A Novel Encoder–Decoder Model for Short-Term Multistep Wind Power Prediction

Wind power has emerged as a vital renewable energy source. However, its inherent temporal variability and non-stationarity pose significant challenges for accurate forecasting. To solve this problem, this study proposes a novel multi-step wind power forecasting model based on an encoder-decoder architecture. It incorporates a multi-frequency attention mechanism into a multi-layer long short-term memory (LSTM) network. The former increases the ability of the model to capture long-term dependencies and global features, while the later focuses on short-term internal dynamics. Together, they allow the model to accurately extract critical frequency information across various time scales, and adaptively emphasize key features within the temporal domain. The experimental results show that the proposed model outperforms different benchmark methods in terms of many performance indicators, including the mean absolute error, mean squared error, and root mean squared error. This demonstrates its high potential to provide technical support for the precise wind power forecasting, and to contribute to the efficient integration of wind energy into power grids.