Abstract
Capacity Expansion Models (CEMs) are widely used in the academic literature to understand the needs and dynamics of highly renewable energy systems. Due to computational constraints, it is common to aggregate time-series data such as hourly power output from Variable Renewable Energy Sources (VRES) using clustering algorithms. However, there is evidence that the presence of wind data leads to increased clustering errors and biased investment decisions. With the above motivation, we combine two approaches from the literature and compare them against the state-of-the-art approach. For a small number of clusters, the proposed approach recovers 95% of the original variance and correlation. This leads to more robust investment decisions. However, we stress the increased computational burden involved.