Abstract
In these days, many people introduce their solar panels on their roof and sell the electricity generated to the electric power company. To keep the quality of the electric power, the electric power company might want to control the magnitude of the generated power from the solar panels according to the current demand. In this paper, we assume that the electric power company will change the buying price to realize the control in the near future. Under such circumstances, we should controlling the selling power according to the buying price, the battery charging ratio, the current generated power and so on. There are several learning methods to construct the intelligent trading system. Almost all of them, however, need a huge number of iterations to complete the learning. To speed up the learning, we propose a novel "Smart-grid-game" system that accelerates the learning speed of the reinforcement learning for the control. Therefore, the system monitors the game players’ behaviors and use the log data for supervising an actor-critic model.
