Instruments and Systems: Monitoring, Control, and Diagnostics Annotation << Back Gradient-Based Method for Neural Network Self-Learning Control of Multi-Loop Nonlinear Time-Varying Stochastically Disturbed System S.V. FROLOV, A.YU. POTLOV, A.A. KOROBOV, K.S. SAVINOVA A gradient-based method for neural network self-learning control of multi-loop nonlinear time-varying stochastically disturbed objects using a multilayer perceptron was proposed. The stability of the control system is achieved through the use of the regularization technique. A numerical example with two controlled parameters was given. A of multi-loop timevarying stochastically disturbed object was described by a system of nonlinear differential equations A neural network with 100 neurons in the inner layer was used. Neural network also characterized an activation function in the form of a hyperbolic tangent. The presented gradient-based method can be used for effi cient control of multi-loop time-varying stochastically disturbed objects in robotics, unmanned aircraft systems, energy and oil and gas processing complexes, bioengineering system and other complex dynamic systems. The results of the research are consistent with the conclusions about the unity and general approaches to control in living and technical systems was made in 1948 by the mathematician N. Wiener. Keywords: neural network self-learning control, gradient-based control method, time-varying system systems, stochastically disturbed system. DOI: 10.25791/pribor.5.2021.1262 Pp. 41-48.