The diakoptics approach is interesting for avoiding excessive system matrices when simulating large grids.
In some test investigations, this was particularly interesting for the refactorization step. For system matrix changes, local refactoring of the submatrix is sufficient in contrast to refactorization of the entire matrix. This has the potential to speed up the step enormously and is beneficial in addition to parallelization. This would therefore mainly benefit simulations of e.g. cascading fault events, where multiple recalculations are necessary and system matrixes can not be precalculated for every case.
The tests were implemented on an older dpsim version and do not take into account implementations of, for example, PARTIAL_REFACTORIZATION_METHOD in the KLU solver as a comparative value, which could already provide an alternative solution.
Proposed procedure:
- Test and check whether the same improvement in performance is evident and the approach remains interesting.
- Implement the diakoptics functionality on other components (e.g. 3ph).
- Add an example case with a high number of nodes to show benefits.
The diakoptics approach is interesting for avoiding excessive system matrices when simulating large grids.
In some test investigations, this was particularly interesting for the refactorization step. For system matrix changes, local refactoring of the submatrix is sufficient in contrast to refactorization of the entire matrix. This has the potential to speed up the step enormously and is beneficial in addition to parallelization. This would therefore mainly benefit simulations of e.g. cascading fault events, where multiple recalculations are necessary and system matrixes can not be precalculated for every case.
The tests were implemented on an older dpsim version and do not take into account implementations of, for example, PARTIAL_REFACTORIZATION_METHOD in the KLU solver as a comparative value, which could already provide an alternative solution.
Proposed procedure: