Optimal DC motor speed controller design with reinforcement learning algorithm

In this study, a reinforcement learning based adaptive PID contoller is designed. The reinforcement learning based controllers are designed in the literature and it has been seen that they give successful results. In our study, the Q-Learning algorithm, which is one of the reinforcement learning methods, was used in the design of the PID controller, which is the most widely used controller structure. Q-Learning algorithm was applied in three different methods in our study. In the first method, an agent is created and agent can increase or decrease all of the PID parameters. In the second method, an agent is assigned for each PID parameter and each agent can increase or decrease the revelant PID parameter. In the third method, an agent using deep learning-based Q-Learning algorithm is created and can adjust each PID parameter. Controllers designed with the Q-Learning method gave as successful results as model-based PID controllers. The advantages and disadvantages of each method are examined.