Research on Recognition of Substation Secondary Screen Cabinet Wiring Diagrams Based on VLMs

Existing drawing recognition methods struggle with format dependency and limited interpretability, hindering parsing of inconsistent drawings from different institutes. To tackle these issues, this paper proposes a wiring diagram recognition approach leveraging large visual models (such as vision-language models [VLMs]) guided by prompt engineering, enabling analysis of diverse secondary screen cabinet wiring diagrams. Firstly, the influence of prompt engineering styles on VLMs' performance was investigated. Then, using optimized prompts, different VLMs' reliability and parsing accuracy were evaluated. Finally, comparisons with traditional methods and worklist for future engineering applications were made. Results show that, with effective prompt engineering, all tested models demonstrate acceptable drawing recognition ability, with the highest average accuracy reaching 93.48%. However, line complexity and character blurriness introduce interference. Relying on rule-based understanding reduces the reliance on large-scale data training, rendering the method applicable for substation secondary panel wiring diagrams.