Validation Methods for Energy Time Series Scenarios From Deep Generative Models

The design and operation of modern energy systems are heavily influenced by time-dependent uncertain parameters, e.g., renewable electricity generation, load-demand, prices. These typically represented a set discrete realizations known as scenarios. A popular scenario generation approach uses deep generative models (DGM) that allow without prior assumptions about the data distribution. However, validation generated scenarios is difficult, comprehensive discussion appropriate methods currently lacking. To start this discussion, we provide critical assessment used in literature. In particular, assess based on probability density, auto-correlation, power spectral density. Furthermore, propose using multifractal detrended fluctuation analysis (MFDFA) an additional method for non-trivial features like peaks, bursts, plateaus. As representative examples, train adversarial networks (GANs), Wasserstein GANs (WGANs), variational autoencoders (VAEs) two time series (photovoltaic wind from Germany 2013 to 2015) intra-day price form European Energy Exchange 2017 2019. We apply four both historical discuss interpretation results well common mistakes, pitfalls, limitations methods. Our shows no single sufficiently characterizes but ideally should include multiple be interpreted carefully context over short periods.