Application of Regression Analysis in Novel Power System Stabilizer Design

The paper presents the methodology for designing knowledge based power system stabilizers (PSSs). The proposed methodology relies on extensive nonlinear digital or analog simulation to detect correlation patterns between input damping states and the PSS output signal. Use of linear and nonlinear least square regression analysis results in a linear and nonlinear state feedback structure that is easier to implement and effective in damping either local or interarea electromechanical modes of oscillation. The PSS design technique is a two-step process: firstly, to identify the most effective damping signals for feedback via fast Fourier spectra analysis and SISO transfer function identification, secondly, to search for the most effective linear]nonlinear regression law that ensures the damping effectiveness. The regression law is based on an ideal PSS analog 'model' with excellent damping performance with the objective of emulating, or duplicating, such performance over a given time period (3-5s). PSS designs are compared using a specified state variation weighting function J with the objectives of minimizing the machine speed and acceleration power, as well as the active and reactive power deviations. This is done iteratively for each design simulation run and the best regression law is adopted.