Open Access

ARTICLE

An Intelligent Diagnostic Method Based on Automatic Learning of Complex Fault Signatures for Multiple Coupled Faults in Photovoltaic Arrays

Jianjun He^1,*, Hai Zhang¹, Junzhe Tian¹, Hongxiang Luo², Yuexin Du¹, Zhiqing Deng¹
1 School of Energy and Power Engineering, Changsha University of Science and Technology, Changsha, China
2 Hunan Wuling Electric Power Technology Co., Ltd., Changsha, China
* Corresponding Author: Jianjun He. Email:
(This article belongs to the Special Issue: AI and Advanced Computational Techniques for Sustainable Renewable Energy Systems)

Energy Engineering https://doi.org/10.32604/ee.2026.078867

Received 09 January 2026; Accepted 13 February 2026; Published online 17 March 2026

Abstract

To address the limitations of traditional deep learning models in diagnosing multiple coupled faults in photovoltaic arrays—such as manual hyperparameter tuning, susceptibility to local optima, and limited diagnostic accuracy—this study proposes an intelligent diagnostic method. The method integrates an improved sparrow search algorithm (ISSA) with a CNN-Transformer model. First, the original SSA is enhanced by incorporating a Circle chaotic map, a dynamic adaptive weight, and a sine-cosine search strategy to improve its global optimization capability and convergence stability. Second, a CNN-Transformer base model is constructed. This model employs a 1D-CNN to extract local fault features from PV I-V curves, uses a mobile inverted bottleneck convolution (MBConv) to strengthen feature fusion, leverages a Transformer to capture global temporal dependencies, and incorporates residual connections to mitigate gradient vanishing. Finally, the ISSA is applied to optimize the key hyperparameters of the model, resulting in the complete ISSA-CNN-Transformer diagnostic framework. Experimental results on a simulation dataset of a 3 × 3 PV array built in MATLAB demonstrate the effectiveness of the proposed approach. The ISSA outperforms PSO, WOA, and the original SSA in convergence precision and speed on benchmark functions. The ISSA-CNN-Transformer model achieves an overall diagnostic accuracy of 98.75%, representing a 2.25% improvement over the unoptimized model, with a 4.41% accuracy gain specifically under triple-fault coupling conditions. Moreover, it maintains an accuracy of 94.25% at a signal-to-noise ratio of 2 dB, showing significantly superior noise resistance compared to benchmark models such as ResNet-108 and SOM-BP. In conclusion, the ISSA-CNN-Transformer method improves the accuracy and robustness of multiple coupled fault diagnosis in PV arrays to a certain extent. It offers a promising technical approach for PV array condition monitoring and fault diagnosis, exhibiting potential application value for ensuring the safe and stable operation of photovoltaic systems.

---

Keywords

Photovoltaic array; multiple coupled faults; fault diagnosis; improved sparrow search algorithm (ISSA); CNN-transformer; hyperparameter optimization

---