An application of variance reduction techniques in freeway simulation

Computer simulation models are used in a variety of applications in transportation engineering and have become a prime aid in decision making. The applications range from evaluating traffic control strategies for single intersections to such complex decision processes as evaluating the impact of removing toll facilities at the George Washington Bridge in New York City. While it is widely accepted that simulation offers an unmatchable capability of evaluating alternate control policies, the high variance of the output variable presents a critical problem in such comparative analyses. The simulation models with high output variance must be run longer or replicated many times to achieve a desired precision level, and that corresponds to increased cost of computer resources. This paper describes and illustrates the application of variance reduction concepts that can improve the reliability and efficiency of the simulation experimental process by taking advantage of the simulation model structure. The two variance reduction concepts (common random numbers and antithetic variates) reduce the variance of the output variable by replacing the original sampling procedure with a new procedure that yields the same expected value but with a smaller variance. The application of the variance reduction concept was illustrated using results from experiments with a freeway simulation model. The results indicate that both common random numbers and antithetic variates sampling procedures appreciably reduce the variance of the simulation output measure.