Energy-Efficient Analysis of Synchrophasor Data using the NVIDIA Jetson Nano

Abstract

Smart Grid Technology is an important part of increasing resilience and reliability of power grids. Applying Phasor Measurement Units (PMUs) to obtain synchronized phasor measurements, or synchrophasors, provides more detailed, higher fidelity data that can enhance situational awareness by rapidly detecting anomalous conditions. However, sample rates of PMUs are up to three orders of magnitude faster than traditional telemetry, resulting in large datasets that require novel computing methods to process the data quickly and efficiently. This work aims to improve calculation speed and energy efficiency of anomaly detection by leveraging manycore computing on a NVIDIA Jetson Nano. This work translates an existing PMU anomaly detection scheme into a novel GPU-compute algorithm and compares the computational performance and energy efficiency of the GPU approach to serial and multicore CPU methods. The GPU algorithm was benchmarked on a real dataset of 11.3 million measurements derived from 8 PMUs from a 1:1000 scale emulation of a power grid, and two additional datasets derived from the original dataset. Results show that the GPU detection scheme is up to 51.91 times faster than the serial method, and over 13 times faster than the multicore method. Additionally, the GPU approach exhibits up to 92.3% run-time energy reduction compared to serial method and 78.4% reduction compared to the multicore approach.