LSTM-Based Rapid Identification of Dominant Low Frequency Oscillation Modal Features in Power System

Authors:

Abstract
Low frequency oscillation (LFO) analysis is quite important for the stability and security of modern power system. The most popular LFO identification methods, such as FFT, Prony, HHT, and even recently proposed algorithms based on neural networks, regarding the measured signal as a stationary and time-invariant signal. Moreover, some methods need tens of seconds or even minutes of measured data before analysis. These are not fit for the modern power system with high proportion of renewables and high proportion of power electronics, where the system status is fast time-varying and the LFO is with more complex characteristics. In this paper, a rapid LFO analysis method based on long-short-term memory neural network (LSTM) has been proposed to solve this problem. The identification process can be divided into three steps. Firstly, determine the order of LFO dominant modes. Then according to the result of order identification, identify frequency feature of each mode. Lastly, qualitatively evaluate attenuation factor of each mode. Experiments are made to prove effectiveness of the proposed method. Comparison work with Prony method demonstrates the advantages of this approach when the LFO signal is with noise and time-varying.
Funding Information
  • Research and Development

This publication has 18 references indexed in Scilit: