Determining the Number of Simulations Required for Statistically Valid Results

Lift simulations use a random number generator to create a list of passengers based on anticipated passenger demand. Depending on the random number seed, different lists of passengers and resulting lift journeys will occur. Each random number seed scenario yields a different simulation run with different results. An infinite number of runs would yield results including a mean average waiting time and standard deviation which would be fully representative of the data. But only a finite number of runs can be completed as there are practical limitations on time and processing resources. How many runs need to be completed until the mean average waiting time can be said to be statistically valid? Different approaches to assessing the number of runs required for statistically valid results are proposed and discussed. The preferred approach allows the user to specify the required confidence level and acceptable range. The method can be applied to both dispatcher-based and Monte Carlo simulation.