Energy Scheduling for Multi-Energy Systems via Deep Reinforcement Learning

With the development of smart infrastructures, especially energy hubs (EHs), traditional power systems transform into the multi-energy systems. This paper investigates a long term profit maximizing energy scheduling problem for multi-energy systems from the perspective of prosumers. Most existing methods assume that future market prices or demand information of prosumers are known to the decision makers. In this paper, we model the multi-energy scheduling strategy in the presence of unknown information as a Markov Decision Process (MDP) problem. We first establish an energy scheduling mechanism by exploring the unique features of EHs. The concept of valid actions is then proposed to ensure the balance between supply and demand. A deep Q-learning algorithm is developed to obtain the scheduling strategy without any prior information. Simulation results demonstrate the effectiveness and efficiency of the proposed strategy.