Modeling the electricity system in integrated assessment models - a time-series clustering approach for diverse futures

The growing role of variable renewable electricity calls for high time resolution when modeling the electricity sector. However, this can be computationally too expensive for models that combine different sectors and run over long timeframes, such as Integrated Assessment Models (IAMs). The computational burden can be reduced by, e.g., timeslicing or time-series clustering. Clustering approaches have the benefit of maintaining chronology and thus enabling electricity storage modeling; however, which hourly-level features clustering should capture for the diverse long-term scenarios calculated with IAMs have remained unclear. This paper makes a two-fold contribution to address these challenges. First, we propose an hourly-resolution electricity sub-model that links to the annual level of a suitable IAM, and uses time-series clustering to reduce the hourly dimension to a small number of representative periods. Second, we propose a new clustering approach – the multi-scenario clustering – that functions robustly across a range of future scenarios that might be calculated with an IAM. In this approach, outputs from multiple model runs covering different future scenarios are used in clustering to define representative periods. Our results show that this approach produces a clustering that works for a wide range of future scenarios nearly as accurately as a-posteriori clustering tailored for each scenario individually. Therefore, the multi-scenario clustering approach provides a robust way to save on computational burden while offering a relatively good level of accuracy in comparison to using the full, multi-year, hourly time-horizon, which would be prohibitively costly to use in IAMs.