Application and performance evaluation of a parallel solver for energy system models with high spatial and temporal resolution

Abstract

The increasing complexity of optimizing energy system models to answer research questions in more and more detail is one of the main drivers towards high performance computing. One of the current limitations is solving these large scale optimization problems with respect to memory constraints and a reasonable wall-clock time. The open source solver PIPS-IPM addresses this problem by providing access to distributed memory architectures and exploiting the block structure of the mathematical optimization problems. Application of PIPS-IPM requires additional information about the underlying block structure of the energy system model which is provided via model annotation based on the expert knowledge of the modeller. Further development of the PIPS-IPM solver and application to several optimizing energy system models are one of the key achievements of the BEAM-ME project. This required bringing together interdisciplinary expertise from the fields of energy systems analysis, mathematics, operations research, and informatics in order to establish efficient solution strategies for complex energy system models. This talk addresses the necessary steps of bringing the energy system model REMix to high performance computers and utilizing the parallel interior-point solver PIPS-IPM. Furthermore the performance of the solver is compared for several high spatial and temporal model instances and different types of decomposition.