Development of a multi-output feed-forward neural network for fault detection in Photovoltaic Systems

Abstract

The main role of a fault detection and identification technique is to interpret what is​ causing a PVS’s real-time energy production to drop. To keep photovoltaic (PV) systems performing at their best, fault detection is crucial. Various fault detection systems have been proposed in the literature over the last decade. Some of the fault detection methods presented solely used data from I–V curves, while others relied on ANN techniques to find faults. This paper presents the development of a multi-output Artificial Neural Network (ANN), for fault detection and identification on the DC side of a Photovoltaic System (PVS). Measurements and benchmarking were carried out in order to validate the algorithm’s capacity to detect both normal operation at maximum power and the ability to determine and identify faults introduced during the experiment’s procedure. It is observed that the classification output, compared with the regression output, provides a better assessment of the operational status of the photovoltaic cell. The developed method has a 93.4 percent accuracy in detecting open-circuit, short-circuit, and mismatch faults, using the classification output, while the regression output provides an operational status estimation with mean absolute error (MAE) equal to 0.305. When compared to other implementations, the findings are encouraging, indicating that the approach could be used in PVS.

Keywords

• Photovoltaic Systems
• Photovoltaic fault detection algorithms
• I–V curves
• Artificial Neural Networks