Measurement of Photovoltaic Device Performance Via Model Parameter Inference
- Thursday, October 15, 2015 from 4:00pm to 5:00pm
- Wilson Hall - view map
Mark Campanelli from Workiva, Bozeman, will give an Applied Mathematics Seminar on Thursday October 15th, 4:10-5:00 in Wilson Hall 1-144.
Photovoltaic (PV) devices convert light into electricity without direct carbon emissions. This talk describes an alternative to the traditional method for measuring the performance of a PV device under standard test conditions (STC). We use Bayesian inference to infer, with quantified uncertainty, the 5 parameters of a lumped-parameter, single-diode circuit-model for PV devices with series and shunt resistance. For reasonably good PV devices and a sufficiently good measurement system, model discrepancy near STC is negligible for this single-diode model. This approach is advantageous because it eliminates irradiance-based corrections to the current values that can introduce voltage-dependent measurement artifacts. However, in addition to measurement noise, the uncertainty from both the measurement-calibration chain and any systematic-error corrections must be considered, and this leads to the simultaneous inference of additional parameters and the potential for parameter non-identifiability. Non-identifiability may not ultimately be an issue for prediction of quantities of interest, but can be problematic in the exchange of supposed unique "best" model parameter values for use in device performance modeling.