Adaptive fault diagnosis method for offshore floating photovoltaic arrays considering wave-induced disturbances

Offshore floating photovoltaic (OFPV) system suffers from harsh environment and frequent occurrence of faults. Wave-induced disturbances reduce the accuracy of existing fault diagnosis methods that are applied to land-based PV systems. Consequently, this paper proposes an adaptive fault diagnosis method for OFPV arrays considering the wave-induced disturbances. A mathematical model of the OFPV array is first established by combining the irradiance of the modules and validated with real data from scaled model testing. Based on this model, the I-V curves of OFPV under different fault conditions are simulated and ten key feature variables are selected to capture the temporal characteristics of evolving faults under wave movements. A method combining piecewise linear regression with curvature analysis is developed to extract important inflection point features. The time series of environmental and electrical features are utilized to diagnose the faults by a CNN-BiGRU-MHA model that combines convolutional neural networks (CNN), bidirectional gated recurrent units (BiGRU) with residual connections, and multi-head attention (MHA) mechanisms. Adaptive sliding time window technique is further incorporated with fluctuation rate clustering and feature cosine similarity to enhance fault features under wave disturbances. The effectiveness of the proposed method is validated via simulated and experimental data obtained from the simulation model and actual PV platform. Compared with the fixed window methods, the adaptive fault diagnosis improves the average F1 scores by 0.51 %–13.20 % across different fault types. The average diagnosis accuracy reaches 99.20 %, outperforming other state-of-the-art PV fault diagnosis models.