Environment-enabled fault diagnosis system for photovoltaic power stations

Photovoltaic (PV) power plants are prone to various faults during long-term outdoor operation, including short circuits, open circuits, module degradation, and partial shading. These faults not only significantly reduce system power generation efficiency but also may trigger equipment damage and even safety accidents. Existing fault detection methods mostly construct static diagnosis models based on electrical parameters alone, resulting in limited adaptability to complex and dynamic operating environments, restricted classification accuracy, and weak physical interpretability of diagnostic features. To address these challenges, this paper proposes an online fault detection and classification method for PV power plants with integrated environmental information. By introducing key environmental factors, such as irradiance and module temperature, the method realizes multimodal perception and modeling of PV system operating states. First, a multidimensional feature space integrating both electrical and environmental parameters is constructed, and the feature distribution of different fault types is analyzed based on kernel density estimation, which reveals the influence mechanism of environmental factors on fault feature representation. Then, a lightweight fully connected neural network model is designed with ReLU activation function and Adam optimization strategy, which enhances the fitting ability for complex nonlinear fault patterns. Experimental results show that the proposed method achieves a classification accuracy of 99.67% on the test set, which verifies the effectiveness of environmental information integration. The proposed method significantly improves the performance of PV fault detection and provides an efficient, robust, and interpretable technical solution for online monitoring and intelligent operation of PV systems in complex field environments.