Generating synthetic energy time series: A review
Marian Turowski, Benedikt Heidrich, L. Weingärtner, L. Springer, Kaleb Phipps, Benjamin Schäfer, Ralf Mikut, Veit Hagenmeyer
Abstract
As the energy system transitions to an intelligent smart grid with a mostly renewable energy supply, synthetic energy time series are required to facilitate the development and improvement of methods for smart grid applications. These synthetic energy time series must exhibit characteristics similar to the real energy time series and applicable to specific use cases. Furthermore, evaluation methods must be applied to verify that synthetic energy time series have the desired characteristics. Whilst many methods exist in the literature to generate synthetic energy time series, up until now, no work has focused on analysing and comparing these methods. Therefore, this study provides a structured literature review of generating synthetic energy time series. The review focuses on five aspects: (1) Identifying methods used to generate synthetic energy time series, (2) categorising these methods according to the generation approach taken, (3) analysing the characteristics of these generated synthetic energy time series, (4) identifying the uses cases for which the time series are generated, and (5) considering how the generated synthetic energy time series are evaluated. In total, this study reviews 169 articles focusing on generating synthetic energy time series and identifies several key research gaps leading to multiple open research fields. The most important open research fields include the need for a standardised evaluation, more generation methods for synthetic time series from generation and battery storage systems, and a stronger focus on further use cases. Word count: 11395 • A large range of methods are used to generate synthetic energy time series. • Most generation methods are applied with bottom-up or direct generation approaches. • Most generated energy time series are low aggregated residential consumption. • Typical use cases are customer profiling, power load, & energy output forecasts. • Most identified articles evaluate the fidelity of generated synthetic time series.