Short-term Wind Power Prediction Based on CEEMDAN De-composition and Spatiotemporal Feature Fusion

Due to the randomness of wind power generation, the output power of a wind farm will fluctuate. As a result, the power grid can face a need for increased reserve capacity, increasing scheduling difficulties, and wind farm abandonment. An effective way to address these problems is to accurately predict the output power of wind farms. Traditional prediction methods usually make predictions based on wind data obtained at a single height. However, with long prediction periods, prediction errors are relatively large because the wind speed and direction at different heights have spatiotemporal correlations within a wind farm. In the model presented here, the wind power data are first decomposed by a “complete ensemble empirical mode decomposition with adaptive noise” model to obtain modal components with different fluctuation characteristics. Then, the characteristics of wind speed, wind direction, air pressure, and other data at different heights are extracted for spatiotemporal feature fusion. Actual measurement data from a wind farm in Chongqing are used to verify the feasibility and effectiveness of the proposed method.