Training system for AI assisted micro-PMU gradient protection

¹ Óbuda University, 1034 Bécsi út 96, Budapest
² Technical University of Kosice, 040 01 Letná 1/9, Kosice

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Published online: 11 June 2026

Abstract

Traditional power grid protections often suffer from latency due to the observation window required for Root Mean Square (RMS) and harmonic calculations. To overcome this barrier, this research proposes a local multilayer perceptron (MLP) model integrated into a Micro-PMU device to provide ultrafast, gradient based fault detection. Since training such models requires high fidelity data —which is scarce due to the variable nature of grid fault events— this paper presents a hardware in the loop (HIL) training system utilizing a high precision digital to analog Converter (DAC) interface. By injecting simulated fault and nominal transients directly into the Micro-PMU analog input, a realistic training environment is achieved where the protection logic is optimized via supervised learning. Experimental results demonstrate a detection latency of 60-90 us, representing a significant advancement over traditional industrial relays. The design of the DAC feedback system and the implementation of the MLP driven differential gradient logic are discussed.