Reinforcement Learning Based Energy Management for Public Buildings Considering Multiple Energy Resources

The combination of expert knowledge and data-driven intelligent algorithms is of great significance for the optimal operation of public building energy supply systems. This paper proposes an adversarial generation strategy learning algorithm that can implicitly learn expert experience to solve the economic scheduling problem of public building energy supply systems in real time. Considering the interaction between electric vehicles and the power grid, an economic dispatch model for the public building energy supply system was exploited. A strategy network and an evaluation network were established, where the actions output by the strategy network were treated as apprentices' actions. The strategy network and evaluation network were alternately trained with the goal of maximizing the score of the strategy network's output actions and accurately distinguishing between expert and apprentice actions. Finally, the proposed solution is assessed through experiments and the effectiveness is confirmed.