In the practice of urban planning, travel demand forecasts are generally obtainedby using the four-step model (generation, distribution, modal split and assignment), despite considerable theoretical advances in the field. This persistence can be explained by the relative ease of implementation of the four-step modelling sequence, which is related, in particular, to the kind of data available that could be used as an input in a model. However, the static nature of the approach raises questions as it pertains to its relevance in producing medium and long range forecasts. This thesis investigates the validity of the hypothesis of temporal stability of the parameters of the first three stages of the traditional forecasting sequence. To do this, taking the Lyon conurbation as our case study, we coded the road and transit networks at different points in time (1985, 1995 and 2006). We then combine this temporal data, which is generally unavailable, with the corresponding household travel surveys in order to calibrate the first three steps of the traditional model and test their predictive ability. For the generation models tested, we note acceptable performance for a 10-year forecast. For a 20-year forecast, some changes in lifestyles have resulted in a decrease in the average number of work trips that traditional models do not predict accurately. Regarding trip distribution, the increase in travel distances observed for certain purposes is reproduced fairly well by the gravity model. At the modal split level, the parameters are not stable and the estimated models would be unable to predict accurately the recent increase in the market share of public transport.
