A Dual Decomposition and Dynamic Weighting-Based Integrated Forecasting Method for Ultra-Short-Term Wind Power

Accurate wind power forecasting is crucial for the safe scheduling, stable operation, and economic benefits of the power grid. However, the volatility and randomness of wind power present significant challenges in developing high-precision forecasting models. This study proposes an ultra-short-term wind power integrated forecasting model based on dual decomposition and intelligent optimization algorithms. The model first employs seasonal-trend (STL) decomposition to decompose the wind power time series into long-term trends, seasonal components, and residuals, revealing the multi-scale characteristics of the data. Then, various forecasting models are applied to model each decomposed component in order to capture their distinct characteristics. Subsequently, the Stacking method is used to integrate the predictions of these models, with linear regression (LR) serving as the meta-learner to combine the results. An intelligent optimization algorithm is introduced to tune the model parameters, thereby enhancing the forecasting performance. Additionally, to address the error between the predicted and actual values, variational mode decomposition (VMD) is applied to further decompose the errors, and a long short-term memory (LSTM)network is employed for dynamic correction, thus improving the final prediction accuracy. Experimental results on data from a wind farm in Xinjiang, China, show that compared to traditional methods and other advanced techniques, the proposed model demonstrates significant advantages in forecasting accuracy, stability, and handling the nonlinear and non-stationary characteristics of wind power.