Forming Set of Factor Values to Remodel Dynamic Objects

The paper considers the problem of forming an array of factors values for remodeling (which is understood as constructing the model based on existing model). This problem is close to the design of experiment (choosing the optimal plan). The main difference is that the experiment is carried out on the existing model. In case the model is not stochastic, there is no need to repeat the experiment (to calculate responses based on the existing model). Classes of initial models can vary significantly, it affects the procedure of forming an initial factors values array for the remodeling. The problem of dynamic systems with variable structure remodeling is under investigation. Neural networks are used as a remodeling class. The suggested algorithm to select initial values is based on the methods of planning experiments for nonlinear models. As an example, the model of the inertial torque transformer (ITT) is considered. The model is a composite system of nonlinear differential equations.