A Gaussian Process Regression IV Model for PV Outdoor Data

Outdoor data are essential to study the reliability of PV modules and systems. Each electrical performance measure is dependent on the conditions the measurement is conducted at and, therefore, needs to be considered in the context of dynamically changing outdoor conditions. In this paper, we introduce a statistical model designed to analyze PV outdoor data. This model uses a timeseries of current-voltage (IV) characteristics, alongside meteorological data, including plane-of-array irradiance ( $G_{\text{POA}}$) and module temperature ( $T_{\text{Mod}}$). The model aims to utilize all available information to predict the respective performance measure as well as its uncertainty at arbitrary conditions and times. First, to ensure its quality and relevance, a suitable filtering approach is applied to the IV curves, $G_{\text{POA}}$ and $T_{\text{Mod}}$ data from nine modules from five locations (Arizona USA, Germany, India, Italy, and Saudi Arabia) observed for over 2 years. Following this, we utilize the extended solar cell parameters (ESPs), a descriptive model for IV characteristics using 10 parameters. The ESPs, then, undergo a principal component analysis (PCA), which transforms the EPSs into a set of uncorrelated principal components (PCs). Individual Gaussian process regressions (GPRs) are then trained on these principal components (PCs). Once the GPRs are trained, the model is capable of reproducing and predicting the complete IV characteristics at any given time $t$, for specified values of $G_{\text{POA}}$ and $T_{\text{Mod}}$. This prediction includes an assessment of its standard deviation, which is derived from data noise and the distance from the observations. This model serves as a versatile tool for various applications, such as analyzing acclimatization effects, degradation trends, seasonal variations, and the performance ratio (PR) of PV modules or systems.