Towards Emulating Arbitrary-Scale Electromechanical Transients for Anomaly Detection

This paper provides a novel contribution to the field of Machine Learning Anomaly Detection (MLAD) for Wide Area Monitoring, Protection, and Control for Power Grids by developing a hardware-based emulation of electromechanical transient anomalies at multiple scales and developing a mathematical model capable of creating new anomalies at arbitrary scales for use in future work of emulation via controllable sources. The work is critical to creating realistic, real-world data capable of providing comprehensive validation of MLAD algorithms at non-trivial scales. It provides a thorough methodology for understanding the entire experimental and modeling process all the way from inducing electromechanical transients on a 3-phase, 4-pole PID-controlled synchronous generator via toggling of discrete 3 phase loads, to data collection through Phasor Measurement Units (PMUs) and a Phasor Data Concentrator (PDC), to data processing and mathematical modeling. In this exploration, it also provides helpful views into the characteristics of electromechanical anomalies far exceeding those in simulation as seen from the perspective of emulation, real-hardware, and all the subsequent real-world artifacts, including accumulating loads causing transients, real noise, and PMU to PDC sampling artifacts to name a few. Furthermore, all data and analysis are published concurrently to allow replication by the broader community.