Glossary - Scientific Machine Learning (SciML) for Power Systems (PS)

The same word often means one thing in a power-system course and another in a machine-learning course. This page gives both readings. Symbols and unit rules are on Notation. Textbook depth: Kundur & Malik (2022) for the machine, Bishop & Bishop (2024) for training.

Contributions that add a row, or that correct a translation, are welcome: Join and contribute.

The machine and the grid¶

Term	Plain meaning	Power-systems view	ML view	First used on
Grid	Generators, loads, and lines that must stay in step	The network under study	The plant that generates the data	Power systems for ML people
Generator	A machine that converts mechanical rotation into electrical power	The device whose rotor angle we track	The source of the state trajectory	SMIB dynamics
Infinite bus	A grid so large its voltage and frequency stay fixed	The far end of a SMIB model	A constant boundary condition	SMIB dynamics
SMIB	Single machine, infinite bus	The shared teaching plant in every lab	The data-generating system reused across tasks	Foundations overview
Frequency	How fast the shared electrical rhythm repeats (50 Hz or 60 Hz)	A system operating quantity	A slow variable beside rotor speed	Power systems for ML people
Fault	A short circuit or similar sudden event	The contingency in a stability study	An input event that changes the dynamics	Stability labels and CCT
Clearing	Breakers remove the fault at a chosen time	A protection action and a study parameter	An event time in the input	Lab 1
Synchronism	The machine stays in step with the grid	The operational stability outcome	The positive class, when labels are used	Stability labels and CCT

States, parameters, and units¶

Term	Plain meaning	Power-systems view	ML view	First used on
State	A quantity that moves in time	Rotor angle and speed	The sequence the model predicts	Problem framing
Parameter	A quantity treated as fixed during one simulation	Inertia, damping, network data	A named unknown, separate from network weights	Problem framing
Rotor angle $\delta$	How far the shaft has drifted from the grid rhythm	A state, in radians	One channel of the trajectory	SMIB dynamics
Rotor speed $\omega$	How fast the angle is changing	A state	The other channel of the trajectory	SMIB dynamics
Inertia $H$	Resistance to a change in speed	A machine constant, in seconds	A named estimand in Lab 3	SMIB dynamics
Damping $D$	A term that pulls speed back toward the grid	A machine constant in the swing equation	The other Lab 3 estimand	SMIB dynamics
Per-unit	A scaling so powers and voltages sit near 1	The unit system of the labs	A preprocessing choice that must match at test time	Notation
Operating point	Power, voltage, and equilibrium before the event	The prefault condition	A feature, or a conditioner, of the example	Problem framing
Trajectory	Angle and speed over a time window	A simulation result	The regression target	Reference simulation
Disturbance	The event you apply (fault, clearing time, input)	A study scenario	The axis you split train and test on	Lab 1

Learning and losses¶

Term	Plain meaning	Power-systems view	ML view	First used on
Neural network	A flexible function with many knobs	A curve-fit or a dynamics model, depending on the lab	The parameterized function being trained	Machine learning for power-systems people
Weights	The knobs inside that function	Distinct from $H$ and $D$	The parameters gradient descent updates by default	Machine learning for power-systems people
Loss	One number that is small when the model agrees with the evidence you named	Comparable to a weighted squared error	The training objective	Classical estimation
Data loss	The part of the loss that matches samples	Fit to the measured or simulated trajectory	Supervised error	Forward PINNs
Physics residual	The swing equation evaluated on the network’s prediction	The equation you already trust, written as an error	An extra term in the loss	Forward PINNs
Training	Adjusting knobs so the loss falls	An estimation procedure	Gradient descent for a set number of epochs	Automatic differentiation
Epoch	One pass through the training examples	One sweep of the identification records	One optimization epoch	Machine learning for power-systems people
Surrogate	A learned stand-in for the simulator	A faster model of trajectories or labels	A supervised model	Data and surrogates
PINN	A network trained with data loss and a physics residual	The equation is part of the fit	Collocation-style physics-informed training	Forward PINNs
Neural ODE	A numerical solver whose right-hand side includes a learned piece	The integrator stays visible	A different SciML branch from collocation PINNs	Neural ODEs and hybrid models
Overfitting	A close fit on the training runs and a poor fit on new ones	A model that fails on the next disturbance	Memorization of the training set	Machine learning for power-systems people

Checks and honesty¶

Term	Plain meaning	Power-systems view	ML view	First used on
CCT	Latest clearing time that still leaves the machine in step, for one scenario	A security margin from a stated fault and window	A decision margin for that scenario	Stability labels and CCT
False-safe	Calling a case stable when the reference says it is unstable	The dangerous direction of a stability error	A one-sided classification error	Stability labels and CCT
Identifiability	Whether the measurements can determine the named parameters	Excitation, noise, and model structure	Whether the loss has a unique parameter answer	Identifiability
Physical parameter	The constant in the model you wrote down	Nameplate or model-sheet meaning, with units	Keeps the nameplate meaning when the model matches the plant	Physical vs equivalent parameters
Equivalent parameter	The number that fits the records when the model is incomplete	An effective value, still useful inside its test	A fitted value whose meaning is the fit, not the nameplate	Physical vs equivalent parameters
Scenario split	Train on some disturbances and test on others	Held-out faults or clearing times	The generalization check this book asks for	Lab 1

References¶

Kundur, P., & Malik, O. P. (2022). Power System Stability and Control (Second). McGraw-Hill Education. https://openlibrary.org/isbn/9781260473544
Bishop, C. M., & Bishop, H. (2024). Deep Learning: Foundations and Concepts. Springer. 10.1007/978-3-031-45468-4