can be seen that drivers decide also for deviating detour routes between interchanges in any order. - Temporal Analysis: In case of emerging congestion on the freeway both main and alternate routes are unbalanced and so it is possible to gain large travel time advantages by detouring. Over time the travel times on main and alternate routes equalize again and with decreasing congestion on the freeway also the advantage of detours vanishes. - Spatial Analysis: Based on historical data some detours between distinct interchanges prove to be more suitable than others. Some detour routes are on average always slower while others are always faster. - Detour Rates: The analysis of detour rates yields as expected a positive correlation between both expected and empirical travel time differences and the ratio of detouring vehicles. With more expected and realized travel time advantage also more drivers decide to drive on an alternate route. Overall drivers can reduce their travel time by detours in the range of some minutes (< 5min). However, there are also lots of drivers increasing their travel time by deciding for an alternate route. This can have various reasons like increasing traffic in the secondary road network or weaker congestion on the freeway. With perfect prediction all these dynamic routing problems would vanish. But especially the temporal traffic dynamics on main and alternate routes are hardly to predict. Firstly the routing decisions of the drivers are unknown and secondly the period of perturbations due to accidents is not known.
However, heuristics can be developed in order to improve the routing decisions for detours of individual drivers. The first proposed heuristic is based on the average historical travel time differences between main and alternate route. If on average a detour route was 𝑥 minutes slower than the main route, a future detour route should be only suggested with an expected travel time advantage of at least 𝑥 minutes. The second heuristic exploits the relation between predicted and empirical travel time in order to obtain an optimal routing criterion.},
    note = {},
    
    school = {Universität der Bundeswehr München},
    
}