Architectures of neural networks based on the physical characteristics of the object

Neural networks based on the physical characteristics of an object have been actively used in recent time in order to solve combined problems that require multi-step calculations. Therefore this technology shall be thoroughly studied and properly used. Neural networks solve many problems in different fields of science. Today there are several different ways of application, one of the possible applications is the solution of the equation of partial variables, in special cases the approximation of piecewise functional systems can be used. This article presents an overview of neural network architecture based on physical characteristics of the object. The application of physical informative neural networks (PINNs) for approximation of piecewise functional systems is considered. It is shown that PINNs allow to use information about the physical equations of the system in training, automatically determine boundaries between ranges of input variables, and use various activation functions and hidden layers.