A methodological review of cost-effective data-driven fault detection and diagnosis in distributed photovoltaic systems

The rapid evolution of Photovoltaic (PV) technologies and the widespread adoption of PV systems highlight the growing need for more efficient and cost-effective monitoring strategies to ensure reliable operation and optimal energy performance. This review presents a methodological approach, incorporating case-based measurements, for performance monitoring of distributed PV systems. It focuses on cost-effective data, such as time-series electrical parameters, which are crucial for accurate fault detection and diagnosis while identifying the constraints that limit the effectiveness of current performance monitoring algorithms. The review first categorises systematic faults in PV systems using two approaches: DC-side vs. AC-side faults, and soft vs. hard faults. It then discusses data availability and processing, highlighting the importance of publicly accessible, cost-effective datasets and suitable data processing methods. Traditional statistical algorithms based on cost-effective data are examined in detail, with an emphasis on their practical applicability. In addition, machine learning-based and edge computing algorithms are critically reviewed and classified according to data availability and task requirements, with a high-level evaluation of their performance. This methodological review aims to support both industry practitioners and researchers in selecting suitable algorithms based on data availability and specific application purposes. Finally, the limitations of current fault detection and diagnosis methods based on cost-effective data are critically examined, particularly their reliance on small-scale or laboratory-based datasets. Building on this comprehensive high-level review, key challenges, emerging trends, and potential gaps between industrial practice and academic research are identified. At the same time, certain challenges, such as the development of fault libraries, have begun to be addressed through the use of real-world datasets.