Online identification of cascading event sequences in power systems using deep learning

The framework proposed in this paper focuses on the online identification of the reason of cascading event sequences in power systems with renewable generation. Cascading events involve highly complex dynamic phenomena and can in some cases severely compromise the security of modern power systems, leading even to blackouts. The proposed framework takes into account uncertainties associated with network operating conditions, contingencies and renewable generation. By utilizing measurement data, the methods within the proposed framework can predict in close to real time the reason of upcoming cascading events, as defined by the action of protection devices capturing dynamic phenomena related to voltage, frequency or transient instability. The framework is evaluated on a modified version of the IEEE-39 bus model, augmented with renewable generation and protection devices. The results demonstrate that the proposed method can successfully predict the reason of cascading events as they appear in a sequence with a mean accuracy of 97.4% with an online computation time of 1.09ms on average. The scalability of the method is showcased on a modified version of the IEEE-118 bus system.