Trip-matrices have traditionally been estimated based on comprehensive and often costly traffic surveys. However, in the recent years the amount ofcollected data on actual counted flows has increased significantly and the quality of counts have become more reliable due to improved automatic counting methods. Methods for estimating matrices from traffic counts are therefore interesting and relevant alternatives to more expensive survey methods. One such method is the Multi Path Matrix Estimation (MPME) formulated by O Anker Nielsen in 1994. It is a heuristic method where an existing trip-matrix is adjusted based on traffic counts. This is done in such a waythat each cell in the matrix is adjusted according to all counts found onthe routes between the zones in network. The method is integrated with the assignment model which makes it possible to use more complex route choice models like the stochastic user equilibrium, where counts are weighted according to the probabilities of the routes. Overall the weighted square deviation between assigned matrices and counts are minimized by changing a base-matrix as little as possible (again by minimizing the weighted squaredeviation). Another advantage is that the method can be applied on large-scale networks and that it can handle non-consistent counts. The main objective of this paper is to evaluate the MPME-method and how it performs in large-scale networks. Secondly data requirements for the method will be investigated and especially the number and geographical spreading of counts are analyzed with respect to the quality of the matrix estimation. Finallythe paper will introduce improvements to the method. The MPME-method has recently been used for updating matrices for the regional model in Copenhagen. Likewise the method has been used to adjust matrices to the new freight model covering the east Denmark and southern parts of Sweden (the Oeresund freight model), matrices for the urban train system in Copenhagen, andEuropean transport matrices in the EU Trans-tools model (as being re-estimated in the TEN-CONNECT project). Based on knowledge gained in the projects some key measurements to evaluate the adjustments are defined. The results of the adjusted matrices are discussed and the methods strengths and weaknesses are revealed. To a get detailed insight in the methods validity,further tests are carried out based on selected cases. This is done by first modeling traffic flows into a network to obtain synthetically counts on all links. Then the MPME-methods ability to reproduce the original matrix when fed with a different matrix is tested. By removing counts by randomas well as in geographically patterns the necessity of counts is evaluated. Finally it is also analyzed how big an impact the correctness of the input-matrix has on the methods ability to reproduce the original matrix. The results show that MPME adjustments did improve the trip-matrices for thelarge-scale-models. In assignments carried out with the adjusted matricesthe flows are more consistent with the counts. The results also reveal that the method has a tendency to reduce the average trip lengths, especially in cases with fewer counts in the network. The tests carried out furthermore shows that the result of the adjustment is highly depending on the input-matrix. This is especially true in case with few and/or geographicallypoorly distributed counts. To improve the method a couple of suggestions are presented. These are designed to solve the tendency of the method to reduce the average trip lengths. For the covering abstract see ITRD E145999
Samenvatting