Faults detection and diagnosis of photovoltaic systems using artificial intelligence tools

Abstract

In this thesis, two efficient strategies of faults detection and diagnosis in Photovoltaic (PV) systems are developed. The first strategy uses the probabilistic neural networks (PNN) classifiers to detect and diagnose faults in the Direct Current (DC) side of Grid Connected Photovoltaic (GCPV) systems. The second strategy suggests the development of two statistical methods (such as the improved-ratio and control charts based methods) to detect and diagnose the faults. The improved ratio based method consists on the evaluation of three coefficients to detect and diagnose short-circuits and open-circuits faults. While, the control charts based method applies the exponentially weighted moving average (EWMA) and Shewhart charts to detect and diagnose the faults in GCPV systems. However, the developed strategies require the availability of a high-quality database that describes the system behavior for both healthy and faulty operations. To deal with this concern, a PSIMTM/MatlabTM co-simulation strategy is developed to elaborate a trusted simulation model. This model requires the use of the One Diode Model (ODM) electrical parameters. For this, an efficient strategy, based on the artificial bee colony (ABC) and the best-so-far ABC algorithms, are developed to identify the ODM parameters. Finally, the ODM identified parameters are used to elaborate an efficient strategy of maximum power point (MPP) estimation. The efficiency of the developed strategies is experimentally evaluated by using real measured data, collected from two actual GCPV systems. The first one is a 9.54 kWp PV system located at Algiers (Algeria), while the second one is a 0.9 kWp PV system, located at Jaen University (Spain).