A Novel Framework for Wind Power Prediction Based on Optimized Variational Mode Decomposition and Temporal Convolutional Network With Hybrid Attention

Ensuring accurate and reliable wind power prediction is paramount for efficient scheduling and safe operation of power systems. However, the inherent instability of wind poses challenges for traditional prediction models, often resulting in low accuracy and inadequate data fitting. To overcome these hurdles, a novel hybrid approach is proposed in this study. Initially, the northern goshawk optimization algorithm (NGO) is employed to optimize the parameters of variational mode decomposition (VMD), thereby enhancing the decomposition performance and mitigating parameter related uncertainties. Throughout the optimization process, fuzzy entropy (FE) serves as the fitness function. Subsequently, a hybrid attention temporal convolutional network (HATCN) is proposed to extract deep temporal features across multiple variables and time steps. Further, cascaded gated recurrent unit(GRU) layers are employed to delve into the temporal correlations of these features. The dataset utilized in this study was sourced from Jiangsu wind farms in 2019 and validated using quarterly wind power data. When compared to empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD) and VMD signal decomposition methods, the proposed NVMDHATCN-GRU model exhibited a notable decrease in the RMSE evaluation index by 344.77%, 128.11% and 19.41% respectively, in the second-quarter data. These experimental findings underscore the high accuracy and effectiveness of the proposed method.