link to homepage

Project BEAM-ME

Navigation and service

BEAM-ME Session at OR Conference

Implementation of Acceleration Strategies from Mathematics and Computational Sciences for Optimizing Energy System Models

Stream: Software Applications and Modelling Systems
Chair: Felix Cebulla

1 - Getting linear optimising energy system models ready for High Performance Computing
Manuel Wetzel, Karl-Kien Cao, Frederik Fiand, Hans Christian Gils

State-of-the-art energy system models include a comprehensive representation of energy sectors and their associated technologies in high spatial and temporal resolution. This complexity is increased further by technologies linking the system along temporal, spatial and sectoral dimensions such as electrical energy storage units, transmission grids or combined heat and power plants. One of the downsides of the increasing detail in linear optimising energy system models is the necessary time to solve the model. The BEAM-ME project addresses the need for efficient solution strategies for complex energy system models. The project brings together researchers from the fields of energy systems analysis, mathematics, operations research, and informatics and aims at developing technical and conceptual strategies for every step of the solution process. This includes changes to the formulation of the energy system model, improving the solvers and utilising the resources of high performance computing. This talk provides an overview of the challenges in adapting the formulation of the energy system model REMix to utilise a solver based on a parallel interior point method. The solver exploits the underlying block structure of the model while maintaining the ability to find the global optimal solution. The block structure has to be communicated to the solver by annotating the energy system model. First insights of current results are discussed and an outlook on future steps is given.

2 - Optimizing large-scale linear energy problems with block diagonal structure by using parallel interior-point methods
Daniel Rehfeldt, Ambros Gleixner, Thorsten Koch

In the wake of the Fukushima nuclear accident in 2011, the German government initiated a policy (named “Energiewende”) to shift energy systems towards sustainable, renewable technologies. Since this transition comes with a massive decentralization and a concomitant increase in the size of realistic energy models, the project BEAM-ME has been launched to develop methods for solving currently intractable energy optimization problems. These models are encoded as large-scale linear programs (LPs) that exhibit a block-diagonal structure with both linking constraints and linking variables. In this talk, initial results of adapting the parallel interior point solver PIPS-IPM for these energy LPs are presented. In particular, the underlying mathematically approach will be presented and the practical implications of employing supercomputers to implement this approach

3 - High Performance Computing with GAMS
Frederik Fiand, Michael Bussieck

BEAM-ME is a project funded by the German Federal Ministry for Economic Affairs and Energy and addresses the need for new and improved solution approaches for energy system models of vast size. The project unites various partners with complementary expertise from the fields of algorithms, computing and application development.
The main focus is on large-scale linear programs (LPs) arising from energy system models. Such models have a block structure that is not well exploited by state-of-the-art LP solvers. Without considering this structure the models become quickly computationally intractable. Within the BEAM-ME project, new solution algorithms that exploit the block structure and utilize the power of High Performance Computers (HPC) are developed and will be made available to energy system modelers. Automatic detection of block structures in models has its limits and hence the user needs to provide block structure information via some model annotation. GAMS for some time has facilities to annotate a model. We discuss some extensions to the GAMS language which forms the interface between the energy system modeler and the newly developed algorithms.
We provide an overview on the large variety of challenges we are facing within this project, present current solution approaches and provide first results.

4 - High Performance Computing for Energy System Modelling
Thomas Breuer, Dmitry Khabi

The common approach to solve linear program (LP) arising from an energy system model (ESM) is based on the shared memory paradigm. However, the apparent simplicity of the non-uniform memory access architecture (NUMA) limits the scalability of the underlying hardware components, such as the memory and compute units. On the other hand, increasing of the resolution in spatial and temporal data of ESM requires more and more computational resources. High Performance Computing (HPC) provides a technical solution to overcome these limitations. Use of HPC demands changes to the existing ESM solving framework. This talk presents the latest results obtained in the BEAM-ME project. We discuss the central questions that came up during integration of the parallel optimization solver PIPS in the framework of GAMS language, which is widely used in the field of ESM. Based on a block structured (stochastic) linear problem, the results regarding the scalability and efficiency of the HPC solution were obtained on two Petaflops supercomputers: Hazel Hen, a Cray XC40-system, at High Performance Computing Center Stuttgart (HLRS) and JURECA, a fat tree EDR-InfiniBand cluster, at Jülich Supercomputing Centre (JSC).