Research map / Graph learning / Graph self-supervised learning
Disentangled Table-Graph Representation for Interpretable Transmission Line Fault Location
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes DTG4Power, an interpretable disentangled graph (line-graph) framework that jointly locates faulty transmission lines and fault points from tabular measurements on 7-, 36- and 325-bus power systems.
Problem
Fault location in power grids relies on tabular state records that ignore transmission line topology, forcing multi-stage pipelines that separately locate fault lines and fault points.
Approach
Converts bus-level graphs into line graphs whose nodes are transmission lines, then uses a GAT-style disentangling network that learns interpretable edge-attribution coefficients to build factor subgraphs, and a self-attention head that weights lines to jointly predict fault line (classification) and fault point (regression) end-to-end.
Key results
- On the 325-bus system, DTG4Power reaches 0.988 fault point accuracy and 0.998 fault line accuracy, versus 0.975 for MLP at fault point on the same system.
- On the 7-bus and 36-bus systems, DTG4Power attains 0.990 and 0.996 fault point accuracy, and 1.000 fault line accuracy.
- Mixed training across fault lines achieves 0.988 Single-SC and 0.974 Three-TG accuracy in Scenario 1 on the 325-bus system, comparable to or above separate training.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use self-attention mapping instead of direct mapping (ablation) | DTG4Power · fault location | accuracy | Helped |
| Use self-attention mapping instead of fully connected (MLP) mapping (ablation) | DTG4Power · fault location | accuracy | Helped |
| Train on mixed fault lines instead of training separately per line (ablation) | DTG4Power · 325-bus, fault location | accuracy: Separate: 0.944; Mixed: 0.988 (Single-SC, S-1) | Helped |
| Increase the fault-location error interval (ablation) | DTG4Power · 325-bus, multiple scenarios | accuracy | Helped |
| Vary the number of factors (ablation) | DTG4Power · fault location, fault types 1 and 4 | accuracy | Mixed |
Surprising findings
- Lines with high attention mostly cluster around faulty lines, while faulty lines themselves are not necessarily high-attention lines.
- The MLP baseline performs strongly, outperforming the GCN baseline on fault point location in several settings.
Limitations
- Scarcity of real fault data; evaluation relies on PSASP simulation.
- Sensitive to number of factor graphs, requiring per-dataset tuning.
- Noisy data robustness not addressed.
Experimental setup
Datasets: PSASP-simulated fault data (7-bus, 36-bus, 325-bus systems)
Benchmarks: PSASP-simulated 7-bus, 36-bus, and 325-bus fault datasets
Metrics: accuracy (within 5% error range)
Baselines: KNN Random Forest MLP GCN
Builds on: Graph Attention Networks line graph Graph Convolutional Network
Closest papers by meaning
- PowerGraph: A power grid benchmark dataset for graph neural networks (NeurIPS 2024 · similarity 0.83)
- Power Grid Anomaly Detection via Hybrid LSTM-GIN Model (Student Abstract) (AAAI 2024 · similarity 0.82)
- A Simple yet Scalable Granger Causal Structural Learning Approach for Topological Event Sequences (NeurIPS 2024 · similarity 0.81)
- Graph Neural Networks for Edge Signals: Orientation Equivariance and Invariance (ICLR 2025 · similarity 0.80)
- Disentanglement-Guided Spatial-Temporal Graph Neural Network for Metro Flow Forecasting (Student Abstract) (AAAI 2024 · similarity 0.80)
- Functional Matching of Logic Subgraphs: Beyond Structural Isomorphism (NeurIPS 2025 · similarity 0.79)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.