VMD-SCINet: a hybrid model for improved wind speed forecasting

Wind energy is gaining importance owing to its renewable and environmentally friendly characteristics. However, the variability and stochastic nature of wind speed makes accurate forecasting difficult. Hence, this study introduces a novel approach (VMD-SCINet) for wind speed forecasting (WSF) by integrating the strengths of variational mode decomposition (VMD) and sample convolution and interaction network (SCINet) architecture for the prediction of wind speed. This study utilizes VMD as a denoising technique for wind speed data and incorporates SCINet to capture global patterns and long-range dependencies for the WSF. The wind speed data acquired from two distinct sites: Leicester, and Portland is used for the evaluation. To evaluate the WSF capability of the proposed hybrid model, it's performance is compared to robust models using data from two wind farms across six different time horizons such as 5-min, 10-min, 15-min, 30-min, 1-hour, and 2-hours. The results from two experiments demonstrate that the proposed approach outperforms other models, leading to a significant improvement in WSF accuracy across all evaluated time intervals.