Dynamic Energy Management for IoT-enabled Smart Microgrid using Deep Reinforcement Learning

Smart Microgrid with distributed energy resources is prevalent due to its flexibility and adaptability. Benefiting from Internet of Things technology, massive real-time energy data can be gathered for intelligent energy management. However, the high stochasticity of renewable energy resources makes it difficult to design an optimal scheduling strategy for smart microgrid. To address this problem, a dynamic energy management strategy based on model-free reinforcement leaning is proposed. Considering the uncertainties of renewable resources and energy demands, the dynamic energy management problem is formulated as a Markov decision process with unknown transition probabilities. Then, an intelligent energy management strategy based on soft actor-critic is proposed. Moreover, gate recurrent unit is incorporated to enhance the extraction of time-series characteristics. Finally, simulation results demonstrate that the proposed method significantly reduces the energy bills by up to 24.28%, compared with other strategies.