Application of Artificial Intelligence to Diagnose Faults of Grid-tied Solar Power Systems

Solar energy has always been an extensively researched field worldwide. One of the key concerns is diagnosing and addressing faults in grid-tied solar power systems. Solar panels, as vital components of these systems, are prone to various issues including short circuit, open circuit, discoloration and surface cracks. Similarly, DC/DC or DC/AC converters often encounter semiconductor component failures due to prolonged high-frequency switching. This paper aims to comprehensively investigate and propose diagnostic methods for grid-tied solar power systems. The suggested faults for diagnosis encompass shading, shorts, and mixed faults in PV array, as well as shorts and open circuits in diodes and switches of the DC/DC boost converter. Additionally, shorts in sub-modules of the modular multilevel converter (MMC) are considered. To identify these faults, analysis of I-V curve of PV strings, diode voltage signals from the boost converter and three-phase output voltage of MMC is conducted. Finally, the faults are extracted by feature extraction algorithms, preprocessed by Principal Component Analysis (PCA) and classified by Support Vector Machines (SVM) – an artificial intelligence technique. Overall, this research provides novel insights and develops high-accuracy predictive models, achieving accuracy levels of nearly 90%.