Planning the decarbonisation of energy systems: the importance of applying time series clustering to long-term models

The study of future energy scenarios with high shares of variable renewable energy sources (VRES) requires an accurate representation of VRES variability and storage capacity. However, long-term optimal expansion models, which are typically used to prescribe the evolution of energy systems, make use of coarse time series to limit computational effort. This weakness can entail an incorrect sizing of VRES plants and storage facilities. In this work, a novel method is proposed to mitigate the current limitations and enable accurate long-term planning of high-VRES decarbonisation pathways. Clustering methods are applied to time series, preserving the possibility of having inter-day and intra-day energy storage. To this end, the temporal framework of an open-source energy system model, OSeMOSYS, is modified to allow the implementation of interconnected, clustered representative days. Traditional and novel approaches are compared and benchmarked for a reference case study, i.e., a remote island. The results show that time series clustering can significantly improve the evaluation of the overall system cost, leading to a relative error of -5% (novel approach) instead of -35% (traditional approach) when 24 representative days are considered. Similarly, the new approach improves the sizing of VRES and storage facilities. The new technique is found to require three orders of magnitude less computation time than the traditional technique to achieve a comparable level of accuracy.