Oscillation mode identification based on wide-area ambient measurements using multivariate empirical mode decomposition

Shutang You, Jiahui Guo, Gefei Kou, Yong Liu, Yilu Liu
2016 Electric power systems research  
Wide-area synchrophasor ambient measurements provide a valuable data source for real-time oscillation mode monitoring and analysis. This paper introduces a novel method for identifying inter-area oscillation modes using wide-area ambient measurements. Based on multivariate empirical mode decomposition (MEMD), which can analyze multi-channel non-stationary and nonlinear signals, the proposed method is capable of detecting the common oscillation mode that exists in multiple synchrophasor
more » ... nts at low amplitudes. Test results based on two real-world datasets validate the effectiveness of the proposed method. Keywords Wide-area synchrophasor measurement; ambient oscillation mode identification; multivariate empirical mode decomposition (MEMD). 2 level [7] . These ambient sychrophasor measurements have been recently used as a data source to extract real-time inter-area oscillation information [8, [10] [11] [12] [13] [14] [15] [16] [17] [18] . Since ambient measurements were first used in Ref. [11] to analyze the electromechanical oscillation, multiple approaches have been developed based on various signal processing techniques. There are two main categories of methods for oscillation analysis based on ambient and event data: transfer function based methods and subspace methods. Transfer-function-based methods directly estimate mode shape through treating measurements as system outputs. Typical transfer function based methods include Fourier transform [12-14], the Prony's method [15], the Matrix-Pencil method [16], Empirical Mode Decomposition (EMD) [17], the Yule-Walker method [18], and the singular value decomposition method [19], etc. To increase analysis efficiency, Ref. [13] adopted a FFT-based distributed optimization method to select the dominant measurement channels for estimating each oscillation mode based on ambient data. Ref. [20] proposed a two-step method which comprised of independent component analysis and random decrement to estimate the oscillation mode. Different from transfer-function-based methods, subspace methods obtain the oscillation mode information through identifying the system state space model using the measurements [11]. Typical subspace methods include the Canonical Variate Algorithm [21], the N4SID algorithm [22], and the autoregressive moving average block-processing method [23]. Recently, Ref. [24] proposed the robust recursive least square algorithm to analyze measurement data. Ref. [25] improved this method by proposing a regularized robust recursive least square method. Existing transfer function based methods have limitations in one or two of the following aspects. a) The capability to analyze drifting, non-stationary signals and provide localized results. Oscillation may drift frequently in ambient measurements [26] . For example, a high damping local oscillation mode may stimulate an inter-area oscillation mode with a pseudo negative damping ratio within a short duration. Existing methods may not be able to find the appropriate time window for analyzing this inter-area oscillation due to this drifting [27] . In addition, some measurements are mixed with fluctuations unrelated to oscillations, such as the frequency fluctuations caused by normal system operation and regulations [28] . These non-stationary measurements usually lead to difficulties to accurately analyze the subtle oscillation modes [29] . For example, the Prony's method requires the signal to be zero-mean and stationary, which may result to difficulties to analyze signals with high amplitude trends [23] . b) The capability to analyze multi-channel signals. Most existing methods can only process one signal so they have to analyze multiple signals in a separate way. they usually require measurements from critical devices, which has good observability (e.g., branch flow and bus frequency) on certain inter-area oscillation modes [8, 11, 17, 19, [23] [24] [25] [29] [30] [31] [32] [33] [34] [35] [36] . If the measurement that has good oscillation observability is not available in some areas, single-channel methods may not be able to provide oscillation information in these areas. Therefore, there is a need of developing multi-channel methodology to extract oscillation
doi:10.1016/j.epsr.2016.01.012 fatcat:f4ag5axjpjbudh5zdvrekuhne4