Super short term combined power prediction for wind power hydrogen production

Abstract

A combined ultra-short-term wind power prediction strategy with high robustness based on least squares support vector machine (LSSVM) has been proposed, in order to solve the wind abandonment caused by wind power randomness and realize efficient hydrogen production under wide power fluctuation. Firstly, the original wind power data is decomposed into sub-modes with different bandwidth by variational modal decomposition (VMD), which reduces the influence of random noise and mode mixing significantly. Then dragonfly algorithm (DA) is introduced to optimize LSSVM kernel function and the combined ultra-short-term wind power prediction strategy which meets the time resolution and accuracy requirements of electrolytic cell control has been established finally. This model is validated by a wind power hydrogen production demonstration project output in the middle east of China. The superior prediction accuracy for high volatility wind power data is verified and the algorithm provides theoretical basis to improve the control of wind power hydrogen production system.