Core Concepts
Data-driven dynamic equivalence models are crucial for power systems, with NeuDyE and PI-NeuDyE offering innovative solutions.
Abstract
The content discusses the challenges of traditional grid analytics in acquiring accurate power system models due to inaccessible parameters. It introduces NeuDyE and PI-NeuDyE as data-driven dynamic equivalence methods using neural networks. The article details the formulations, training, testing, and results of these methods on the NPCC system. It also explores DP-NeuDyE as a more practical variant that reduces input variables. Extensive case studies validate the effectiveness and scalability of these approaches.
Introduction:
Challenges in acquiring accurate power system models.
Introduction of NeuDyE and PI-NeuDyE for dynamic equivalence modeling.
Problem Formulation:
Partitioning interconnections into internal and external systems.
Formulation of InSys and ExSys components.
ODE-NET-Enabled Dynamic Equivalence:
Necessity of continuous-time learning.
Physics-informed continuous backpropagation technique.
Seen from Driving Port Equivalence:
Algebraic component separation for ExSys modeling.
Formulation of ODE-NET based Driving Port Equivalence.
Case Study:
Algorithm settings for training and testing.
Simulation results validating PI-NeuDyE under various scenarios.
Discussion:
Comparison of training time efficiency between different methods.
Generalizability analysis based on electrical distance.
Conclusion:
Importance of data-driven dynamic equivalence models like NeuDyE and PI-NeuDyE for power systems analysis.
Stats
Recent advancements in Phasor Measurement Units (PMUs) provide rich history measurements [4].
DP-NeuDyE only needs 4 dimensions of InSys features [5].
Quotes
"Physics-Informed Neural Networks are engineered to leverage physical knowledge"
"DP-NeuDyE reduces the number of inputs required for training"