A novel hybrid system for wind speed forecasting

Accurate wind speed prediction is vital to enhance wind energy utilization efficiency and maintaining the stability of societal structures. In this paper, the primordial wind speed data is decomposed via variational mode decomposition (VMD) into sub-sequences spanning low to high frequency components. The low-frequency sequences are then predicted using a bidirectional gated recurrent unit (BiGRU), while the high-frequency sequences are forecasted using an extreme learning machine (ELM). This paper introduces a novel method that employs cosine similarity (CS) to determine the length of the input in a deep learning model, which is termed lookback. Finally, the individual predictions are effectively integrated to produce the final forecasts. This hybrid system is referred to as VMD-CS-BiGRU-ELM. Predictions based on real wind speed data and different error criteria indicate that the proposed system performs well.