SCADA Data Based Wind Power Interval Prediction Using LUBE-Based Deep Residual Networks

Abstract

Wind is a pollution-free renewable energy source. It has attracted increasing attention owing to the decarbonization of electricity generation. However, owing to the dynamic nature of wind speed, ensuring a stable supply of wind energy to electric grid networks is challenging. Therefore, accurate short-term forecasting of wind power prediction plays a key role for wind farm engineers. With the boom in AI technologies, deep-learning-based forecasting models have demonstrated superior performance in wind power forecasting. This paper proposes a short-term deep-learning-based interval prediction algorithm for forecasting short-term wind power generation in wind farms. The proposed approach combines the lower upper bound estimation (LUBE) method and a deep residual network (DRN). Wind farm data collected in northwestern China are selected for this empirical study. The proposed approach is compared with three benchmark short-term forecasting approaches. Extensive experiments conducted on the data collected from five wind turbines in 2021 indicate that the proposed algorithm is efficient, stable, and reliable.