EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2906. Decomposing stochastic energy system optimization models – time-splitting in Benders Decomposition vs. PIPS-IPM++

Invited abstract in session TD-19: Decomposition techniques applied to energy problems, stream OR in Energy.

Tuesday, 14:30-16:00
Room: 44 (building: 116)

Authors (first author is the speaker)

1. Shima Sasanpour
Energy Systems Analysis, German Aerospace Center (DLR)
2. Manuel Wetzel
Energy Systems Analysis, German Aerospace Center (DLR)
3. Karl-Kiên Cao
Energy Systems Analysis, German Aerospace Center (DLR), Institute of Networked Energy Systems
4. Andres Ramos
Departamento de Organizacion Industrial - Instituto de Investigacion Tecnologica, Universidad Pontificia Comillas

Abstract

Energy system optimization models (ESOMs) are a helpful tool to design future energy systems. Although uncertainties in the input data assumptions can significantly influence the structure of the energy system, they remain unconsidered in many cases. Through the application of stochastic programming (SP) it is possible to obtain expansion decisions that offer risk hedging with regard to the uncertainty of future developments.
ESOMs become large in size when a broad regional scope and technological diversity for sector coupling are considered, even without taking uncertainties into account. Therefore, speed-up techniques are needed to keep the models solvable, especially if SP is additionally considered. Benders Decomposition (BD) is a widely used method to solve SP models since the second stage can be solved in independent subproblems, allowing for a parallelization along the stochastic scenario dimension. However, the cardinality of the scenario set is usually much smaller than the number of time steps.
In our analysis we expand the parallelization potential of stochastic ESOMs by additionally splitting the scenarios along the time dimension. We apply this decomposition technique to two methods: First, we consider time-splitting in BD in combination with MPI. This is then compared to the parallel high-performance computing solver PIPS-IPM++, which was recently extended to also consider stochastic optimization, allowing a decomposition along the scenario and time dimension.

Keywords

Status: accepted

Back to the list of papers