Predictive analysis of power degradation rate in solar PV systems emphasizing hot spots and visual effects-based failure modes

A definite and exclusive long-term PV degradation projection is required to reduce financial vulnerability in the solar photovoltaic plant-based energy transition markets. So, in this paper beside a traditional linear rate approach, a non-linear design tool for quantifying long-term failure mode-based power degradation rate (R-d), is proposed. An on-field experimental investigation incorporating visual effects-based Solar I-V characterization and thermal imaging for a duration of 1 year (February 2022 to January 2023) is carried out on a realistic solar PV system exposed to field conditions for 5-6 years. The experimentally evaluated R-d is 7.7 %/year. Two non-linear failure mode-based R-d models employing machine learning approaches, with differences in inputs are developed and are proven economically useable against the reported assessment techniques. The root mean square error (RMSE) for model 1 and model 2 during testing is 0.0019 and 0.0020 respectively (LSTM based approach) whereas testing error for model 1 and model 2 using back propagation based approach is 0.00182 and 0.00132 respectively. The proposed tool is also validated for a realistic PV plant in Telangana. A closeness of prediction against the actual of 9.0 % is observed for model 1 during validation, whereas the same for model 2 is 9.6 %.