Dynamic baseline-based fault diagnosis method for photovoltaic arrays integrating degradation modeling and CNN-transformer

Photovoltaic (PV) arrays operating over long periods are affected by coupled interference from performance degradation and multi-source faults. This leads to static diagnostic baselines, distorted fault features, high false alarm rates, and weak detection capability for slowly evolving faults.Conventional methods can hardly satisfy the operation and maintenance demands of large-scale PV plants.To address these issues, this paper proposes a multi-source fault diagnosis method for PV arrays by integrating a performance degradation process model with a CNN-Transformer hybrid network. First, a dynamic performance degradation baseline model is constructed using uncertainty theory and nonparametric kernel density estimation. This model evolves over time and is used to correct electrical parameters, thereby eliminating the influence of natural aging. Second, deviations between measured data and the dynamic baseline are calculated to form baseline deviation feature vectors, which enhance real fault signals. Finally, a CNN-Transformer hybrid network is designed. The CNN captures local abnormal features of abrupt faults such as short circuits and open circuits, while the Transformer models long-term temporal dependencies of slowly evolving faults such as degradation, shading, and mismatch. This enables accurate classification of multi-source faults. Experiments based on operational data from a 10 MW PV power plant (2018–2024) show that the proposed method achieves a macro F1-score and balanced accuracy of 97%. This significantly outperforms the conventional CNN-LSTM method (69%), with a false alarm rate below 2.5%. The proposed approach effectively decouples degradation from faults and suppresses false alarms, laying a foundation for reliable intelligent fault diagnosis in large-scale PV systems.