Performance Analysis and Improvement of Power Systems Ring-Down Electromechanical Mode Identification Algorithms

Abstract

With the commissioning of Wide Area Measurement Systems (WAMS) in large power grids, measurement-based mode identification is finding wide application. From power system stability viewpoint, mode identification from ring-down signals is important. Although ring-down identification algorithms have been studied for a few decades, these still have a scope for improvement. For example, Signal-to-Estimation-error Ratio (SER), which is the recommended fitness metric to compare original and estimated signals in iterative Prony method, sometimes performs suboptimally. So a superior metric is proposed here by combining SER withMean Absolute Percentage Error (MAPE). Another popular ring-down algorithm is matrix pencil, which is normally presented in a non-iterative formulation. It is shown here that iterative formulation of matrix pencil is feasible and is slightly faster than iterative Prony. From the viewpoint of mode identification of noisy signals, Singular Value Decomposition (SVD)-based non-iterative algorithms are reported to be superior. Hence three such algorithms, namely, Total Least Squares matrix pencil (TLS matrix pencil), Hankel Total Least Squares (HTLS) and Eigensystem Realization Algorithm (ERA) are evaluated comparatively. In the process, it is shown that TLS matrix pencil and HTLS algorithms are equivalent. Evident improvement in matrix pencil algorithm performance by incorporation of SVD suggests the same possibility in Prony algorithm. So a customized formulation of Structured Total Least Squares-Prony (STLS-Prony) algorithm is developed for application to power systems. This is compared with two known formulations of SVD-augmented Prony algorithm, namely, Principal Eigenvector-Prony (PE-Prony) and Total Least Squares-Prony (TLS-Prony). A Taylor series-augmented Fourier transform called Digital Taylor-Fourier Transform (DTFT) is examined for its ability to handle exponentially varying sinusoids and a novel concept termed neper response is put forth to characterize the same. It is shown that the computational efficiency of DTFT-based mode identification can be improved greatly by raising the Taylor series order.