Multi-step-ahead wind speed forecasting based on optimal feature selection and a modified bat algorithm with the cognition strategy

Abstract

With the arrival of big data, data mining analysis and high-performance forecasting of wind speed is increasingly attracting close attention. Despite the fact that massive investigations concerning wind speed forecasting in theory and practice have been conducted by multiple researchers, studies concerning multi-step-ahead forecasting are still lacking, impeding the further development in the field. In this study, a novel hybrid approach is proposed for multi-step-ahead wind speed forecasting utilizing optimal feature selection and an artificial neural network optimized by a modified bat algorithm with cognition strategy. The proposed hybrid model can largely remedy the deficiencies of neural networks for multi-step-ahead forecasting, which is validated for different forecasting horizons, and is shown to work effectively. Finally, experiments based on three verification units from the city of Penglai in China are conducted effectively, illustrating that the proposed model not only has advantages when compared with benchmark models, but also has great potential for application to wind power system.