ABSTRACT
A third group of approaches uses other kinds of surrogate simulators, different than convolution equations. Examples are statistically derived regression equations, power functions, and arti‹cial neural networks (ANNs). Figure 9.1 details S-O model actions when a user provides statistical equations or power functions derived from existing data and the S-O model will not change these expressions. Figure 9.2 details S-O model actions when using a calibrated numerical simulator to provide data to train ANN ¡ow and transport simulators that are used with the optimizer. This ‹gure assumes no retraining, but Chapter 15 describes that process. Table 9.1 summarizes the presentation of these response surface approaches.
