Photovoltaic system fault diagnosis method based on physics-Constrained causal discovery and causal perception graph neural network

Photovoltaic power generation, as a strategic cornerstone in the global energy metamorphosis, necessitates effective fault diagnostic protocols to ensure secure and efficient system operation. While extant fault diagnostic methodologies have demonstrated considerable utility, they predominantly concentrate on statistical correlations between datasets while neglecting underlying causal mechanisms. This investigation proposes a sophisticated photovoltaic system fault diagnosis framework predicated on causal discovery and graph neural networks, comprising two fundamental modules: the Physical-Constrained Causal Discovery Method, which elucidates nonlinear causal architectures between variables from multi-source heterogeneous data; and the Causal Perception Graph Neural Network, which implements structured information propagation to enhance feature extraction capabilities. Experimental validation, conducted on a comprehensive dataset encompassing both intermediate power point tracking (IPPT) and maximum power point tracking (MPPT) operational conditions, demonstrates that the accuracy under variable operating conditions exceeds 99%. Under constant operating conditions, the model achieves an exceptional average accuracy of 99.46% (IPPT) and 99.76% (MPPT), substantially outperforming contemporary methodologies. This approach introduces a critical causal dimension to photovoltaic system fault diagnostics, establishing a novel technological pathway for enhancing the reliability and operational efficiency of renewable energy infrastructure.