Application of Neural Networks in Control Systems with Limiters

This paper discusses a method for developing neural network controllers that can be used in systems with limiters when there is no information about the plant to build the necessary mathematical model. Artificial neural networks can be used to develop such controllers. To synthesize a neural network controller, reinforcement learning principles are applied using a neural network simulator of the plant. This approach is effective when classical methods of control system synthesis cannot be applied due to significant nonlinearity and inability to create an accurate mathematical model of the plant. The proposed methods allow us to expand the class of technical systems for which direct synthesis of control laws close to optimal ones is possible. Neural network controllers also provide robustness and adaptability, which makes them attractive for practical application. The paper discusses in detail the choice of structure and parameters of neural networks in the simulator and controller, as well as the formation of training samples taking into account the constraints in the plants.