Priori-guided and data-driven hybrid model for wind power forecasting

Abstract

To overcome the high uncertainty and randomness of wind and enable the grid to optimize advance preparation, a priori-guided and data-driven hybrid method is proposed to provide accurate and reasonable wind power forecasting results. Fuzzy C-Means (FCM) clustering algorithm is used first to recognize the characteristics of the weather in different regions. Then, for the purpose of making full use of both priori information and collected measured data, a three-stage hierarchical framework is designed. First, via fuzzy inference and dimension reduction of Numerical Weather Prediction (NWP), more applicable wind speed information is obtained. Second, the accessible wind power generation patterns are served as a guide for mining the actual power curve. Third, the forecasted power is derived through the recorded data and the predictable wind conditions via data-driven model. This forecasting framework ingeniously introduces a gateway that can import priori knowledge to steer the iterative learning, thus possessing both adaptive learning ability and Volterra polynomial representation, and can present forecasted outcomes with robustness, accuracy and interpretability. Finally, a real-world dataset of a wind farm as well as an open source dataset are used to verify the performance of the proposed forecasting method. Results of the ablation analyses and comparative experiments demonstrate that the introduction of domain knowledge improves the forecasting performance.

Keywords: Explainable representation; Practical power curve; Priori-guided machine learning; Ultra-short-term forecasting; Wind power forecasting.