Data-driven solar spectral irradiance characterization and application for PV evaluation

Accurately calculating photovoltaic (PV) module efficiency is a key step in solar energy applications. As PV efficiency is influenced by the spatiotemporal variability of the solar spectrum, it is essential to employ local spectra as a replacement for the single AM1.5 G standard spectrum to assess PV modules’ performance. Focusing on the PV-relevant spectral range (280–1400 nm), a total of 186,357 valid solar spectra measured in Beijing over one year were compiled, and a data-driven approach was used to establish a set of five local reference (LR) spectra. An autoencoder was used to reduce spectral dimensionality to five, and k-means clustering (k = 5) was applied to group the spectra, with the cluster-centroid spectra adopted as the LR spectra. Cluster 1 deviated most from AM1.5 G in band-integrated irradiance, with a maximum difference of 77.14%. APE increased from 1.713 eV for AM1.5 G to 1.81 eV for Cluster 5, corresponding to a maximum relative change of 5.66% among the five LR spectra. For common PV materials with different bandgaps, the spectral factor under the LR spectra versus AM1.5 G differed by up to 9.6%, and AM1.5 G tended to overestimate PV efficiency for most of the time in Beijing. The proposed LR spectra provide a practical regional spectral baseline for PV performance evaluation. • Long-term spectra measurements were conducted to capture spectral characteristics. • The significant difference between the spectra in Beijing and AM1.5 G was detected. • Local reference spectra were generated by data-driven techniques. • Most PV modules’ performance is overestimated by the AM1.5 G spectrum in Beijing. • APE can’t characterize the impact of the spectrum on PV modules alone.