Statistical Approach for Validating Simulation Models by Comparison with Operational Systems-Illustrated for Traffic Flow

摘要

Development of computerized simulations including random elements is sometimes desired for types of s'stems that are operational for special cases. Then, the suitability of a prospective simulation model can be checked for these special cases. That is, some of the important kinds of properties for these systems can be compared for agreement. The simulation model is unsatisfactory when there is too much disagreement. The first step is to decide on the properties to be compared. Suitable standardization is important for obtaining comparisons that are both feasible and valid. The next step is to quantitatively determine how decisions on comparisons would be made if the properties considered, or their representative values (for example, expected values), were exactly determined. Finally, statistical methods are developed for making these comparisons on the basis of observed values for these properties. When the simulation is of a random nature but the cost of a run is relatively small, expected values of the properties can be almost exactly determined for the simulation model. This is accomplished by repeating simulations a very large number of times. Nonparametric methods, including asymptotic considerations, furnish the basis for the statistical methods. An example pertaining to traffic flow is used to illustrate the overall approach. (Author)