Increasing amounts of non-synchronous generation in power grids are bringing reductions in system inertia. In a grid with extremely low inertia, the estimation of frequency indicators such as the frequency nadir can be used to feed into predictive system controls that would avoid nuisances such as triggering system protection systems, avoiding needless blackouts. In this paper, the timing of a frequency nadir is predicted using a Nonlinear Auto-Regressive (NAR) model based on an Artificial Neural Network (ANN). The estimation method is tested under a gradual inertia reduction in order to observe the adaptability of the method, under various prediction horizons.
