Uncertainty quantification in deterministic parameterization of single diode model of a silicon solar cell

摘要

The paper presents a probabilistic approach to study uncertainty quantification in electrical parameters associated with single diode model of a solar cell. Five parameters which governs the electrical behavior of a solar cell needs to be estimated for analyzing the performance of a solar cell. Here, three deterministic techniques for parameter estimation namely Black Widow Optimization, multi-variable Newton Raphson and Particle Swarm Optimization are presented. For verifying the proposed techniques, Solarex MSX83 experimental (Current-Voltage) I-V data is considered. I-V and (Power-Voltage) P-V curves are plotted under standard test conditions (1000 W/m(2), 1.5 A.M. spectrum, 25 degrees C) for three deterministic techniques and are compared with experimental curves. Black Widow Optimization shows increased convergence speed and accuracy. Additionally, uncertainty quantification analysis is presented to study the behaviour of deterministic response of single diode model under Gaussian random parametric uncertainity. Random samples of uncertain electrical parameters are obtained using Latin hypercube and Monte Carlo sampling methods to estimate the output probabilistic response. It can be observed that the objective function shows a bimodal distribution under random parametric variations primarily due to measurement error, shading losses, surface defects, uneven thermal management and manufacturing issues. Alternatively, Bayesian approach is performed using Markov chain Monte Carlo sampling to estimate the probabilistic distribution of model parameters from given set of observations. To avoid computational complexity, Gaussian process model is also considered using polynomial choas expansion. The expected variance in the model response is statistically explained using multi-variate global sensitivity analysis based on Sobol Indices. Based on computed values of first order, second order and total Sobol indices, it can be concluded that reverse saturation current 'I-O' is the most influential parameter among five electrical parameters and contributes about 68.9% to the variance in model response under random uncertainties.