Optimized SVM and feature selection for fault detection and classification in GFM-IBR system

Due to the increasing integration of new energy sources and the critical need for power grid stability, accurate fault detection and classification in power systems has become paramount. Therefore, this paper presents a new, optimized framework for fault detection in Grid-Forming Inverter-Based Resource (GFM-IBR) power systems. This framework is based on frequency feature extraction using the Fast Fourier Transform (FFT) and a Support Vector Machine (SVM) model. First, we sample the voltage and current signals from the IBR side. Then, the angle and phase values of these signals are extracted using FFT. The average angle and phase values obtained within the 10-millisecond period after a fault in each phase are then designated as the main features. The distinctive aspect of this study is its dual optimization approach. This involves both adjusting the parameters of the SVM model and optimally selecting features using several evolutionary and metaheuristic algorithms, specifically HHO, PSO, GA, GWO, WOA, GOA, DO, and AEO. The method presented in this paper was tested on a power system connected to a GFM-IBR, simulated in PSCAD software. The data generated in PSCAD was then transferred to a Google Colab environment for feature extraction and SVM model training. • A novel framework for fault detection and classification in power systems with GFM-IBR integration is proposed. • Frequency-domain features are extracted using time-windowed Fast Fourier Transform (FFT) of network signals. • Support Vector Machine (SVM) is employed for fault classification, optimized through dual-stage tuning. • The dual optimization strategy includes both SVM parameter tuning and feature subset selection. • Eight evolutionary/metaheuristic algorithms (e.g., HHO, PSO, GA, GWO, etc.) are evaluated for optimization effectiveness. • Simulation and testing in PSCAD and Google Colab demonstrate improved classification accuracy and reduced model complexity.