Ensemble forecasting system for short-term wind speed forecasting based on optimal sub-model selection and multi-objective version of mayfly optimization algorithm

Wind energy has attracted considerable attention in the past decades as a low-carbon, environmentally friendly, and efficient renewable energy. However, the irregularity of wind speed makes it difficult to integrate wind energy into smart grids. Thus, achieving credible and effective wind speed forecasting results is crucial for the operation and management of wind energy. In this study, we propose an ensemble forecasting system that integrates data decomposition technology, sub-model selection, a novel multi-objective version of the Mayfly algorithm, and different predictors to better demonstrate the stochasticity and fluctuation of wind speed data. After decomposition using the data decomposition technology, each decomposed wind speed series is considered as the input to multiple predictors, from which the optimal forecasting model for each sub-series is determined based on sub-model selection. To obtain reliable forecasting results, a novel multi-objective version of the Mayfly algorithm is proposed to estimate the optimal weight coefficients for integrating the forecasting values of the subseries. Based on three experiments and four analyses, the proposed ensemble system is verified as effective for obtaining accurate and stable point forecasting and interval forecasting performances, thus aiding in the planning and dispatching of power grids.

Automated summary

Wind energy as a low-carbon, environmentally friendly, and efficient renewable energy source has attracted considerable attention, but the irregularity of wind speed poses challenges in integrating wind energy into smart grids. This study proposes an ensemble forecasting system to accurately forecast wind speed and is verified through experiments and analyses for its effective performance.