Analisi comportamentale della scelta del percorso attraverso l'utilizzo di nuove tecnologie di acquisizione delle informazioni

In travel demand modeling, route choice is one of the most complex decision-making contexts to understand and mathematically represent for several reasons. Firstly, a large number of available paths may exist between the same origin-destination (OD) pair. Secondly, neither the traveler nor the modeler are aware of all the available alternatives. Thirdly, individual choices are dictated by different constraints and preferences that are difficult to capture by modelers who face increasingly larger datasets where retrieving the exact path chosen by travelers is not always straightforward. Last, there is a lot of uncertainty about travelers’ perceptions of route characteristics as well as other characteristics that can influence their choices, such as age, gender, habit, weather conditions and network conditions. This highlights the difficulties encountered for interpreting individual user behavior in greater depth. The rapid advances in GPS devices, has resulted in major benefits for data collection, which now can be recorded automatically and with greater accuracy compared to the techniques used in the past (phone calls, e-mails, face-to-face interviews, laboratory experiments.). On these basis the main objective of the thesis is then to study route choice using a GPS database. The data were acquired during a survey, named "Casteddu Mobility Styles” (CMS), conducted by the University of Cagliari (Italy) in the metropolitan area of Cagliari between February 2011 and June 2012. Each participant was asked to carry a smartphone with builtin GPS in which an application called “Activity Locator” – implemented by CRiMM (Centre for Research on Mobility and Modeling) – had been installed. A total of 8831 trips were recorded by 109 individuals, of which 4791 referring to the car driver mode. Each GPS track (consisting of a sequence of referenced position points) was then treated with map-matching techniques, through which it was possible to associate each “GPS point” to a link of the network, thus creating the observed route database. The first objective of the thesis is to understand which are the characteristics of the data acquired during the CMS survey, doing firstly the same analysis that other authors did in their researches based on GPS data. In almost all the previous researches, the GPS data were collected through in-vehicle surveys that make it possible to gather objective information on trips (travel times and distances). Pre-and post-analysis interviews were conducted to gather information about the subjective characteristics of the individuals and GIS platforms were used to study the routes. In the present study, the data were collected using an integrated system able to also record the activities conducted, along with all the characteristics associated thereto. In this way a complete database was created containing all the information (objective and not) concerning the trips. For comparisons with the objectively most convenient paths, then, was used a static macrosimulation model (implemented in CUBE, Citilabs Ltd.) of the entire study area (Cagliari and its metropolitan area), which reproduces the network characteristics actually encountered by the drivers referring to the data used. From this first analysis it was observed that when more than one route is taken for repetitive trips between the same OD. In order to understand these particular behavior of users, named also intravariability, discrete choice models were estimated. It’s important to note that in the previous GPS-based researches this particular behavior was only identified, without studying it in depth. Several other studies, focused on route switching behavior, tried to understand it applying discrete choice models, but their database were based on data acquired through questionnaires or laboratory experiments, and for the majority the route switching behavior was studied in relation to the trip information provision. The objective of this analysis is then to combine the two fields of the research on route switching, trying to understand it estimating discrete choice models using a GPS based database, closing the gap of the previous researches. The final goal of the model estimations is to understand which are the main attributes of the routes and the characteristics of the users that most influence the choice of an habitual route for the same origin-destination (OD) trip. After these first analysis, the final objective of the thesis is to apply a route choice model to GPS-based data. Modeling route choice behavior is generally framed as a two-stage process: generation of the alternative routes and modeling of the choice from the generated choice set. The focus of this step of the research is on the bias that might be introduced in the model estimation by the choice set generation process. Specifically, although several explicit choice set generation techniques are found in the literature, the focus is on stochastic route generation and the correction for unequal sampling probability of routes when applying this technique that is easily applicable to large-scale networks. Indeed, stochastic route generation is a case of importance sampling where the selection of the path depends on its own properties, so route choice models based on stochastic route generation must include a sampling correction coefficient that accounts for the different selection probability. In this study is proposed a methodology for calculating and considering this correction factor into MNL-based models with choice sets generated by means of stochastic route generation. Specifically, was decided to look at the sampling correction factor proposed for the random walk algorithm and to calculate the route selection probability in order to exploit this expression. Therefore, a procedure is proposed for the computation of the selection probabilities on the basis of the stochastic generation principle, then the correction factor and last the EPS for model estimation. The modeling analysis confirms the functionality of the proposed approach that has great advantages: (i) it provides insight into the application of stochastic generation in route choice modeling, especially in large-scale networks where the only need is a standard random number generator and a Dijkstra algorithm; it proposes a simple and manageable procedure from the computational perspective for the calculation of route selection probabilities and hence the correction factor and EPS for model estimation; it proves the efficiency of the proposed methodology on revealed preference data in a dense urban network by showing an increase in goodness-of-fit of the model and a shift from illogical to logical sign in parameters estimated for key variables such as travel time.