| The task of simulation input modeling is arguably one of the most important aspects of a simulation study. The process of building the overall structure of a simulation model is crucial, but is also more intuitive. The actual task of input modeling is certainly less intuitive and traditional methodologies can be repetitive and time consuming. Some exploratory work in using neural networks to assist in this effort has been performed, but is restricted to the realm of classification to trained categories. The thrust of this research is the development of a neural network system that first selects a probability distribution family based on shape, and then determines the parameters of the distribution(s) with an interpolative capability, which reduces the training effort from exhaustively learning each distribution with an infinite range of parameter values. |
