Ultra-Short-Term Wind Power Forecasting Based on the Strategy of “Dynamic Matching and Online Modeling”
Yuhao Li, Han Wang, Jie Yan, Chang Ge, Shuang Han, Yongqian Liu
Abstract
Ultra-short-term wind power forecasting plays a vital role in real-time scheduling, frequency regulation, and intraday market transactions. Due to the complexity of weather systems, unit aging, wind farm control strategies, etc., the temporal dependency relationship in wind power series changes from time to time (known as concept drift), which leads to the low forecasting accuracy of the commonly used offline modeling methods. Online modeling can effectively deal with concept drift by utilizing the latest information in the flow data and capturing the latest concepts during the modeling process. However, the existing online modeling methods cannot meet the timeliness requirements of the power grid for ultra-short-term wind power forecasting. Therefore, a strategy of “dynamic matching and online modeling” for ultra-short-term wind power forecasting is proposed in this paper. Training samples are dynamically selected according to the characteristic similarity of amplitude and fluctuation, aiming to improve the representativeness of samples and reduce the training time simultaneously. In addition to historical power, Numerical Weather Prediction wind speed is also introduced in the process of “dynamic matching” to improve the forecasting accuracy. Operation data from three wind farms in China is used to validate the effectiveness and robustness of the proposed method. The results show that the forecasting accuracy can be improved by 1.18%–4.32% for 4 hours in advance.