Power-SAM: An Advanced Abnormality Change Detection Algorithm for Intelligent Monitoring in Power Equipment

Timely detection and management of abnormalities are critical to ensuring the safe and stable operation of the smart grid. To address this, a novel abnormal change detection (ACD) model, termed Power-SAM, is proposed for smart grids based on Artificial Intelligence. Firstly, integrate the strong generalization capabilities and efficient image segmentation performance of the Segment Anything Model (SAM) while employing fine-tuning strategies to enhance its applicability for processing images in power scenarios. Secondly, the Time Difference Feature Generator (TDFG) was designed to generate feature vectors suitable for ACD, allowing SAM to adapt to power ACD tasks. Thirdly, a feature fusion network, which combined feature aggregation with differential computation, was adopted, enhancing the model’s ability to detect subtle changes in power scenes. Lastly, CBAM-FPN by integrating Feature Pyramid Network (FPN) and lightweight Convolutional Block Attention Module (CBAM) was introduced to Power-SAM, improving ACD performance for multiscale targets. An image dataset for substation change detection was constructed from real-world power scenarios and ACD requirements. This dataset was used to evaluate the performance of the proposed Power-SAM model. The results demonstrate that the algorithm achieves precision, recall, F1-score, and IoU metrics of 95.25%, 93.35%, 94.41%, and 89.41%, respectively, outperforming current state-of-the-art artificial intelligence models.