Anomaly diagnosis of photovoltaic power generation system through fusing time series data with its same-source images

With the increasing use of photovoltaic (PV) systems in the global energy transition, accurate anomaly diagnosis is critical for ensuring stable and efficient operation. However, traditional time series methods often suffer from limited feature extraction and insufficient information when using single time series signals. In order to overcome these limitations, we propose a novel method through fusing the time series data with its same-source images. This is achieved by combining the Markov Cross-State Transition Field (MCSTF) and Same-Source Spatio-Temporal Fusion Network (SSSTFN). The MCSTF is adopted to capture the global dynamics and local state transition features of PV power and solar irradiance to construct three-channel images for data enhancement. And, the SSSTFN model integrates Spatio-Temporal Fusion Attention Module (STFAM) and Skip-Connection components to fuse spatio-temporal features while preserving spatial information, thereby improving diagnostic performance and efficiency. Experimental results on a multi-class PV anomaly dataset demonstrate that the proposed method can achieve high diagnostic accuracy, outperforming other mainstream models while maintaining high efficiency. These findings validate the scalability and effectiveness of the proposed method in real-world PV anomaly detection scenarios with high accuracy and low latency requirements.