MACHINE LEARNING OF CONTROL SYSTEMS WITH FEEDBACK BASED ON THE PRINCIPLE OF SYNTHESIZED OPTIMAL CONTROL

• Published in: Надежность и качество сложных систем no. 4
• Main Author: Shmalko, Elizaveta
• Format: Journal Article
• Language: English
• Published: Penza State University Publishing House 01-01-2024
• Subjects: machine learning; optimal control; quadcopter; synthesis
• ISSN: 2307-4205
• DOI: 10.21685/2307-4205-2023-4-2

Background. In an effort to automate various life processes to improve their quality, the need to automate the development of control systems becomes obvious to make it fast and universal. This sounds especially relevant in the context of ever-increasing robotization and the emergence of various robots as control objects. The most common task of robotics is the synthesis of feedback control. It assumes that the control system that ensures the achievement of the goal by the object is designed, depending on the state of the object, optimally according to specified criteria. The task of synthesis is relevant, but there are no general approaches to its solution today. In this paper, an inverse approach is proposed to the synthesis of an optimal feedback control system based on machine learning methods to obtain realizable solutions to the optimal control problem. Materials and methods. The paper presents the principle of synthesized optimal control. The general idea is as follows. First, the object is stabilized with respect to some point in the state space, through the solution of the problem of synthesis of the stabilization system. Adding a stabilization system to the object model gives it a new property: at each moment of time, the object has a point of equilibrium. Near the equilibrium point, all solutions converge. Thus, the problem of optimal control is solved through the optimal position of the equilibrium point. Results. Substantiations are given and the principle of synthesized optimal control is formulated, which includes the stage of synthesis of the stabilization system. The implementation of the quadrocopter control system based on the principle of synthesized optimal control is presented. Conclusions. When solving the problem of optimal control, it is necessary to additionally ensure the movement of the object along the obtained trajectory to compensate for possible constantly existing uncertainties. In the presented synthesized optimal control approach, the uncertainty is compensated by the stability of the system with respect to a point in the state space. The approach is universal and is not limited to certain types of control object models or control quality functionals. It can be argued that this approach is machine learning of control systems with feedback.