Artificial neural network metamodels of stochastic computer simulations

A computer simulation model can be thought of as a relation which connects input parameters to output measures. Since these models can become computationally expensive in terms of processing time and/or memory requirements, there are many reasons why it would be beneficial to be able to approximate these models in a computationally expedient manner. This research examines the use of artificial neural networks (ANN), to develop a metamodel of computer simulations. The development and use of the Baseline ANN Metamodel Approach is provided and is shown to outperform traditional regression approaches. The results provide a solid foundation and methodological direction for developing ANN metamodels to perform complex tasks such as simulation 'optimization', sensitivity analysis, and simulation aggregation/reduction.