Development of a Reference Signal Self-Organizing Control System Based on Deep Reinforcement Learning
Hiromichi Iwasaki, Atsushi Okuyama · 2021
Intelligent control has received a significant amount of attention in recent years owing to its use in autonomous driving technology and other applications(1)-(3). Intelligent control is a control theory that uses machine learning algorithms to build control systems. In this study, we develop an intelligent control theory based on deep reinforcement learning. We proposed a reference signal self-organizing control system based on a deep deterministic policy gradient (DDPG). This proposed system is an extension of an existing control system using DDPG. We verify the effectiveness of the proposed system through swing-up and stabilizing control simulations using an inverted pendulum with an inertia rotor. We confirmed that the pendulum was inverted by the swing-up control at approximately 1.2 s and the pendulum was stabilized for approximately 8.8 s. Therefore, we confirmed the effectiveness of the proposed reference signal self-organizing control system.