Optimal computing budget allocation in selecting the best design via discrete event simulation

The methodology based on computing budget allocation is an effective tool in solving the problem of selecting the best design out of a group of stochastic systems via simulation. It can intelligently determine the best simulation lengths for all simulation experiment and thus significantly reduce the total computation cost to obtain the same confidence level. After a comprehensive review of previous works, this dissertation extends the previous efforts to investigate improved technologies which provided very different view of how to efficiently allocate the computing budget. Not only can the new approaches fix some unsolved difficulty in the previous works, but also improve the speed of the earlier approaches. Moreover, comparing with other methods including two-stage procedures, and the other sequential procedures in some specific problems, the numerical experiments show that all of our approaches are superior to the compared methods with various same confidence levels.