High Performance Computing vs. Heuristic: A performance benchmark for optimization problems with linear power flows

Abstract

We address the crucial aspect of unmanageable computing times of large-scale Energy System Optimization Models. Such models provide insights into future energy supply systems while keeping an overall perspective. However, the degree of phenomena or processes to be modeled is ever-increasing. With PIPS-IPM++ a novel solver is presented which is designed for linear optimization problems with linking variables (i.e. investment decisions) and linking constraints (i.e. power flow constraints). Compared to existing approaches for computing time reduction it makes use of High Performance Computing. Here, we present a benchmark study that compares the performance of PIPS-IPM++ with a usual speed-up heuristic (Temporal Zooming). Our results show that speed-ups between factor 10 and 15 are achievable with PIPS-IPM++ and Temporal Zooming, respectively. Despite PIPS-IPM++ has a great potential to parallelize solving, the tested version of the solver is especially useful for models without investment decisions (i.e. optimal power flow problems).