Research on State Evaluation Algorithm for Urban Rail Transit Substation Equipment Based on Improved YOLOv8

The integration of artificial intelligence is transforming substation inspection in the power industry. Accurate equipment defect diagnosis is critical for preventing failures that could disrupt rail transit operations, thereby enhancing grid reliability. However, the diverse equipment morphologies in complex substation environments often cause traditional detection algorithms to underperform on small targets. To address this, we propose a visual fault detection algorithm based on an improved YOLOv8. The model incorporates a Deformable Convolutional Module in its backbone for adaptive multi-scale feature learning and employs the Focaler-IoU loss to refine bounding box regression. Experimental results show our method achieves significant gains in power equipment defect detection: a 3.6% increase in recall, a 4.3% rise in precision, and a 4.4% improvement in , validating the algorithm's practical value for intelligent substation inspection.