Surrogate Models for Investigating Dynamic Security Regions of Renewables-Dominated Grids

Conventional grid operational planning usually entails dynamic security assessment performed by running time-domain simulations. This requires detailed modelling and is computationally burdensome. An alternative is to use a machine learning approach as a lightweight surrogate to evaluate stability for a set of operational inputs. Any supervised learning approach used will only be as good as the exemplar data it has been trained on. In this paper we propose the use of synthetic resampling to deal with lack of operational edge case examples and super resolution to improve coverage without additional samples. The contribution demonstrates improved accuracy in security assessment for an illustrative transmission network test case over a number of scenarios.