Prediction Reliability of the Transport Simulation Models: a Before and After Study in Naples
E Cascetta, A Papola, University of Naples, IT; A Carteni, University of Salerno, IT
This paper investigates the prediction reliability of the transportation system models, through a ?before and after? study. Moreover a new procedure for the OD demand matrix correction is proposed.
A transportation system can be defined as the combination of elements and their interactions, which produce the demand for travel within a given area and the supply of transportation services to satisfy this demand.
The relevant interactions among the various elements of a transportation system can be simulated with mathematical models. Supply models simulate the performances of the transportation services available among the different zones and demand models simulate the relevant aspects of travel demand as a function of the activity system and of the supply performances. Typically, the characteristics of travel demand simulated include the number of trips in the reference period and their distribution among the different zones, the different transport modes, and the different paths. The interaction between demand and supply can be simulated trough assignment models which allow the calculation of link flows.
In literature there are a great number of applications of these models. They are generally estimated so as to reproduce the observed actual system conditions and then applied to predict the future system conditions due to possible changes in the supply system (T*I, 1990; HCG, 1991; Algers S., 1994; Cascetta E., 1995; Gunn H. and de Jong G., 2000; Cascetta E., 2001; Coppola P. and Cartenì A., 2001).
But what is the reliability of these predictions? Very few studies have been conducted in order to verify the reliability of these models by comparing the real future system conditions with those predicted by the model.
This paper investigates the prediction reliability of the transportation system models, through a ?before and after? study, taking advantage of the opening of a new part of a Neapolitan subway in December 2002. Moreover a new procedure for the OD demand matrix correction was proposed able to identify possible macro mistakes in the supply/assignment model.
The study can be organized in three different moments.
First of all, the transportation model system of the Naples metropolitan area was implemented (specified and estimated).
Then, the set of surveys needed for making up the database for the before and after study was planned and executed.
Finally the before and after study was conducted by comparing link flows predicted by the model with those observed in both before and after scenarios.
Concerning the model system, two supply models were implemented for both before and after scenarios. Moreover, a standard four step demand model was used considering four purposes: commuting, high school student, university student and other purposes. Finally, the network assignment was conducted through a user equilibrium model with stochastic (probit) path choice model for the car mode and through a network loading model with deterministic hyperpath choice model for public modes.
Concerning the survey planning, the before and after observed variables are link flows and the relative location of the traffic count sections which was identified with the following methodology:
-first, the demand predicted by the model in the before scenario was assigned to the network of the after scenario (including the new subway line) so as to identify the sub OD demand matrix composed by all and only demand flows which the model predict on the new subway links;
-then, this sub matrix was assigned to the network of the before scenario so as to identify services and links currently used by this demand. The traffic count sections are an opportune subset of these links.
On these sections, the traffic counts were obviously conducted both before (November-December 2002) and after (May 2003) the opening of the new subway line.
Concerning the before and after study, first of all the whole model system (the topology of the network, the parameters of the generalized cost function, the parameters of the route choice model and the demand matrix elements) was corrected by using link flows observed in the before scenario.
It is important to underline that a new GLS procedure for the demand matrix correction was proposed able to identify possible macro mistakes in the supply/assignment model. This procedure, indeed, correct the demand matrix so as to perfectly reproduce the observed link flows. Differently from the typical GLS problem, the feasibility set of the proposed GLS estimator can be empty; in other words, a non negative demand matrix which assigned to the network exactly reproduces the observed link flows, could not exist. This possibility would show an incompatibility between assignment matrix and traffic counts and, consequently, some possible errors in the supply and/or assignment models. The produced software implementing this procedure is able to locate such incompatibilities and to identify the traffic counts that determine them. In this way it is generally possible to identify and correct the possible errors in the supply and/or assignment models.
After the correction of the whole model system, the link flows predicted by the corrected model in the after scenario were compared with those observed and the percentage difference resulted quite low. This is a further empiric evidence that the OD matrix correction with traffic flows is a procedure able to substantially increase the prediction reliability of a transport system model if well and correctly used, that is by opportunely locating the traffic count sections and by using the observed data for the simultaneous correction of the whole model.
Association for European Transport