WSFNet: An efficient wind speed forecasting model using channel attention-based densely connected convolutional neural network

This paper introduces a novel deep neural network (WSFNet) to efficiently forecast multi-step ahead wind speed. WSFNet forms the basis of the stacked convolutional neural network (CNN) with dense connections of different blocks equipped with the channel attention (CA) module. Dense connections create direct transition paths between the input and all subsequent convolutional blocks. This encourages the reuse of all activations at the network input without loss of gradients in subsequent layers. The CA modules contribute significantly to the performance of the network by suppressing non-useful features extracted by each convolution block. In the proposed method, variational mode decomposition (VMD) was utilized to provide an effective preprocessing and improve the forecasting ability. The case study was conducted on publicly available data from Sotavento Galicia (SG) wind farm. In the evaluations, three variants of the proposed network were analyzed and compared with state-of-the-art deep learning methods. When the results were analyzed, the overall correlation coefficient (R), root mean square error (RMSE), mean absolute error (MAE), and symmetric mean absolute percentage error (SMAPE) were obtained as 0.9705, 0.7383, 0.5826, and 0.0466, respectively. The obtained results indicate that the proposed method achieved a competitive performance and can be effectively used for smart grid operations. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This paper introduces a novel deep neural network (WSFNet) for efficiently forecasting multi-step ahead wind speed, incorporating dense connections and channel attention modules, as well as utilizing variational mode decomposition for preprocessing, achieving competitive performance.