Algorithms for Synthesis of Adaptive Neural Network Control Systems Based on the Velocity Gradient Method

In this paper, algorithms for the synthesis of adaptive neural network control systems based on the velocity gradient method are considered. According to the method of analytical design of aggregated controllers, a control law is determined that transfers the representative point from an arbitrary initial state to a final state in the extended phase space. To adjust the weight coefficients, the control law's type and form are determined, ensuring the fulfillment of the limiting ratio, which guarantees the achievement of the target control by the system. The velocity gradient method is used in the adaptive control law in the combined form. The results obtained make it possible to carry out the first stage of the synthesis of adaptive neural network control systems, which consists in choosing the object output function in the state space of a nonlinear system used in neural network systems as generalized errors in the learning functions of multilayer neural networks.