Online Ultra-Short-Term Wind Power Prediction

Ultra-short-term wind power prediction (UST WPP) is the key technique to guide the Power Grid to make planned dispatching for electricity on a daily basis and to maintain a steady operation for electricity trading market. Due to the fluctuating and intermittent characteristics of wind speed, the forecasting wind speed from numerical weather prediction (NWP) has large deviations from real wind speed, and it is difficult to meet the high demande of for UST WPP accuracy. To address this problem, this paper proposes an online wind speed revision model based on Kalman filter (KF) and multiple meteorological sources, and builds a wind-power conversion model based on light gradient boosting machine (LightGBM) algorithm to accelerate the calculating speed and to improve the accuracy of UST WPP. Based on the data analysis of a costal wind farm in Zhejiang province, the wind speed data after the revision of KF can effectively reflect the trend of real wind speed, so that the mean absolute error decreases dramatically. After inputting the revised wind speed into LightGBM model, the obtainted UST WPP gets very close to real power such that the root mean square error decrease from 16.36% to 5.64% after revision of KF, which means UST WPP accuracy improves effectively.