Deep Learning-Based Intelligent Recognition Method for Power Grid Station Wiring Diagram via Multi-Module Collaboration

Abstract

Aiming at the engineering pain points in the intelligent recognition of power grid station wiring diagrams, including missed detection of tiny electrical symbols, low recognition presision of non-standard text, incomplete extraction of wiring line segments and disordered topology reconstruction, this paper proposes a multi-module collaborative deep learning scheme for intelligent recognition and topology reconstruction. The open-source YOLOv11 model from Ultralytics is improved to enhance the detection accuracy of tiny electrical symbols, and the Focal loss function is introduced to alleviate sample class imbalance. The PP-OCRv5 model is fine-tuned via transfer learning with layer-wise unfreezing to improve the recognition effect of non-standard power text. The EDLine-based wiring segment extraction algorithm is improved by combining binarization and morphological skeletonization to fully restore broken wires. NetworkX is adopted to design pixel-level fusion topology reconstruction with multi-source features, and finally output the power topology in JSON format conforming to the CIM standard. Experimental results show that compared with the traditional step-by-step recognition scheme, the proposed method achieves a relative improvement of 2.17% in mAP@0.5 for tiny electrical symbols, an increase of 12.64 percentage points in the recognition precision of non-standard power text, and increases of 4.5 and 5.5 percentage points in the precision and recall of wiring recognition respectively. The topology reconstruction error rate is reduced to 1.2%, invalid isolated island topologies are completely eliminated, and the compliance rate of CIM modeling reaches $\mathbf{9 9. 4 \%}$. The proposed algorithm has significant anti-interference performance and applicability against noise interference from non-standard drawings, and is suitable for processing large quantities of stock station wiring diagram and tasks related to digital power grid modeling.