Two-Step Wind Power Prediction Approach With Improved Complementary Ensemble Empirical Mode Decomposition and Reinforcement Learning

The strong stochastic nature of wind power generation makes it extremely challenging to accurately predict and support the planning and operation of modern power systems with significant penetration of renewable energy. This article proposes a two-step wind power prediction method, which consists of two phases: long time-scale coarse prediction and short time-scale fine correction. In the long time-scale phase, a complementary ensemble empirical mode decomposition-based sigma point Kalman filter approach is proposed to coarsely predict wind power merely with historical data. In the short time-scale phase, a deep deterministic policy gradient approach learns from real-time weather information to correct the coarse prediction result, which results in an improved prediction accuracy. A real-life case study confirms that the proposed method can properly predict wind power generation and have a better prediction accuracy than existing techniques, thus offering a viable and promising alternative for predicting wind power generation.

Automated summary

This article proposes a two-step wind power prediction method, which involves coarse prediction at a long time-scale and fine correction at a short time-scale, ultimately improving prediction accuracy with the guidance of real-time weather information.