Optimization based priority clustering for time series in power system models - improving system robustness and storage deployment

Climate variability is a threat to the security of supply of decarbonized power systems. Therefore, modelers should include multiple weather years and climate projections into their power system optimization models. However, computational resources limit the amount of input data in a complex model. Therefore, time series clustering methods are commonly used. Yet, research has shown that applying these methods leads to errors especially for storage deployment, reducing security of supply. We therefore present the novel clustering algorithm obpc (Optimization Based Priority Clustering) which we test on a model of a fully decarbonized European electricity system in 2050 with the aim of achieving high security of supply under climate and weather variability. Obpc performs the clustering in the result data space and allows for the assignment of priority to days with certain characteristics. Using results from single-day optimizations for the clustering, we find that it is beneficial to cluster on investment decisions from the single-day optimization instead of generated electricity or residual load, reducing unmet demand by up to 81.6 % compared to input data based clustering. For the priority assignments, it was beneficial to give higher priority to days with high single-day system costs, reducing unmet demand by additional 88.5 % on average. In total, the unmet demand was reduced by 99 % on average while only prolonging computational time by a constant summand of 8 min compared to input data based clustering, making obpc suitable to plan power systems that perform robustly under a high number of weather years and climate scenarios. • Novel time series clustering method obpc used to enhance security of supply. • Clustering based on optimization results with priority assignment to time periods. • Reduction of unmet demand on average by 99 % compared to input data based methods.