An efficient approach for diagnosing faults in photovoltaic array using 1D-CNN and feature selection Techniques

Diagnosing faults in Photovoltaic (PV) systems is essential for operation and maintenance. Selecting relevant features is necessary for successful fault diagnosis because redundant and irrelevant features reduce fault diagnosing accuracy. This paper proposes a novel and efficient approach to diagnos...

Full description

Saved in:
Bibliographic Details
Main Authors: Yousif Mahmoud Ali, Lei Ding, Shiyao Qin
Format: Article
Language:English
Published: Elsevier 2025-05-01
Series:International Journal of Electrical Power & Energy Systems
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S0142061525000778
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Diagnosing faults in Photovoltaic (PV) systems is essential for operation and maintenance. Selecting relevant features is necessary for successful fault diagnosis because redundant and irrelevant features reduce fault diagnosing accuracy. This paper proposes a novel and efficient approach to diagnosing faults in PV systems. The Feature Selection and Fault Diagnosis (FSFD) method is executed for diagnosing five types of faults in PV array (PVA): partial shading condition, line-line fault, arc fault, open-circuit fault, and degradation fault. Firstly, a PVA modeling method using MATLAB/Simulink is employed to simulate I-V curves and extract their features. Next, a feature permutation technique-based method is proposed for selecting the most relevant features. A simple and accurate one-dimensional convolutional neural network (1D-CNN) model is developed to classify the faults based on the selected features. Finally, a confusion matrix is utilized to evaluate the performance of the trained model. Three datasets of PVAs have been utilized to evaluate the effectiveness of the proposed FSFD method. The results indicate that the FSFD method has effectively identified the best five features out of eight for training the 1D-CNN model. The trained model has achieved diagnosing accuracy rates of 99.85%, 99.73%, and 99.97% in series–parallel PVA, total cross-tied PVA, and series PVA datasets, respectively. The proposed method accurately diagnoses single faults in three PVA configurations. Therefore, we recommend conducting additional studies to improve the proposed method for diagnosing hybrid faults.
ISSN:0142-0615