On the State-of-the-art Demand Forecasting Model Developed by Netherlands Railways
B de Vries, Netherlands Railways, NL; J Willigers, Significance, NL
Dutch Railways (NS) and Significance have developed a new model system to provide train travel demand forecasts. The design of the system provides the necessary technical abilities, while still meeting user requirements such as transparency.
Dutch Railways (NS) and Significance have developed a new model system to provide train travel demand forecasts for NS's strategic decision making. The model is among others used for the evaluation of timetable alternatives, forecasting of volumes of new stations, evaluation of tariff policies, all under different exogenous scenarios for demographic, economic and car-related developments.
This paper focuses on the design of this model system from two, seemingly conflicting, perspectives.
The first perspective is about the user requirements which were formulated and relate to how the model should work in practice, such as:
1) the model should have a clear structure and be transparent in how its results have been calculated,
2) input data and model coefficients should be up-to-date,
3) the model should be able to interact and be consistent with other, complementary models currently in use by NS, and
4) the model should be user friendly and easy to maintain.
These user requirements were derived from a thorough investigation within NS among users of forecasts in a wide range of disciplines. The bottom line from these interviews was that forecasts, even based on complex models, should be in a way easy to understand and made plausible. This transparency is the most important condition for forecasts really to be used by the decision makers.
Hence from the second perspective the model system has to have sufficient technical detail, complexity and flexibility to provide passenger forecasts for different scenarios, time horizons and study areas and to be adequate in its forecasting precision. It is this trade-off between easy-to-understand forecasts and the necessary complexity of the model to be adequate in its forecasting results which has been the main challenge of this project. As a solution a modular system has been developed to fulfill all these requirements, while finding a trade-off point where both perspectives might oppose each other.
Significance contributed in developing a new module to the model system for forecasting demand volumes per station pair for different exogenous scenarios. The module consists of four sub-modules, each with a clear function and output that can be viewed and confirmed. The first sub-module uses an exogenous scenario of demographic, economic and car-related variables to create a year-by-year dataset on the municipal level. A station assignment sub-module forms a station level dataset of these same variables. An elasticity sub-module uses growth factors for scenario and time table developments to forecast demand volumes between station pairs, separately for each of six travel purposes. And a final sub-module forecasts demand volumes for newly opened stations. All sub-modules have been estimated and validated by time series and cross-sectional data.
The model system's design has been greatly influenced by the precondition of transparency: the contribution of separate modules in the final forecasting result can be easily visualised. This has contributed to the understanding and acceptance of the model results by the decision makers. Yet the model system is capable of modelling complex processes, such as competition between stations and lagged response to time table changes. The year-by-year forecasts thereby provide insights into the timing and development of dynamic processes. After completion of the first version of the model system in 2009, it has been used for a wide range of forecasting projects, varying from infrastructural studies for Dutch government to minor timetable studies for Dutch Railways.
Association for European Transport