Development of a methodology for the use of neural networks and simulation modeling in system design

In this paper the use of metamodels to approximate the reverse of simulation models is explored. This purpose of the approach is to achieve the opposite of what a simulation model can do. That is, given a set of desired performance measures, the metamodels output a design to meet management goals. The performance of several neural network simulation metamodels was compared to the performance of a stepwise regression metamodel in terms of accuracy. It was found that in most cases, neural network metamodels outperform the regression metamodel. It was also found that a modular neural network performed the best in terms of minimizing the error of prediction.