Automated Identification of Re-Closing Events in an Operational Smart Power Grid

Electric power grid “smart” re-closing devices employ the latest microprocessor and telecommunications technologies to provide a unique and powerful fault interruption and re-closing capability for automated restoration. These devices work in unison to isolate faulted power line segments while minimizing the number of customers negatively affected by extended power outages. These devices also measure and store all monitored voltage and current waveforms associated with electrical disturbance events, which enables power personnel to conduct analysis to aid in repairs, conduct corrective actions, and implement mitigation strategies as needed. However, the amount of generated data has become prohibitive to traditional “by-hand” analysis techniques. This has led researchers to investigate and develop machine learning approaches capable of processing, analyzing, and identifying all of this “smart” device generated data. This work employs common machine learning (ML) techniques to automate the identification of the IEEE COMmon format for TRAnsient Data Exchange (COMTRADE) files generated during smart re-closing operations within an operational smart power distribution system. Based on individual waveform processing and classification, each COMTRADE file is assigned one of twelve labels. The combination of neural networks and logic driven classification resulted in a single-phase classification accuracy of 96.83% and a COMTRADE file identification accuracy of 92%.