Wasserstein-Distance-Based Temporal Clustering for Capacity-Expansion Planning in Power Systems

Abstract

As variable renewable energy sources are steadily incorporated in European power systems, the need for higher temporal resolution in capacity-expansion models also increases.Naturally, there exists a trade-off between the amount of temporal data used to plan power systems for decades ahead and time resolution needed to represent renewable energy variability accurately. We propose the use of the Wasserstein distance as a measure of cluster discrepancy using it to cluster demand, wind availability, and solar availability data. When compared to the Euclidean distance and the maximal distance, the hierarchical clustering performed using the Wasserstein distance leads to capacity-expansion planning that 1) more accurately estimates system costs and 2) more efficiently adopts storage resources. Numerical results indicate an improvement in cost estimation by up to 5% vis-à-vis the Euclidean distance and a reduction of storage investment that is equivalent to nearly 100% of the installed capacity under the benchmark full time resolution.

Lucas Condeixa

Doctoral Researcher

Lucas Condeixa is a Doctoral Candidate in the Systems Analysis Laboratory in the department of Mathematics and Systems Analysis in Aalto University.

Fabricio Oliveira

Associate Professor of Operational Research

Fabricio Oliveira is an Associate Professor of Operational Research in the Department of Mathematics and Systems Analysis.