ML speeds up Industrial ­development processes

Parameter analysis and optimization are common tasks in engineering applications. SCAI develops DesParO, a machine learning software package for intuitive exploration, ­automatic analysis, and optimization of parametrized problems. DesParO can be coupled with simulation packages or used for measurement data. It minimizes the number of simulations needed for building up a model adaptively. Fraunhofer SCAI applies this approach in particular for energy networks.

Networks for transport of gas, electricity, water, steam or chemicals are essential for modern economy. The ­Energiewende (energy turnaround) requires optimization of energy flows in order to reduce consumption of non-­renewable energy sources. The transformation of our ­energy backbone needs modeling, simulation and ­optimization of the corresponding networks.

The different energy networks can be ­mathematically modeled in a very similar way. Their numerical simulation can be based on the same ­numerical kernels. Our MultiphYsical NeTwork Simulation framework (MYNTS) is suited for all energy ­networks. MYNTS uses HPC technology. For parameter analysis and optimization, MYNTS provides an interface to DesParO.

Software MINTS
© MEV Agency UG – Merten Hans-Peter / Fraunhofer SCAI

Simulation, analysis and optimization

In order to set up an energy network model, first, the ­topology of the network is created. Then all the ­network ­elements are modeled. For physical devices, ­technical ­documentation is evaluated. If the behavior of a device or ­subsystem is not known yet, it can be learned with ­DesParO, based on ­measurement results. After network ­initialization, ­important operating scenarios have to be learned with ­DesParO, e.g. by clustering and / or classifying sensor data appropriately. Finally, optimization goals must be established, reflecting the ­intentions of relevant stakeholders.

Besides simulation and optimization workflows, MYNTS offers ensemble analysis based on DesParO. So, the impact of (smaller or larger) variations of parameters can be identified. Combinations of geometrical properties, physical parameters of ­devices, operating scenarios or profiles of ­consumers, ­generators, or prosumers can be varied and analyzed ­systematically. In addition, MYNTS offers modules for ­physically equivalent graph reduction and for graph ­clustering. The latter can be used for analysis of main flow ­directions or comparative analysis of several operating scenarios.



  • Mathematical techniques for transforming energy ­networks, e.g. gas and electrical power on long-distance and distribution levels.
  • Making the energy infrastructure (electrical power, gas, ­district heating with converters and storages) of ­cross-sectoral distribution grids more flexible.
  • Simulation and optimization of data center energy flows.
  • Energy management (DIN EN ISO 50001) and optimization for the chemical industry based on online process mining.
  • Chemical micro- and macrokinetics of fuel cells.