This repository contains the source code, trained models, and experimental data for the paper:

Hybrid Intelligence for Topology Anomaly Detection and Correction in Power Distribution Networks Submitted to Frontiers of Computer Science (FCS), 2026

We propose a three-layer hybrid intelligence framework that integrates symbolic rule reasoning, physics-informed state estimation, and graph neural network (GNN)-based adaptive detection for comprehensive topology anomaly detection and correction in power distribution networks. The framework achieves 78.1% global anomaly recall (83.5% with improved GNN) across 30 test networks spanning 3 to 1,888 buses.

鈹溾攢鈹€ paper/                  # LaTeX source and figures
鈹?  鈹溾攢鈹€ main.tex           # Paper source
鈹?  鈹溾攢鈹€ ref.bib            # Bibliography (63 references)
鈹?  鈹溾攢鈹€ figures/           # All figures (PDF + PNG)
鈹?  鈹斺攢鈹€ FCS_Highlights_3pages.pptx
鈹溾攢鈹€ src/                   # Source code
鈹?  鈹溾攢鈹€ anomaly_detection/ # Core detection engine
鈹?  鈹溾攢鈹€ correction_engine/ # Correction logic
鈹?  鈹溾攢鈹€ data_preprocessing/# Data pipeline
鈹?  鈹溾攢鈹€ api/               # FastAPI backend
鈹?  鈹溾攢鈹€ visualization/     # D3.js frontend
鈹?  鈹溾攢鈹€ utils/             # Utility functions
鈹?  鈹溾攢鈹€ config.py          # Configuration
鈹?  鈹斺攢鈹€ run_mvp.py         # Main entry point
鈹溾攢鈹€ models/                # Trained GNN models
鈹?  鈹溾攢鈹€ gnn_model_best.pt  # Best GNN model
鈹?  鈹溾攢鈹€ gnn_binary_best.pt # Binary classifier
鈹?  鈹斺攢鈹€ gnn_gae_model.pt   # Graph autoencoder
鈹溾攢鈹€ data/
鈹?  鈹溾攢鈹€ benchmarks/        # Benchmark results (JSON)
鈹?  鈹斺攢鈹€ experiments/       # Paper experiment data
鈹溾攢鈹€ tests/                 # Test suite (pytest)
鈹斺攢鈹€ docs/                  # Documentation

# Install dependencies
pip install -r src/requirements.txt

# Run the MVP
python src/run_mvp.py

# Run tests
pytest tests/ -v

• PandaPower (46 networks)
• SimBench (123 networks)
• IEEE PES Test Feeders
• CIGRE Benchmark Systems
• Chinese distribution network models (14 models)

• Chinese 10kV Distribution Networks: 14 models covering urban, suburban, rural, industrial, and DER-integrated configurations. See data/networks/.
• Benchmark Results: Complete experimental results for RQ1-RQ9. See data/experiments/.
• Trained Models: GNN model weights (.pt format). See models/.

PandaPower (97 built-in networks), SimBench (123 networks), IEEE PES test feeders, and CIGRE test networks are publicly available. See data/README.md for details.

@article{topology_anomaly_2026,
  title={Hybrid Intelligence for Topology Anomaly Detection and Correction in Power Distribution Networks},
  author={Hanbo Wang},
  journal={Frontiers of Computer Science},
  year={2026}
}

This project is licensed under the MIT License - see LICENSE for details.