Transformer-based deep learning model for forced oscillation localization

Abstract

Accurately locating Forced Oscillations (FOs) source(s) in a large-scale power system is a challenging task, and an important aspect of power system operation. In this paper, a complementary use of Deep Learning (DL)-based and Dissipating Energy Flow (DEF)-based methods are proposed to localize forced oscillation source(s) using data from Phasor Measurement Units (PMUs), by tracing the forced oscillations source(s) on the branch level in the power system network. The robustness, effectiveness and speed of the proposed approach is demonstrated in a WECC 240-bus test system, with high renewable integration in the system. Several simulated cases were tested, including non-gaussian noise, partially observable system, and operational topology variations in the system which correspond to real-world challenges. Timely localization of forced oscillation at an early stage provides the opportunity for taking remedial reaction. The results show that without the information of system operational topology, the proposed method can achieve high localization accuracy in only 0.33 s.