Principal component density estimation for scenario generation using normalizing flows

Abstract Neural networks-based learning of the distribution non-dispatchable renewable electricity generation from sources, such as photovoltaics (PV) and wind well load demands, has recently gained attention. Normalizing flow density models are particularly suited for this task due to training through direct log-likelihood maximization. However, research field image shown that standard normalizing flows can only learn smeared-out versions manifold distributions. Previous works on flow-based scenario do not address issue, distributions result in sampling noisy time series. In paper, we exploit isometry principal component analysis (PCA), which sets up a lower-dimensional space while maintaining computationally efficient likelihood We train resulting (PCF) data PV power demand Germany years 2013–2015. The results investigation show PCF preserves critical features original distributions, probability frequency behavior application is, however, limited but rather extends any dataset, series, or otherwise, be efficiently reduced using PCA.