Ultra-Short-Term Wind Power Forecasting Based on Adaptive Time-Frequency Decoupling and Dual-Channel Fusion

To address the strong non-stationarity inherent in wind power generation sequences and the limitations of conventional deep learning models in capturing global linear evolution trends, this paper proposes an ultra-short-term wind power forecasting method based on Optimized Variational Mode Decomposition (OVMD) and a deep linear fusion network, termed WTG-FusionNet. OVMD is employed as the core time-frequency analysis technique to adaptively decompose the original non-stationary wind power series into multiple narrowband intrinsic mode functions (IMFs) under a variational constraint framework, enabling effective separation of long-term evolution trends and short-term fluctuations in the time-frequency domain. Based on the decomposed components, a master-slave parallel prediction architecture is constructed. The nonlinear branch integrates a multi-scale temporal convolutional network (MS-TCN), a bidirectional gated recurrent unit (BiGRU), and an improved self-attention mechanism (ISA) to comprehensively extract multi-scale temporal features. In parallel, the linear branch explicitly models the global linear trend through an autoregressive (AR) pathway, thereby compensating for the fitting bias of deep nonlinear models. The final prediction is obtained via a residual fusion mechanism that integrates the outputs of both branches. Experimental results on a wind power dataset from the State Grid Corporation of China demonstrate that the proposed model achieves a coefficient of determination (R2) of 0.94 on the test set, with the root mean square error (RMSE) and other evaluation metrics outperforming those of benchmark models. These results validate the effectiveness of the proposed fusion architecture in handling highly non-stationary wind power time series. This study provides both theoretical significance and practical value for improving the reliability and economic efficiency of renewable energy grid integration and dispatching.