Multi-Step Prediction Method for Wind Power: A Framework Integrating CNN–RNN–LGBM Models

Abstract

Wind power prediction plays a significant role in enhancing the effectiveness of power system operation and decision-making. Given the inherent stochastic nature of meteorological events, achieving highly accurate forecasts for wind power poses considerable challenges. To address this challenge, this paper initially leverages the time series learning capability of recurrent neural networks (RNN) to extract sequential information from historical wind power data. Subsequently, the information extracted from the convolutional layer is transferred to the light gradient boosting machine (LGBM), utilizing the feature extraction capability of convolutional neural networks (CNN). Furthermore, an optimal weighted combination is employed for the short-term prediction of wind power. Finally, a multi-step wind power prediction method of integrated CNN–RNN–LGBM is proposed in this paper. Simulation results demonstrate that the proposed CNN–RNN–LGBM framework outperforms other models during global training. Meanwhile, transferring the information from CNN to LGBM can improve its performance, proving the feature extraction ability of CNN.