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

Abstract

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 (Rd), 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 Rd is 7.7 %/year. Two non-linear failure mode-based Rd 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 %.