Manifold Learning in Power System Transient Stability Assessment (TSA)
Following embedding or dimmensionality reduction methods are examined:
- Principal components analysis
- Kernelized principal components analysis
- Truncated singular value decomposition
- Isomap embedding
- t-distributed stochastic neighbor embedding
- Locally linear embedding
- Modified locally linear embedding
- Locally linear embedding with local tangent space alignment algorithm
- Locally linear embedding with Hessian eigenmap method
- Spectral embedding
- Multi-dimensional scaling
Supervised and unsupervised training, using cross-validation with stratified shuffle split, with hyperparameters optimization tackled by means of the simulating annealing algorithm.
Dataset is derived from the extensive set of numerical electro-mechanical simulations of the IEEE New England 39-bus benchmark electric power system.