Wavelet-Based Fault Detection of a Single-Phase Transformerless Grid-Connected Photovoltaic Inverter System

Abstract

The increasing global transition towards renewable energy has highlighted the critical role of photovoltaic (PV) systems in sustainable electricity generation. This research investigates fault detection techniques for single-phase transformerless grid-connected PV systems using advanced wavelet-based analysis and machine learning classifiers. Focusing on critical system components, the study explores fault scenarios including DC link capacitor, IGBT, ground fault, AC filter capacitor, and short circuit conditions. By implementing Discrete Wavelet Transform (DWT) with Daubechies-4 mother wavelet and extracting statistical features, the research developed a comprehensive fault detection framework. Three machine learning models Support Vector Classification, Random Forest, and Neural Network were evaluated for fault classification accuracy. Results of the research reveals that the neural network model achieved the highest overall accuracy at 99.46%, followed by Random Forest at 99.40%, and Support Vector Classification at 99.20%. The precision metrics indicate superior performance of the SVC model in correctly identifying positive cases across all fault categories. The recall values demonstrate the models' effectiveness in identifying all relevant instances of each fault type, with the neural network model showing strength in this aspect. The results demonstrated exceptional performance, highlighting the potential of wavelet-based techniques in enhancing PV system reliability and safety. The methodology provides a robust approach for near real-time capable fault detection in simulation environments, offering significant implications for improving renewable energy infrastructure resilience.