Wind Power Forecasting: LSTM-Combined Deep Reinforcement Learning Approach

With the high penetration of renewable energies connected to smart grids, accurate wind power forecasting becomes more and more crucial to cope with its intermittency to achieve power balance. However, traditional forecasting methods face several challenges from model complexity, prediction delay, and insufficient training data. To address the mentioned challenges, we propose a forecasting scheme that combines long short-term memory(LSTM) and reinforcement learning (RL), to achieve accurate forecasting of short-term wind power. First, we train an LSTM model based on historical data as the basic model, which can effectively improve the input information accuracy. Secondly, we utilize the wind power output predicted by LSTM to form the state space in RL. Meanwhile, a well-designed reward function is proposed to guide the RL agent's training, which does not need the pre-prepared data and model knowledge. Finally, the case studies verify the superiority of the proposed method in terms of wind power forecasting accuracy.