Detecting heterogeneity in PV modules from massive real-world “step” I-V curves: A machine learning approach

摘要

We demonstrate that I-V curves with bypass diodes in forward bias can be useful in learning the heterogeneity in PV modules. In the laboratory-based experiments, we show that heterogeneity in a PV module can be detected from “step” IV curves that are collected under non-uniform irradiance. On the other hand, heterogeneous cell performance can lead to bypassing even under uniform irradiance. This hypothesis was tested using a fabricated 4-cell mini-module with cells that were engineered to have highly heterogeneous front contact resistivity and a SPICE-based circuit model. We find good agreement between the experimentally determined curve and simulations. We illustrate a technique for automatically classifying and analyzing massive real-world I-V curves and for gaining insights into the performance of PV modules. By classifying 3.7 million I-V curves, we demonstrate the occurrence of “step” I-V curves under two irradiance conditions: under non-uniform irradiance condition, mirror augmented PV module in Cleveland, Ohio; and under uniform irraidance condition at the Negev Desert, Israel, Gran Canaria, Spain and Mount Zugspitze, Germany. Under the uniform irradiance conditions, we found that the percentage of “step” I-V curves in all three I-V curve types gradually increase over time. This indicates the electrical characteristics within a PV module change from homogeneous to heterogeneous. Since the “step” I-V curves have a lower maximum power and a lower fill factor than normal I-V curves at the same irradiance condition, the heterogeneity in I-V module directly cause power degradation.