Features of Synthesis of Neural Network Simulators with Limiters

One of the problems of the modern theory of automatic control is the construction of verified mathematical models of objects of regulation. In some cases, obtaining such models is not possible. This paper discusses approaches to the use of neural networks as regulators. Due to the presence of the adaptivity property, neuroregulators in some cases show a better quality of control in comparison with classical ones. The most common architectures of neural networks and methods of their training are considered. A comparative analysis was carried out. The first stage in the design of neural network regulators is the creation of a neuroimitator. An example of the synthesis of a neural network simulator is given. The proposed scheme has been worked out on systems with saturation type limiters and a mechanical stop.