Technical indicator enhanced ultra‐short‐term wind power forecasting based on long short‐term memory network combined XGBoost algorithm

Abstract The growing integration of renewable energy sources into the power grid has introduced unprecedented uncertainty. Ensuring an appropriately scheduled reserve is essential to accommodate renewable energy's intermittent and volatile nature. This study introduces an innovative approach to ultra‐short‐term wind power forecasting, which relies on feature engineering and a hybrid model. The effectiveness of this proposed method is showcased through a case study involving a utility‐scale wind farm in Inner Mongolia, China. The findings indicate that the hybrid model, which combines the XGBoost (Extreme Gradient Boosting) algorithm and LSTM (Long Short‐Term Memory) network with KDJ (Stochastic Oscillator), and MACD (Moving Average Convergence and Divergence), achieves the highest forecasting accuracy. Specifically, the proposed model yields a normalized mean absolute error of 0.0396 for wind power forecasting. The modelling and forecasting process takes approximately 550 s. Furthermore, the suggested method is employed to predict wind power and wind speed for a wind farm in the USA. The experimental results consistently indicate that the proposed model maintains a dependable performance across various raw datasets; it is suitable for use in power system operations.