Extreme learning machine for real-time damping of LFO in power system networks

This article proposes a real-time power system stabilizers (PSS) parameter optimization technique employing extreme learning machine (ELM) to enhance overall system stability by damping out the low-frequency oscillations. It models two electric networks, i.e., single machine infinite bus systems where the first network's synchronous machine is equipped with a PSS only, and the second network's synchronous machine is equipped with a unified power flow controller coordinated PSS. It uses diverse loading conditions to investigate the performance of the proposed ELM model-tuned PSS technique and compares it with conventional PSS and the referenced works in terms of the eigenvalues and minimum damping ratios. Additionally, the satisfactory values of the well-known statistical performance indices including the root mean squared error (RMSE), mean absolute percentage error, RMSE-observations-to-standard deviation ratio, coefficient of determination (R²), Willmott’s index of agreement, and Nash–Sutcliffe model efficiency coefficient provide confidence in the developed technique in predicting PSS parameters. Besides, comparisons of results from time-domain simulation demonstrate the ELM model tuned system's superiority over the conventional approach for both test cases. Furthermore, the ELM models require less than a cycle to predict PSS parameters for any loading condition that endorses the developed technique's real-time application.