Microgrids · Distribution systems
Physics-informed GNN for microgrid N-1 security prediction
This course connects power flow and machine learning to physics-informed graph neural networks and microgrid N-1 security prediction. Start with balanced and unbalanced power flow, build real pandapower networks, learn MLP training and graph message passing, then train a GNN with AC residuals and screen individual line outages.
Balanced Power Flow
Interactive lesson · Microgrids / Distribution systems
Explore balanced three-phase AC power flow through background, equation derivations, a worked feeder example, live parameter controls, and editable Python code.
Unbalanced Power Flow
Interactive lesson · Microgrids / Distribution systems
Derive a three-phase, four-wire feeder model and explore unequal loads, single-phase PV, neutral displacement, sequence components, and editable Python.
PandaPower-based Implementation
Interactive lesson · Real pandapower / Python
Build balanced and three-phase networks, change parameters, run editable Python, inspect result tables, and screen line outages. Download a self-contained Notebook for further experiments.
Open PandaPower-based Implementation →
AI Preliminary · Neural Network Basics
Interactive lesson · Module 4 / Supervised learning
Start with a single-neuron slider experiment, derive a one-hidden-layer MLP and its gradients, train and replay a sine-regression model, and connect the workflow to grid data. Includes editable Python, original slides, and a self-contained Notebook.
GNN Preliminary · Graphs and Message Passing
Interactive lesson · Module 5 / Graph learning
Encode a five-bus microgrid with node and edge features, inspect adjacency and computational arcs, step through message passing, and train a small end-to-end GNN. Change loads and switches, compare a linear voltage reference, and run editable Python or the self-contained Notebook.
Physics-informed GNN · Microgrid N-1 Security
Interactive lesson · Module 6 / AC physics and graph learning
Derive AC mismatch and a normalized joint loss, inspect voltage perturbations, train matched data-only and physics-informed GNNs, and screen line outages for supply, voltage, and current violations. Includes raw-versus-corrected comparisons, editable Python, and a standalone Notebook.