Accounting For Demand Simulation Random Variations In Project-Level Performance Assessment
Corresponding Author: Garett Ballard-Rosa, Sacramento Area Council of Governments
Presented By: Garett Ballard-Rosa and Shengyi Gao, Sacramento Area Council of Governments
The shift to activity-based microsimulation in travel demand modeling supports a transportation planning approach responsive to a range of complex policy questions. Less understood from this shift, however, is the effect of random variation in demand simulation models. In particular, person level simulations require a process to allocate choice at the person level, often by assigning a random seed to each possible outcome for each person. This seed variable is a unique characteristic of simulation models, leading to variation in results from the same input files and processing where random seeds vary from one run to the next.
The primary objective of this research is to explore the effect of random variation in demand simulation modeling, using the DAYSIM sub-model of the Sacramento Area Council of Governments (SACOG) as a case study. DAYSIM is a micro-simulation of household-generated travel that uses a random seed to determine the order of simulation of person-level activities. This research’s assessment of random variation in DAYSIM has progressed on two tracks. First, to develop parameters on changes from random variation by analyzing network indicators such as link-level volumes across multiple model runs where random seeds vary. Second, the effect of random variation on project performance assessment. The work finds random variation of ‘big number’ travel metrics (such as person trips or VMT) to be very small at the full network level, but more pronounced for smaller number metrics and geographies. In particular, the research examines how random variation can skew the assessment of individual transportation projects, such as through a benefit-cost analysis. While the work focuses on a single agency, the findings should be relevant to other users of demand simulation models.