Gaussian process digital twin for voltage stability analysis of complex power networks under perturbation

Static voltage stability is essential for complex power networks incorporating a large share of renewable energy sources (RES). However, its estimation is a challenging task under network perturbation. Data analytics methods can be readily utilized to predict voltage stability index. Hence, the concept of Gaussian process (GP) based digital twin has been utilized in this paper to predict the voltage stability index from topological and electrical network features. The GP based prediction provides the mean and the uncertainties in the network stability margins. Firstly, the critical network features responsible for voltage stability index are identified using the Bayesian feature selection technique. With the critical network features, we train the GP with the stability margin of the power network. We test the GP model to predict the voltage stability margin against topological perturbations like line-trips and also in the line parameters. The results are benchmarked on a few large-scale power networks e.g. IEEE 300 Bus, Texas/ACTIVSg (2000 nodes), Great Britain (2224 nodes) and European power grid case2869pegase (2869 nodes), where the mean and uncertainty of the prediction for different kernel functions are used in the GP regression model. Our results show that there is always a bias in the mean prediction of the stability margin for the critical nodes.