Tackling Challenges in Energy System Research with HPC

Energy system optimization models are one of the central instruments for the successful realization of the energy transition towards renewable sources. We have identified three major challenges to overcome the current limitations in energy system research. First, studying the future is subject to large uncertainties and these uncertainties are usually tackled with modeling of just a small subset of all possible scenarios. This has proven to be inadequate since most models are highly sensitive to input data. Second, the widely-used commercial solvers show poor scalability and are limited to single shared-memory compute nodes. Thus, models are defined with a lower resolution and technological diversity than necessary. The third challenge is that single models usually tend to investigate only certain aspects of an energy system, which do not cover all parts of future pathways. To overcome those limitations, we inspect the conceivable parameter space by using a hitherto unattained number of model-based scenarios. Therefore, we have implemented an automated parameter sampling based on a broad literature review, and a self-developed distributed-memory solver that outperforms commercial solvers. In addition, we have coupled different types of models in an automated, parallelized workflow. We use this workflow for a case study of the German power system. By evaluating more than 3600 scenarios, we observe a clear dominance of photovoltaics in future system designs. Efficiently leveraging the capability of HPC by combining those approaches could be a game changer for the energy-system analysis community and could ensure a better applicability for real world policy support.

---

Contributors:

• Karl-Kiên Cao
• Manuel Wetzel
• Ulrich Frey
• Shima Sasanpour
• Jan Buschmann
• Kai von Krbek
• Aileen Böhme
• Charlie Vanaret