Fault diagnosis in photovoltaic arrays: A robust and efficient approach using feature engineering and 1D-CNN

• A robust and efficient method for faults diagnosing in PV arrays using a 1D-CNN and feature engineering is proposed. • Feature engineering has been introduced to normalize the I-V curve characteristics based on the PV system scale. • The proposed method has high performance for diagnosing faults in PV arrays in noisy and noiseless data. • The pre-trained model can detect and diagnose faults in different-scale PV arrays without retraining. • Experimental results showed the adaptability of the proposed method across different PV array topologies. Fault diagnosis is essential for maintaining the efficiency of the PV systems and ensuring their safety. With increased research into fault diagnosis techniques, artificial intelligence (AI)-based methods have been widely explored. However, most existing studies have concentrated on series and series–parallel topologies of PV systems, which limits the practical application of these methods to other configurations. Furthermore, the quality of the features used for diagnosis is critical for enhancing the model’s generalization and adaptability across various PV scales and topologies. To overcome these challenges and limitations, this study proposes a robust and efficient method based on feature engineering and one-dimensional convolutional neural networks (1D-CNN). Specifically, a novel feature normalization technique is introduced in the feature engineering phase. The current–voltage (I-V) curve characteristics are normalized based on the PV system scale, creating a standard normalization approach for different scales of PV arrays and improving the generalization capability of the proposed method. The proposed 1D-CNN model consists of two convolutional layers, a single max-pooling layer for feature extraction, two fully connected layers, and a softmax layer for fault classification. Performance evaluations of the proposed method were conducted across four PV system topologies: series–parallel (SP), Total-Crossed-Tied (TCT), Honey-Comb (HC), and Bridge-Linked (BL) PV arrays. The experimental results demonstrate that the proposed method performs exceptionally well in diagnosing five types of faults in clean and noisy data. Moreover, the pre-trained 1D-CNN model can diagnose faults across different PV array scales without requiring retraining, highlighting the model transferability in the proposed method.