Abstract: The paper proposes a mode identification method for low frequency oscillations in power system. First, an improved empirical mode decomposition (EMD) denoising algorithm is employed for signal preprocessing. The algorithm superimposes a cosine wave onto the signal and conducts a second decomposition after implementing the interval thresholding. It overcomes the shortcomings of serious aliasing noise and long computing time, which exist in the conventional algorithms, and separates the signal and noise in a fast and effective manner. After signal denoising, the matrix pencil (MP) algorithm is used to extract mode parameters. Introducing relative difference of singular values can solve the key problem of order determination in MP algorithm, thus leading to an accurate estimation for model order. Lastly, numerical signals, system simulation signals and tested power grid signals are analyzed. Simulation results show that the proposed method achieves excellent performance in anti-noise ability, parameter accuracy and computing speed.

Key words: low frequency oscillation, mode identification, empirical mode decomposition (EMD), matrix pencil, singular value decomposition