Acceleration of reinforcement learning via game-based renewal energy management system

Renewal energy generation has improved greatly in the last decade despite its limitations; e.g., the generated power varies depending on weather conditions. As an example, the electricity generated by a photovoltaic system varies based on irradiation and temperature. The variation in the generated power remains a serious problem. To solve this problem, we usually need to use a rechargeable battery to stabilize the generated power. However, high skills are required to meet the conditions for a smooth operation of such batteries. Using a reinforcement learning method to acquire such skills might require a long learning period. To overcome this problem, we introduce herein a new scheme; a smart grid game-based skill acquisition system. In this system, human skills are corrected via a game and such imitated them by using a supervised actor-critic method. As a result, our system acquire a high operation skill about one-year of operation on the smart grid game whereas the conventional approach takes more than two years to obtain such skills.