Accelerating the Performance of Large-Scale TIMES Models in the Modelling of Sustainable Development Goals

Abstract

AbstractEnergy system models become very complex when introducing Sustainable Development Goals (SDGs) in high spatial and temporal detail. This can challenge their solvability and may require aggregation or reformulation of the optimisation problem or even solver-based methods for accelerating the solution time of the models. We provide insights into two powerful solver-based methods using a European TIMES-based model to guide the modeller in applying these methods. The first method involves efficiently parametrising the Barrier interior point solver in a shared-memory system, e.g., a personal computer. We find that with a suitable set of Barrier solver options, the run time of our test model was reduced by 95%. The second solver-based method uses distributed computing systems to solve the model matrix in parallel and across several nodes. We find that by exploiting the new parallel interior point solver PIPS-IPM++, we can scale up the model size several times without increasing solution runtimes when solving across multiple nodes. By combining solver- methods with suitable model reformulations, the energy system modelling research community can accelerate the solution of large-scale models featuring the assessment of the complex interactions between several SDGs.