A comparison of clustering methods for the spatial reduction of renewable electricity optimisation models of Europe

Modeling the optimal design of the future European energy system involves large data volumes and many mathematical constraints, typically resulting in a significant computational burden. As a result, modelers often apply reductions to their model that can have a significant effect on the accuracy of their results. This study investigates methods for spatially clustering electricity system models at transmission level to overcome the computational constraints. Spatial reduction has a strong effect both on flows in the electricity transmission network and on the way wind and solar generators are aggregated. Clustering methods applied in the literature are typically oriented either towards preserving network flows or towards preserving the properties of renewables, but both are important for future energy systems. In this work we adapt clustering algorithms to accurately represent both networks and renewables. To this end we focus on hierarchical clustering, since it preserves the topology of the transmission system. We test improvements to the similarity metrics used in the clustering by evaluating the resulting regions with measures on renewable feed-in and electrical distance between nodes. Then, the models are optimised under a brownfield capacity expansion for the European electricity system for varying spatial resolutions and renewable penetration. Results are compared to each other and to existing clustering approaches in the literature and evaluated on the preciseness of siting renewable capacity and the estimation of power flows. We find that any of the considered methods perform better than the commonly used approach of clustering by country boundaries and that any of the hierarchical methods yield better estimates than the established method of clustering with k-means on the coordinates of the network with respect to the studied parameters