System and Method for Reinforcement Learning Supporting Delayed Rewards

摘要

The present invention relates to a reinforcement learning method for supporting delay compensation in a reinforcement learning system. The reinforcement learning method using the reinforcement learning agent of the present invention comprises the steps of receiving an immediate compensation value and a delay compensation value associated with a control action from an environmental system, and taking into account the received immediate compensation value and the delay compensation value to control the Generating a final reward value corresponding to the action, generating a transition tuple including the final reward value, and applying the generated transition tuple to the reinforcement learning agent to perform reinforcement learning. According to an embodiment of the present invention, since the delay compensation value measured by being delayed in the environmental system can be applied to the directly related control action, the performance and speed of the reinforcement learning system can be increased.