I. Introduction
Large-scale blackouts have led to many component failures, significant economic losses, and severe social impacts [1]. Tra-ditionally cascading failure study heavily relies on simulation models, such as Manchester model [2], hidden failure model [3], and OPA model [4]. Recently efforts have been made on extracting failure propagation patterns from simulated data. This includes branching process (BP) model [5], [6] and multi-type BP model [7] that extract high-level statistical information of outage propagation and the component interaction models such as influence graph [8], interaction network [9]–[11], and coupled interaction network [12] that enable an explicit study of the interactions between component outages.