A Short-Term Wind Power Forecasting Method Based on STL-TimesNet-iTransformer Model

High-accuracy short-term wind power prediction is crucial for ensuring the security of smart grids. Existing wind power prediction models still have significant limitations in addressing the inherent intermittent, random, and multi-scale characteristics of wind energy. Therefore, this paper proposes a novel hybrid prediction framework, STL-TimesNet-iTransformer. First, Seasonal Trend Decomposition (STL) is used to decompose the original power data into low-redundancy components, suppressing non-stationary interference. Then, the framework concurrently integrates the periodic feature extraction capability of TimesNet and the long-range dependency capturing mechanism of iTransformer, creating a dual-path architecture to simultaneously analyze local fluctuation patterns and global evolutionary trends. Experimental results demonstrate that this model accurately fits the dynamic patterns of the power generation sequence, achieving the highest coefficient of determination and the lowest mean absolute error (MAE) and root mean squared error (RMSE), thus validating the superiority of the proposed method.