Deep Reinforcement Learning Based Energy Storage Management Strategy Considering Prediction Intervals of Wind Power

Abstract

Wind power generation development needs new methods to maintain the energy balance and stable operation by using energy storage management strategies. Most of the available storage control strategies do not take into account wind power uncertainty, which can be quantified by prediction intervals of wind power. This paper presents a data-driven energy storage management strategy considering prediction intervals of wind power for wind power microgrid operations. On the one hand, a power interval prediction model based on long short-term memory (LSTM) and lower upper bound estimation (LUBE) is established to quantify the uncertainty of wind power forecasting. On the other hand, the energy storage management problem is formulated as a Markov decision process (MDP), which is solved by deep reinforcement learning. According to the real-time state such as wind power, power prediction intervals, local load, dynamic electricity price and state of charge (SOC), the proposed strategy can make the charge/discharge schedule automatically. Effectiveness of the proposed energy storage management strategy is demonstrated on real wind farm dataset.