Choice of Travel Demand Forecast Models: Comparative Analysis in Urban Rail Route Choice
H Kato, University of Tokyo, JP; Y Kaneko, Nihon University, JP; M Inoue, Creative Research and Planning Co. Ltd., JP
This paper analyzes empirically to what extent the choice of travel demand forecast techniques impact on the estimated results. The urban rail demand forecast in the Tokyo Metropolitan Area will be used for the empirical analysis.
Various types of techniques have been so far used for travel demand forecast in transport planning. In the actual practical transport projects, the choice of the technique often depends on the analyst's preference or his/her knowledge and experience. However, even if the same data set is used, the estimated results may vary among various types of demand forecast techniques. This could cause the inadequate analysis and it may result into the wrong decision-making. This paper analyzes empirically to what extent the choice of travel demand forecast techniques impact on the estimated results. The urban rail demand forecast in the Tokyo Metropolitan Area will be used for the empirical analysis.
The paper focused on the following three factors that should be considered in the urban rail demand analysis particularly in the mega cities such as London, Paris and Tokyo: the in-vehicle congestion; the stochastic route choice; and the route choice set. First, it is often observed that the rail-use commuters suffer from the serious in-vehicle congestion in the morning peak hours. As the in-vehicle congestion varies among the rail routes, the rail-use commuters can choose the route by considering not only the travel time and the travel cost but also the in-vehicle congestion. When considering the in-vehicle congestion explicitly, we may need to consider the user?s equilibrium under which any route from an origin to a destination has the same utility level. Whether the equilibrium is considered or not may impact on the estimated demand. Second, the urban rail network in the mega cities is so complicated that it may make the rail users difficult to understand the network well. If it is assumed that they have the incomplete information, it may be better to apply the stochastic approach. We cannot judge the appropriateness of this assumption a priori. Third, the urban rail network in the mega cities is so dense that it enables the rail-users to choose the various alternative routes. When using the probabilistic?choice-based technique for demand analysis, we should define the individual choice set. We usually generate the choice set with the specific generation rule. What type of choice-set-generation rule is applied may affect the estimated demand.
We select the following four methods as the rail route choice models: the all-or-nothing (AON) assignment model, the multinominal logit (MNL) model, the user equilibrium (UE) model, and the stochastic user equilibrium (SUE) model. We compare them empirically with the completely same dataset.
The data used for the empirical analysis is the Tokyo Metropolitan Travel Census 2000. The rail network including the 1,877 zones, 4,850 nodes and 9,796 links in the Tokyo Metropolitan Area is used. First, we estimate the coefficients of the MNL model with the empirical data. Second, we define the link performance functions with the estimated coefficients. Third, we simulate the travel demand with the above-mentioned models. Then, we compare the fitness of the models among the methods by comparing the estimated link flows with the observed link flows. Fourth, we apply the models to a transport project impact analysis. The infrastructure improvement project along the Odakyu rail line will be used for the impact analysis. We compare the difference among the estimated travel demand among models.
The empirical analysis shows that the SUE has the best fitness. However, although the computation time of the SUE is more than ten times longer than the UE, the difference in fitness between SUE and UE is not so significant. The comparison of model fitness between the AON and the MNL shows that the influence of in-vehicle congestion is significant. The way of defining the rail choice set has also considerable effects the estimated demand.
This paper compared the travel demand analysis methods with the same empirical dataset. Any method compared in this paper is so popular among transport planners that we can expect the results of comparison will give us the useful information for better transport planning.
Association for European Transport