Exploring the Choice of Experimental Design Used to Create the Training Set for a Reverse Neural Network Simulation Metamodel in System Design
In this study the use of reverse simulation metamodels as a decision support tool is explored. In reverse simulation metamodeling the outputs of the simulation model (performance measures) are used as the inputs to the metamodel and the metamodel approximates the inputs of the simulation (controllable factors). The focus of this study is the choice of the experimental design (D-optimal or orthogonal arrays) used to generate the data set used to create the reverse simulation metamodel was investigated using 36 simulation scenarios with different degrees of complexity. It was found that neural network metamodels trained using an orthogonal training set performed better than those trained using a D-optimal training set.