Modelling Detailed Passengers' Behaviours in a Complex Public Transportation Network
Vincent Leblond, RATP, Crystal Langlois, RATP
RATP is operating the Parisian public transportation network which has become very complex and interconnected. Studies requirements have therefore evolved leading to more accurate traffic forecasts in terms of level of details. Our team is working on improving our assignment model of public transportation demand following five topics: competition between public transportation modes, route choice between several railway alternatives, zonal demand disaggregation around network access points, aggregating multiple routes levels of service for mode choice modelling and introducing passengers’ sensitivity to crowding. All theses improvements are implemented in in-house developed programs using the C programming language.
In order to deal with these five topics, we have developed and implemented a stochastic assignment algorithm for the public transportation demand in a new transportation model. We demonstrate that it significantly improves each separate topic and globally enhances the model representativeness and sensitivity.
In the 1970s, RATP began the internal development of GLOBAL, its transportation model. One of the first study was about the central section of the A line, which is today the first line in Paris in terms of traffic. At that time, GLOBAL was a precursor in France. Since then, the model has been used in many projects of urban railways, subway extensions and recently for trams. The Parisian public transportation network has become very complex and interconnected: RATP is now operating 14 subway lines, 2 urban railways, 3 trams and 351 bus lines. Every year, about 3 billion trips are undertaken on the network.
Operating such a complex network brings new challenges in terms of station design, fleet management, system optimization and studies of future projects like network extensions and the conception of new lines. These new infrastructures, such as the Greater Paris public transportation network, will be connected to the existing one and will offer opportunities for new itineraries as well as significant improvements of the level of services in public transportation.
Studies requirements have therefore evolved leading to more accurate traffic forecasts in terms of level of details: simulation of operating schemes, transfer and platform access time optimization, transfer trips extraction on complex multimodal nodes, detailed use of stations.... Regarding these new requirements, our team is working on improving our assignment model of public transportation demand following five topics:
(1) competition between public transportation modes (bus, dedicated-lane bus, underground, urban railway and tram);
(2) route choice between several railway alternatives;
(3) zonal demand disaggregation around network access points;
(4) aggregating multiple routes levels of service for mode choice modelling;
(5) introducing passengers’ sensitivity to crowding.
Our model follows a classic four steps methodology, implemented in in-house developed programs using the C programming language from the Dijkstra with heaps algorithm to the graphical interface. Each program is dedicated to a specific task: data editing, result analysis, GIS operations and four steps calculations. These programs are updated and enhanced everyday by our team.
Using traffic counts on various public transportation modes and surveys on passengers' choices, we have evaluated choice parameters. This way, the model reproduces the different uses of each mode. For instance, in the suburbs the bus can be used as a means of feeder to the railway network, while it appears as an alternative to the metro inside Paris.
We have estimated choice parameters on the railway network, as well as the dispersal of passengers, using disaggregated route data from surveys. Our analysis shows the variability of choices where a simple time and prices weighting is insufficient to explain individual choices. A choice model is built using believable route enumeration rules.
In order to deal with zonal demand disaggregation, we built a specific demand distribution model around network access points. Moreover the zoning definition method is enhanced to help modelling the distribution of users and ensure the matching of the Transportation Analysis Zones (TAZ) and the study.
Finally levels of service are combined with the multiple routes assignment model to determine the choice of the mode. A log-sum form is used in order to avoid under-estimation of the “mean” level of service given by the public transportation network.
Although the public transportation network is being extended, it is not growing as quickly as the demand. Furthermore many urban projects are created along urban railways. Thus, crowding in public transportation is becoming a sensitive issue as well as an everyday situation. In addition to the four previous topics, passengers' sensitivity to crowding is progressively introduced in route and mode choice models.
In order to deal with these five topics, we have developed and implemented a stochastic assignment algorithm for the public transportation demand in a new transportation model. We demonstrate that it significantly improves each separate topic and globally enhances the model representativeness and sensitivity. In addition, the sensitivity is still being tested using scenarios. These works represent major improvements for our studies. They will benefit of a continuous development using user feedbacks, methodology improvements and programming optimization.
Association for European Transport