Photovoltaic Fault Detection and Diagnosis: Which Level of Granularity for PV Modeling?

Abstract

The global photovoltaic annual capacity has increased by around 100 GW (+25%) in one year, reaching 505 GW at the end of 2018. However, photovoltaic modules are installed outdoors and subjected to harsh environmental conditions. These can greatly reduce their performance. Hence, a solution to decrease those losses may be the systematic use of health monitoring systems in photovoltaic (PV) installations which can detect and diagnose faults. The diagnostic methods can be based on comparisons between what an installation should produce and what it actually produces. Production prediction can be carried out either based on historical operating data or detailed modeling. In the latter case, the model must reproduce the operation of the system in real conditions as accurately as possible. The purpose of this research is to study and evaluate four different granularity levels used to model photovoltaic modules: modeling at (i) cell level, (ii) semi-string level, (iii) string level and (iv) the whole module. All these models are based on a mathematical expression derive from the single diode equivalent circuit, implemented in Matlab. After an experimental validation, the proposed method has been tested on the healthy case and shadings faulty cases. Total and partial shadings had been studied. Moreover, a method to identify abnormal aging due to a decrease of the shunt resistance is proposed, using shading associated with cell level modeling. In this paper, we have shown that modeling at module or string level could be used for detection but the identification may require lower granularity, modeling at the cell level.