Reasons behind this project
Dynamic stability studies in modern power systems rely on two types of unit models: generic models, which are standardised but often fail to accurately capture the fast dynamics of real devices; and vendor-specific black-box models, which are precise but opaque — they cannot be inspected, modified or used to diagnose the root cause of stability problems. As inverter-based resources proliferate, neither approach is sufficient on its own.
Compounding this, the process of tuning generic model parameters to match the behaviour of a real device is slow, manual and expertise-intensive. System operators and manufacturers spend significant engineering time on a task that, in principle, could be automated — releasing that effort for more valuable analysis work.
Objectives
DyPSOS (Dynamic Power System model Optimisation and Surrogates) develops an open-source automated framework that bridges the gap between generic and black-box models, without requiring access to proprietary model internals.
The framework combines three complementary techniques. First, heuristic and global optimisation algorithms automatically tune the parameters of generic dynamic models to match the time- and frequency-domain behaviour of a reference (black-box) model, using benchmark simulation results as the fitness signal. Second, the structure of the generic model — not just its parameters — is selected automatically, using feature extraction from simulation results to determine which control blocks and levels of detail are actually needed. Third, where generic models remain insufficient, data-driven methods (neural ODEs, DAEs, surrogate modelling) enhance equation-based models without sacrificing their analytical interpretability.
The resulting framework is validated on pan-European study cases drawing on open models from COLib, and demonstrated on grid-connection test cases where it replaces the costly manual development of manufacturer-model equivalents while preserving confidentiality.
Project partners
Fraunhofer IEE, SPEN, TotalEnergies and Amprion. The project is looking for additional partners.
Image credits: Image by starline on Freepik