Hybrid intelligent framework for one-day ahead wind speed forecasting

Nowadays wind power is considered as one of the fastest growing alternative energies, that is expected to continue to grow rapidly in the upcoming years.Depending on the countries that invent, generate and export the green energy technologies,the world will witness a major leap in this field. In our study, we focused on the wind speed forecasting for the wind farms hosting purpose. A new hybrid architecture is implemented for the forecast of one-day ahead wind speed based on an Adaptive Grey Wolf optimizer (AGWO),the Singular Spectrum Analysis (SSA) and the proposed hybrid Encoder-Decoder-Convolutional-Neural-Network-Gated-Recurrent-Unit (ED-CNNGRU) model.The choice of the proposed framework was to overcome the limitation of the hyper-parameters tuning process that depends mainly on the datasets characteristics, to achieve a standard model capable to deal with several time series without adaptation.Where the AGWO is applied for the optimal combination tuning of the SSA window length with the ED-CNNGRU layers, then the AGWO-SSA is used to decompose the original wind speed series into its trend and details components in order to eliminate the fluctuating noise from the original raw data ,followed by the optimized ED-CNNGRU model designed for one day ahead efficient wind speed prediction. The experimental results showed that the proposed architecture outperformed the benchmark models in matter of adaptability and forecasting performance that were validated by a remarkable decrease in the RMSE (Root Mean Square Error) , MAE (Mean Absolute Error) and MAPE (Mean Absolute Percentage Error) error metrics values. From experimental results of three datasets that have different characteristics to confirm the performance of the proposed approach.

Automated summary

Wind power is considered one of the fastest growing alternative energies, expected to continue its rapid growth. A new hybrid architecture for wind speed forecasting showed superior adaptability and predictive performance, outperforming benchmark models in reducing error metrics values.