Optimized Deep Reinforcement Learning Approach for Dynamic System

Reinforcement learning methods provide significant and impressive improvements in artificial intelligence studies in recent years, especially in Atari and Go games. This development attracts the attention of scientists who try to understand how people learn. In addition, the biggest advantage of reinforcement learning over other learning algorithms is that it does not require any prior data. This feature distinguishes Reinforcement Learning from others. Reinforcement Learning is an approach in which smart programs work in a certain or uncertain environment to constantly adapt and learn based on scoring. The feedback process is also known as a reward or called a penalty. Given Agents and environment, it is determined which action to take. In this study, we apply the success of Deep Q-Learning (DQL) algorithm, one of the model-free based deep reinforcement learning algorithms used in the literature, on the CartPole problem. We also offer a different method to improve the agent's success during the training phase. We present an optimized DQL algorithm using a function that updates the weights of the neural network at every step.