Selection of representative slices for generation expansion planning using regular decomposition

In power and energy system planning tools, the temporal detail is often reduced by selecting representative slices out of longer time series. Various methods exist for the selection task, but they may prove slow or otherwise unfavourable in practical applications. Here, a generalized clustering algorithm, referred to as regular decomposition, is presented and applied to a power system planning study covering countries in the Northern Europe. The algorithm is compared with other selection methods, and the comparison is repeated with various number of representative slices and in three carbon price scenarios in order to provide more robust results. When selecting four weeks or more, regular decomposition is shown to perform relatively well compared to the other selection methods in terms of the total costs resulting from the power system model runs. When applied to inter-annual time series, regular decomposition is demonstrated to scale well. Although random sampling shows the most stable performance overall, the results indicate the need to test several methods for each system. Moreover, the results highlight the need to include net load peaks in the selected slices and to carefully estimate their position in the time series. A two-stage method for including net load peaks is presented.