Traffic Model for Copenhagen: Updating of the Trip Matrices
Otto Anker Nielsen, Christian Overgård Hansen, Technical University of Denmark, DK; Goran Vuk, The Danish Transport Research Institute, DK
The paper describes the undertaken methodology for constructing the travel matrices for Copenhagen, and how the improved matrices influenced the performance of the model.
Traffic planning activities in Copenhagen have over the last 10 years mainly been supported by the so-called OTM traffic model. The behavioural models in OTM include advanced utility-based formulations, which are combined with base-matrices in a pivot-correction procedure. Before and after studies of specific projects have shown, that these matrices might be the Achilles heal of the whole model system. Since January 2005 the OTM model has therefore been in a large-scale process of updating where creation of new trip matrices is in focus.
The OTM model consists of trip production, trip distribution and mode choice models, all following a utility-based framework, and network models based on mixed-Probit formulations and equilibrium formulations. The model system includes feedback cycles to take congestion into account. The behavioural functions have been estimated based on the combination of multiple Revealed and Stated Preference data-set.
Goran Vuk and Christian O. Hansen validated the present OTM 4.0 version, in a forthcoming article in Transportation, 2006, ?Validating the passenger traffic model for Copenhagen?. Poor base-matrices is likely to be the major cause of discrepancies which is revealed in the article.
The presently matrices describing the 1992 travel patterns were built upon travel analyses from the end of 1980?s and the beginning of 1990?s. Those matrices have been adjusted to the counted traffic numerous times since the first version of the model was built in 1995.
Reasons for building new travel matrices for Copenhagen are following:
· Reduction of uncertainty in forecasting, since base matrices are applied in a pivot point procedure in the OTM 5.0, and used for model estimation and calibration.
· Metro is now an existing mode, which helps building the transit matrices.
· The matrices can be applied in numerous types of travel analyses beyond modelling purposes, e.g. for the analyses of travel behaviour, infrastructure car-ownership interaction and importance of closeness to stations.
The 2004 GA-travel matrices describe travel patterns for an average working day (Monday to Friday), for five travel modes (walk, bicycle, public transport, car driver and car passenger), and six travel purposes (home-work, home-education, home-shopping, home-other private trips, non home based private trips and business trips). Day matrices were split by seven day periods. The new model operates therefore with 5 x 6 x 7 = 210 matrices.
2. Data sources
The main approach was to utilise all relevant available data-sources for the matrix estimation using a hybrid estimation approach, including:
· 33.079 personal interviews, which include a one day travel diary.
· A postcard survey in the central Copenhagen based on 61.000 handed out postcards. About 20.000 postcards were returned and coded into the data base.
· 2004 existing traffic counts for road vehicles, public transport modes and bicycle. The vehicle counts were validated by a GIS-based technique. The public transport counts were combined with digital timetable by GIS-based procedures. The bicycle counts were collected for the purpose of the model calibration only.
To improve the accuracy and spatial resolution it was decided to split the model analysis area from 601 zones into 818 zones while the surrounding area was split into 17 port zones.
3. Applied Methods
The estimation of the matrices followed the following steps;
1. Base matrices were estimated at an aggregated zonal level (90 zones) applying personal interview data. First, interview data was expanded to population stratified by socio economy and home location and adjusted for seasonal and weekly variations. Second, while mode was estimated directly on data, splitting by trip purpose was supported by a modelling approach due to the limited number of data.
2. The matrices were then adjusted according to the postcard survey to take account of children and residents outside Copenhagen. (e.g. tourists) and integrated with the postcard data.
3. The zone aggregated matrices were spatial detailed into the finer system of 818 zones by use of zonal generation and attraction rates.
4. The matrices were reformulated into simple GA-based tours by use of interview data and modelling. The distribution into seven day periods was conducted by time-of-day factors (ToD) categorized by trip purpose and eight time zones.
5. The car matrices where re-estimated on the detailed zonal structure based on traffic counts. A modified Multiple Path Matrix Estimation (MPME) procedure where used, which assumes that car users choose routes according to a stochastic user equilibrium framework. The square deviation to counts were minimised under the condition that the square deviations to the original matrices were as small as possible. Since matrices at aggregated zonal level were rather well defined and port zones had fixed observed volumes, the resulting matrices where fitted to these by a modified Furness procedure.
6. The public transport matrices where re-estimated based on traffic counts in busses and at stations. Due to the availability of automatic counting bus data, all stops and lines include passenger data. Since one zone may connect to many stops, and each stop may have zonal-connectors from several zones, a fairly complex procedure was developed for the estimation. The results where correction factors at zonal level, which were used in a modified Furness procedure. A core problem with the estimation of public matrices were, that many trips are longer than the time-intervals of the matrices, i.e. a trip ending at a certain stop may begin in a prior time period.
The paper describes the applied matrix estimation procedures, and demonstrates appraisals of the new matrices. The study had a fairly large budget ? out of the total budget amounting to Euro 710.000, activities related to building of the travel matrices had a budget of about Euro 430.000. However, we believe that quality of modelling depends on the data foundation and the hybrid approach has been cost efficient.
Association for European Transport