A Comparison of Sample Enumeration and Stochastic Microsimulation for Urban Travel Demand Forecasting
BRADLEY M, Independent Consultant and LAWTON T K, Portland Metro, USA
The paper compares two travel demand forecasting approaches that have been applied to forecast travel demand in and around Portland, Oregon. Both approaches apply a system of discrete choice models to a synthetic sample of households, drawn to match the c
The paper compares two travel demand forecasting approaches that have been applied to forecast travel demand in and around Portland, Oregon. Both approaches apply a system of discrete choice models to a synthetic sample of households, drawn to match the characteristics of the actual or forecast population. The system consists of a model to predict a 111 day's schedule of tours (by purpose and trip chain type), a model to predict the times of day at which each tour begins and ends, and a model to predict the locations of all out of home activities and the modes used to reach them The accessibility of travel by all modes and locations also appears in the upper level models to influence the number of tours and stops made at various times of day.
Rather than focusing on the models themselves, the paper contrasts two methods of applying them. In sample enumeration, the probabilities across all possible alternatives are added across all individuals in the sample. The output consists of ongirddestination tour and trip matrices, segmented by purpose, mode, time of day and income class. In stochastic microsimulation, the probabilities are used in a Monte Carlo fashion to predict a single set of tours, destinations and modes for each individual in the sample. Thus, all household and person characteristics can be tied back to each individual trip record so that the output resembles a "synthetic travel diary survey'.
The paper provides various comparisons of the results using the two approaches. A major advantage of the microsimulation method is that much more detail can be retained in the output. The simulated trip records can be aggregated in a flexible way to create trip matrices, perform equity analyses' or input to dynamic traffic simulations. The microsimulation framework also provides more flexibility in applying specific sub-models. For example, models for work-based tours and intermediate stop locations had to be applied in an aggregate manner in the sample enumeration, but could be applied disaggregately with much less processing time in the microsimulation. Disadvantages of the microsimulation approach can include less spatial coverage across possible OD pairs? and less stability in results due to the use of random draws. Both of these disadvantages, however, can be counteracted by using much larger synthetic samples. In Portland, it is possible to simulate the entire Portland population (1.4 million person records) with the microsimulation approach in less run time than is required to run just one tenth as many records in the sample enumeration.
Association for European Transport