Accurate Linear I-V Extraction of Ideality, Series and Shunt Resistances

Linear regression extraction of photovoltaic performance parameters offers fast and repeatable evaluation of the illuminated current-voltage (I-V ) curve versus diode model fitting. However linear fitting gives systematically inaccurate series (R
_s
) and shunt resistance (R
_sh
) values and does not compute the ideality factor (n). Here we demonstrate models for predicting n and R
_s
with improved accuracy over basic regression-extracted I-V features. Features are extracted with linear regression for a dataset of over 1 million I-V curves randomly generated with the single-diode model to simulate the operation of silicon PV modules and cells with a wide variety of sizes, performances, degradation states, and operational conditions. The I-V feature values estimated with linear regression were then used to construct generalized multivariate linear models for the actual single-diode values of n and R
_s
, based on our training set of over 800,000 I-V curves. We demonstrate generalized prediction of n and R
_s
with average absolute errors of 13.68% and 17.85%, respectively on our testing set of over 200,000 I-V curves, using only linear regression-extracted I-V parameters. Accuracy of R
_s
values predicted with this method is highly improved, while prediction of n has not been previously performed with this method.