Wind Power Forecasting Research Based on Autoformer
Yu Hua, Xin Guan, Haoran Xu, Chenhao Zhao · 2024
To address the complexity and uncertainty in wind power forecasting and to enhance prediction accuracy and stability, this study proposes a wind power forecasting model based on Autoformer. By leveraging its self-attention mechanism, the model effectively captures the trend and periodic characteristics inherent in wind power data. The model's performance was validated using data from Wind Farm A in Liaoning Province, comparing its forecasting accuracy against LSTM, Informer, Transformer, and Autoformer models. Prediction outcomes were evaluated using mean squared error (MSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and coefficient of determination (R2) as metrics. Experimental results demonstrate that the Autoformer model outperformed others across all evaluation metrics, with the predicted curve closely aligning with actual values. Compared to other models, Autoformer achieved significantly lower MSE, MAE, and MAPE values, and a higher R2value, proving its substantial advantage in capturing dynamic power fluctuations and enhancing prediction precision in wind power forecasting. Future research may explore multi-model integration to further improve forecast stability and applicability.