The problem of optimal model selection often occurs in real problems of modeling because there can be several equally accurate models. For this case the method of model additional determination by additional bias criterion is proposed. At first the regularity criterion is calculated, then if it is impossible to select the optimal model, the additional bias criterion is calculated for the models that belong to the interval of uncertainty. The examples of model additional determination are shown in the selection problem of aircraft surface material.
The terminology and concept of metamodeling, examples of GMDH-based metamodels and ways to applying the principles of metamodeling for constructing models using the inductive approach are given. Metalearning is considered as a way to the adaptation of metamodels. An approach to the construction of a comprehensive inductive meamodeling system is proposed.
The problem of determining the quality of a cast product is studied by estimating the proximity of the experimental cooling curve to one of the specified reference curves. The paper compares several approaches to finding the distance between curves. The authors propose a new approach based on calculating the mean difference between the coefficients of functions approximating experimental and reference curves. Its effectiveness and reliability is shown on a case study.