IBR Dynamics and Physics-Informed Learning
Physics-Informed Neural ODEs for IBR Dynamics
C3 · Core course · 9 modules
Learn continuous-time dynamics with data, physical constraints, and residual models.
Module framework available; full lessons, runnable experiments, and worked practice are in development.
Prerequisites and learning path
Python, introductory ODEs, and a simple GFM swing model; detailed converter modeling is needed for the advanced project.
Modules
Dynamical Systems and Trajectory Data
Build a simple reference system and construct meaningful training trajectories.
Module outlineNeural Network Foundations
Understand the function approximator used to learn a vector field.
Module outlineNeural ODEs and Learned Vector Fields
Learn a derivative law, then integrate it to predict trajectories.
Module outlineDerivative and Trajectory Matching
Compare noisy derivative labels with gradients through a numerical solver.
Module outlinePhysics Constraints and Residual Backbones
Place physical knowledge in the loss or in the model architecture.
Module outlineContinuous-Time PINNs and Neural ODEs
Distinguish learning a solution function from learning a dynamic law.
Module outlineParameter Identification and Generalization
Separate unknown-parameter estimation from prediction at supplied parameters.
Module outlineStability Diagnostics for Learned Models
Check equilibria, local modes, and finite-horizon recovery before making claims.
Module outlineFrom Swing Models to Converter Dynamics
Plan a learning experiment with converter states, terminal outputs, and events.
Module outline
Course project
Compare pure NODE, grey-box, and physics-aware models on held-out trajectories.
Deliverables
- Trajectory-based train, validation, and test splits.
- Reproducible training with stated observations and losses.
- Rollout and stability diagnostics with honest limitations.