Physics-Informed Neural Network Architecture for Real-Time Dynamic Security Assessment in Low-Inertia Transmission Networks

In power systems, the fast transition towards converter interfaced generations (CIGs) has reduced the inertia available to the transmission system, thus increasing its susceptibility to large disturbances. In low inertia conditions, fault and severe contingencies cause the system to experience sharp frequency fluctuations, rapid rotor angle divergence and poor voltage recovery. Dynamic security assessment (DSA) methods that rely on stiff non-linear differential algebraic equations (DAEs) are computationally expensive and challenging to apply in practice. Purely data driven alternatives often struggle to generalise to unseen operating scenarios. To address these challenges, we propose a physics informed neural network (PINN) that can be used for real time DSA in low inertia transmission networks. This approach includes multi-objective regularisation that incorporates centre of inertia (COI) swing equations, converter current tracking and semiconductor current limits into the loss function using Automatic Differentiation. The network classifies whether a scenario is secure (i.e. transiently unstable is defined as the positive class) and predicts the post-fault trajectory. The ensemble mean accuracy across an adapted IEEE 39 bus, 10 machine benchmark system with renewable penetration levels up to 70% was found to be 98.7% ± 0.2% (98.40% in median representative run). Accuracies in unseen topological N − 2 contingencies were found to be 94.5% ± 0.3%. The average false secure (false negative) rate was found to be 2.22% ± 0.18% (2.22% in representative run). This model can be implemented in real time with inference times of 7.2 ms on a dedicated GPU (21.4 ms estimated contingency screening time). This shows that it could be used for online security screening in the benchmark studied.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-19
DOI
https://doi.org/10.21203/rs.3.rs-11637283/v1
Primary Topic
Power System Optimization and Stability
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Physics-Informed Neural Network Architecture for Real-Time Dynamic Security Assessment in Low-Inertia Transmission Networks

Venkatesh Daggupati
Research Square
Power System Optimization and Stability
preprint

Physics-Informed Neural Network Architecture for Real-Time Dynamic Security Assessment in Low-Inertia Transmission Networks

Venkatesh Daggupati
preprint en

Abstract

In power systems, the fast transition towards converter interfaced generations (CIGs) has reduced the inertia available to the transmission system, thus increasing its susceptibility to large disturbances. In low inertia conditions, fault and severe contingencies cause the system to experience sharp frequency fluctuations, rapid rotor angle divergence and poor voltage recovery. Dynamic security assessment (DSA) methods that rely on stiff non-linear differential algebraic equations (DAEs) are computationally expensive and challenging to apply in practice. Purely data driven alternatives often struggle to generalise to unseen operating scenarios. To address these challenges, we propose a physics informed neural network (PINN) that can be used for real time DSA in low inertia transmission networks. This approach includes multi-objective regularisation that incorporates centre of inertia (COI) swing equations, converter current tracking and semiconductor current limits into the loss function using Automatic Differentiation. The network classifies whether a scenario is secure (i.e. transiently unstable is defined as the positive class) and predicts the post-fault trajectory. The ensemble mean accuracy across an adapted IEEE 39 bus, 10 machine benchmark system with renewable penetration levels up to 70% was found to be 98.7% ± 0.2% (98.40% in median representative run). Accuracies in unseen topological N − 2 contingencies were found to be 94.5% ± 0.3%. The average false secure (false negative) rate was found to be 2.22% ± 0.18% (2.22% in representative run). This model can be implemented in real time with inference times of 7.2 ms on a dedicated GPU (21.4 ms estimated contingency screening time). This shows that it could be used for online security screening in the benchmark studied.

Research Square
Texas Tech University (US)
Power System Optimization and Stability
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon