The FCM-guided deep learning model for low-frequency oscillation damping for electric power networks

Machine learning (ML) techniques have gained substantial attention in many aspects of contemporary life during the last several years. As part of the digital revolution, the electricity industry is one of the leaders in implementing such appealing and effective technology for various applications. In general, low-frequency oscillations (LFO) are nonthreatening but slow-burning issues that, if not addressed appropriately, might lead to complete network collapse. Due to the significance of prominent ML family members in improving LFO damping in electric power system (EPS) networks, the applicability of the fuzzy c-means (FCM) clustering-based deep learning (DL) technique is modeled in this paper for two typical EPS networks by predicting the critical parameters of the power system stabilizers (PSS). The first network is a single-machine infinite bus (SMIB) network where the synchronous generator is equipped with a PSS. On the other hand, a unified power flow controller (UPFC) coordinated PSS is connected at one terminal of the synchronous generator of the second network. The clustering of the datasets obtained through the whale optimization algorithm (WOA) is performed based on the calculated silhouette values for both power networks. Then, several statistical performance indices (SPI) are evaluated to validate the robustness of the training and testing procedure of the DL method for the prepared data clusters using the FCM clustering technique. The efficacy of the proposed FCM-DL strategy in enhancing LFO damping for the two test networks is assessed based on standard analytical and time-domain analysis. Therefore, the minimum damping ratio (MDR), eigenvalue, rotor angle, and angular frequency with respect to time are simulated and analyzed. The article also includes a comparison of the findings of previous studies to illustrate the potential of the proposed FCM-DL strategy in improving EPS stability by damping out undesirable LFOs. It is worth noting that the developed FCM-DL models can predict the candidate parameters with a coefficient of determination (R2) value of more than 0.9974. During the implementation phase, the proposed strategy achieves competitive MDR, for instance, more than 0.50 and 0.74 MDR for the first and second networks, respectively.