Abstract: For detecting evolving faults in an electric power grid, a machine learning based model is trained and the trained model used using events with time information as input. An event is generated by a function in an intelligent electronic device based on one or more values measured from the electric power grid, the event indicating at least the function and its output. In the training, a plurality of event patterns that are extracted from event history data are used, an event pattern including in occurrence order events preceding within a time span a disturbance event. The disturbance event is an event resulting to an interruption in power supply in the electric power grid. The trained predictor is outputting predictions for occurrence times of disturbance events in the electric power grid, and they are displayed for detecting evolving faults in the electric power grid.