On the use of common random numbers in activity-based travel demand modeling for senario comparison
Activity-based travel demand modeling (ABM) provides a very high level of detail when modeling complex travel behavior of individuals. Since stochastic simulation is used, however, the high level of detail may induce a large random fluctuation in the output, necessitating (possibly) excessively many model reruns to produce reliable output. This may especially become prohibitive in terms of computation time when comparing the travel behavior between multiple scenarios, in which case each scenario would require its own simulation. To alleviate this issue, in this paper, we propose the use of common random numbers, which is a technique that reuses the same random numbers for the choices made by the travelers between scenarios. This technique ensures that any observed difference in model output across scenarios cannot be attributed to any mutual difference in drawn random numbers, eliminating an important source of random fluctuation. In particular, we show how to implement this technique in an activity-based travel demand model. Furthermore, we demonstrate by means of a numerical study, based on travel mode choice in the metropolitan region Rotterdam-The Hague in the Netherlands, that the technique of common random numbers can greatly reduce the number of runs needed to obtain output with a random fluctuation that is small enough to draw reliable conclusions. In turn, this also brings down the required computation time significantly.
To reference this document use:
Common random numbers
Mobility & Logistics
TRB 101st Annual Meeting of the Transportation Research Board, 1-18