Extreme scenario generation for renewable energies

Abstract

With a high penetration of renewable energies, scenario generation for wind and solar power is essential for the operation of modern power systems. Beyond the typical scenarios, extreme scenarios like full-capacity generation for consecutive days should also be taken into account. However, developed data-driven methods are unlikely to capture the characteristics of these extreme scenarios because of limited data. To this end, a three-staged extreme scenario generation method is proposed for renewable energies to effectively and efficiently generate extremely high power output scenarios. First, an extreme data augmentation algorithm is designed to push the original data distribution towards the extreme case. Then, based on the extreme value theory, the tail of the augmented dataset is modelled by Generalised Pareto Distribution (GPD). Last, with the augmented dataset and the conditional labels sampled from the fitted GPD, a conditional generative adversarial network is trained with the modified loss function. A case study on a real-world dataset shows that the authors’ proposed method has superior performance over the state-of-the-art generative models in terms of extreme scenario generation. Besides, the generated samples successfully capture the temporal and spatial correlation of real scenarios of renewable energies.