The European Union has set ambitious climate goals for 2050, with a significant emphasis on the contribution of renewable energies, such as photovoltaics, to meet the future energy demand. Recently, the Italian Ministry of Environment and Energy Security submitted a proposal for updating the National Integrated Energy and Climate Plan to the European Commission, foreseeing a substantial increase in photovoltaic installations with a total capacity of 80 GW and an energy production of approximately 100 TWh.

These objectives require an expansion in both the installed capacity of photovoltaic systems and actual energy production. Maintaining high performance for both existing and new installations is essential to achieve these goals. The rapid growth of PV installations poses a challenge for PV plant operators, as they need to manage portfolios of installations on the scale of gigawatts, demanding significant human and financial resources. In this context, the digitalization and robotization of photovoltaic plant inspection activities play a crucial role in reducing Operation & Maintenance (O&M) costs.

Identification and diagnosis of faults through Machine Learning-based algorithms provide valuable information to O&M operators, enabling them to promptly address issues and optimize time and repair costs. This work aims to develop tools to support the O&M sector in maximizing energy production. Through the RSE’s PV Fault Facility, large datasets were generated with faulty conditions over a span of more than 4 years, resulting in over 1.6 million fault-labeled records, for the study of combined faults and partial shading characteristics.

The collected data will be made available to the PV scientific community through a data repository designed during this work. The PV dataset series have also been instrumental in the development of combined fault diagnosis models and the development of retraining techniques for diagnosis models to maintain their high accuracy even in the presence of degradation caused by the natural aging of technology. The combined fault classifier has been able to correctly identify the type of fault in approximately 100% of the considered observations, while the use of a change detector has resulted in an R2 above 0.90 for all years examined.

The Report is available on the Italian site