Determining the Number of Traffic Micro-Simulation Runs Required to Estimate Travel Time Reliability

There is increasing interest in quantifying travel time reliability (TTR) for thepurposes of evaluating the benefits of different road transportation improvementalternatives. Generally two approaches for quantifying TTR have been describedin the literature, namely: (1) empirically based statistical models and (2)simulation models. The use of simulation model modes to estimate TTR requiresrepetitively sampling from a set of input parameter distributions to capture therange of variation in supply and demand expected in the real-world. The primaryadvantage of the simulation models is the ability to make TTR estimates for avariety of improvement alternatives, including new technologies or operationalpolicies for which field data are not available. However, one of the challenges inusing simulation models is determining how many samples must be taken fromthe input distributions and thereby the simulation runs that must be conducted. Atypical approach is to use stopping or convergence criteria. In this paper weshow that this approach can be susceptible to problems of local convergence andwe show that the use of Latin hypercube sampling (LHS) rather than randomsampling can mitigate this problem and reduce the number of simulation runsrequired.

Keywords:

Travel Time Reliability , Microsimulation , Monte Carlo Simulation , Latin Hypercube Sampling