MP-Grid (Power-Grid): Detecting Power Grid Outages with Topological Machine Learning

Official implementation of MP-Grid, a topology-informed learning pipeline for power distribution outage detection and localization using multiparameter persistent homology (MPH) and lightweight classifiers.

Paper: MP-Grid: Detecting Power Grid Outages with Topological Machine Learning
Authors: Md Joshem Uddin, Damilola R. Olojede, Roshni Anna Jacob, Baris Coskunuzer, Jie Zhang
Code + datasets + trained parameters: https://github.com/joshem163/Power-Grid

---

Overview

Power distribution outages (from extreme weather, equipment faults, or cyber-physical events) alter both the state and topology of the grid. MP-Grid captures these changes by computing multiparameter persistent homology summaries using:

• Bus voltages as a node filtration function, and
• Branch flows/currents as an edge filtration function,

then vectorizing topological summaries (e.g., Betti-0 signatures over a 2D threshold grid) into a fixed-length feature vector for classification (default: XGBoost).

---

What’s in this repository

At a high level, the repository contains:

• MP-Grid training/inference scripts (e.g., train_MPgrid.py, runtime scripts),
• Baseline models training scripts (e.g., train_baseline.py, metric/runtime scripts),
• Data utilities (e.g., load_data.py),
• Model utilities (e.g., models.py, module.py),
• Notebooks for experiments/visualization,
• Task-specific folders such as Localization/ and topological feature utilities.

---

Datasets

The paper evaluates MP-Grid on IEEE test feeders and realistic synthetic networks, including:

• IEEE 37-bus
• IEEE 123-bus
• 342-node LVN
• IEEE 8500-bus
• NREL synthetic San Francisco Bay Area network (SMART-DS)

Data are generated via power-flow simulation (e.g., OpenDSS) with outage scenarios and (optionally) partial observability.

---

Requirements

The paper reports experiments in Python 3.11.4 and uses a lightweight stack for MP-Grid:

• numpy, pandas
• networkx
• pyflagser (for topological computations, if used in your scripts)
• scikit-learn
• xgboost

Baselines may additionally require:

• torch, torch-geometric (for GNN baselines),
• matplotlib / seaborn (optional plotting).

---

Runing the Experiments

Run the appropriate training script (e.g., train_MPgrid.py) with the correct data path to reproduce the results.