REINFORCEMENT LEARNING IN CONTROL SYSTEMS OF OBJECTS WITH A TRANSPORT DELAY
V.S. Borovik1, S.V. Shidlovskiy1,2
1National Research Tomsk State University, Tomsk, Russia 2Tomsk Polytechnic University, Tomsk, Russia
Keywords: reinforcement learning, DDPG, control system, simulation, PID controller, formation of control actions, control under conditions of a lack of a priori information
Abstract
In this paper, we consider the possibility of using reinforcement learning systems for solving control problems under conditions of a lack of a priori information about the control object. The paper presents a solution to the problem of training the system by the Deep Deterministic Policy Gradient method for objects with a transport delay, as well as a comparison of the efficiency of the proposed solution with the classical method based on PID control, calculated using extended amplitude-phase-frequency characteristics and the Ziegler-Nichols method.
|