Model Design – Desired and Undesired Model Characteristics and Effects
R. Russell Rhinehart · 2016
The user will choose a model functional form, its architecture. Perhaps it will have been derived from first principles and left in its primitive form, or converted into dimensionless group variables, or scaled variables by range or by standard deviation, or rearranged for computational convenience. Alternately, the user might have decided to use an empirical model, such as a power series, or some set of functional relations that seem to match the data trend, or a fuzzy logic (FL) structure, or a neural network model. This includes the number of neurons in the hidden layer of a network or the number of FL rules and membership categories. In any case, the user will desire that the in-use form of the model be convenient to use and provide a minimum of uncertainty. This chapter reveals issues related to model structure. It is an example from a pseudo-component modeling of stress-strain relationship in a film.