Table 1 Comparative analysis of PV inverter fault diagnosis approaches.

Model-based¹⁶ & Signal-based²⁰

Sliding mode observers; extended Kalman filters; Adaptive observers; FFT; STFT; wavelet transform

Provides insightful diagnostics; good at detecting transient faults; effective for periodic fault detection

Requires accurate system models; performance degrades with noise; limited adaptability; struggles with complex systems

Statistical methods²⁵

K-nearest neighbor; Wolf optimization; independent component analysis; random forest; ensemble methods

Robust pre-processing capabilities; good for handling imbalanced data; effective feature extraction; better performance through hybrid approaches

Limited representation learning; may require extensive feature engineering; performance depends on data quality

Deep learning^33,34

1D-CNN; 2D-CNN; hybrid CNN (HCNN); CNN-LSTM; pyramid-structured networks

High diagnostic accuracy; strong pattern recognition; good at handling complex data; real-time detection capability

Requires extensive preprocessing; limited generalization; high computational demands; limited spatial dependency handling