One of the major prerequisites for planning or operating traffic systems is a knowledge of anticipated or existing traffic flows between defined points or zones. These origin- destination (O-D) flows are generally represented as a matrix. Origin-destination matrices are thus frequently used to represent spatially distributed travel demand. The basis for developing origin-destination matrices are travel surveys and cross-sectional traffic counts.
As O-D matrices cannot be estimated completely through surveys, traffic counts are a necessary condition for calibrating and updating them. Once a prior matrix with a reliable structure is available, it is possible to identify changes in traffic demand and to calibrate them against count data with the aid of suitable calibration methods and cross-sectional traffic counts. When so doing, it is important for the structure of the O-D matrix only to be adapted slightly so that it conforms to spatial differences in the changing figures for average traffic congestion. The matrix structure itself should remain essentially unchanged.
The main purpose of the present research was to recommend a suitable method for developing and calibrating O-D matrices on the basis of such cross-sectional traffic counts. The method had to be suitable for the existing background data and for the intended applications in Switzerland.
Because of their low cost, their simple availability and their increasingly complete coverage, methods based on cross-sectional traffic counts have become increasingly popular in recent years. Estimations of O-D flows making use of such counts attempt to pin- point one solution out of a multitude of possible solutions in which the average traffic congestion figures deduced from the estimated flows match as closely as possible the average traffic congestion figures actually measured. In this way, the method of assignment and a correctly described route choice behaviour assume central importance.
After examining the pertinent literature and taking due consideration of the demands made with respect to developing O-D matrices, the Path Flow Estimator (PFE) model was chosen as the most suitable process.
As the process developed here is to be used on both comparatively uncongested and heavily congested networks, the stochastic user equilibrium forms its major component. The PFE was subjected to further development in the following areas within the course of this investigation:
- a new route choice model was implemented (C-logit model);
- a new speed function was used for the calculation of travel times;
- asymmetric confidence intervals were applied to count locations;
- the trip distance distribution can be used as constrains;
- the total number of trips can be defined;
- calculation times were accelerated.
After this further development, the model was checked on the national road network to see whether it was plausible.
It is apparent that the demand distribution between zones can change dramatically through calibration if a small or insufficient number of count locations are available in relation to the density of the network. In such cases, there is a relatively large degree of freedom for changing the prior matrix. The greater the number of counting stations, however, the more this freedom is restricted. When calibrating, it is thus recommended to use a traffic count network as dense as possible alongside a reliable prior matrix in order to preserve the structure of the original O-D matrix.
The requirements for using the dynamic model are the same as those that apply to using the static model. Additional difficulties are presented by the prior matrices and/or their structure. Time dependent and spatial demand dynamics must be represented when developing hourly matrices.
The existing databases in Switzerland are a limiting factor for the developing of national O-D matrices. Some of the existing traffic counting stations are not suitably located for the purposes of developing matrices. It would be necessary to optimise the distribution of counting stations given the current network density and zoning.