Fault diagnosis of photovoltaic arrays at ports under small-sample and imbalanced data conditions

Photovoltaic (PV) power generation is increasingly deployed in ports to support green and low-carbon development. However, the harsh operating environment and the scarcity of fault data in newly installed PV arrays hinder accurate and reliable fault diagnosis. To address the issues of data imbalance and fault sample scarcity typically encountered during the initial deployment of PV arrays in port areas, this study proposes an enhanced oversampling algorithm, Adaptive K-nearest neighbor and Dynamic Random-disturbance-based Synthetic Minority Over-sampling Technique (AKDRSMOTE), for fault data augmentation. Furthermore, a hybrid strategy based on improved Harris Hawks Optimization (IHHO) optimized support vector machine (SVM) is proposed to enhance the diagnostic performance.Experimental results demonstrate that under small-sample and imbalanced data conditions, the proposed approach effectively identifies various complex PV faults. The model achieves an accuracy of 93.42%, an F1-score of 88.09%, and a Kappa coefficient of 92.89%, all of which outperform traditional fault detection techniques. These findings substantiate the accuracy, robustness, and stability of the proposed method in complex port environments and highlight its strong potential for real-world engineering applications in intelligent PV system operation and maintenance.