Neural networks approach to online identification of multiple failures of protection systems

In complex emergency situations, failed protection relays and circuit breakers (CBs) have to be identified in order to begin the restoration process of a power system. This paper proposes a novel neural-network approach to identify multiple failures of protection relays and/or CBs. The approach uses information received from protection systems in the form of alarms and is able to deal with incomplete and distorted data. All possible emergencies are simulated and analyzed separately for each section of a power system. Taking into consideration supervisory control and data-acquisition system malfunctions, the corrupted patterns are used to train neural networks. The preliminary classification of emergencies into two different classes is applied to improve the system's performance. The evaluation of results shows that the overall error rate does not exceed 5%. The developed system was tested on a real power system.