| name | pinn-small-signal-stability-multi-inverter |
| description | Physics-informed neural network for small-signal stability analysis in multi-inverter power systems — predicts poles/residues of whole-system impedance across full operating space, identifies oscillation risks and optimal generation distribution. |
| metadata | {"arxiv_id":"2607.07523","published":"2026-07-08","authors":"Hanxi Chen, Xiangyu Meng, Jianhong Wang, Yue Zhu","tags":["PINN","small-signal-stability","multi-inverter","power-systems","systems-engineering","impedance-model"]} |
PINN Small-Signal Stability for Multi-Inverter Power Systems
Core Concept
Traditional whole-system impedance models for multi-inverter power systems are limited to small neighborhoods around steady-state operating points due to linearization assumptions. This paper develops a dedicated PINN that:
- Trains on step-response data from limited EMT (electromagnetic transient) simulations
- Predicts poles and residues of whole-system impedance/admittance across the full operating space
- Characterizes how impedance evolves with power flow variations
- Reveals oscillation risks and their root causes under time-varying conditions
Key Innovations
- PINN training: Uses step-response data from limited EMT simulations rather than analytical models
- Full operating space coverage: Predicts transfer functions beyond linearization neighborhood
- Oscillation risk visualization: Direct visualization of possible oscillatory modes under given power flow
- Optimal generation distribution: Enables optimal dispatch while maintaining safe operation
Architecture Pattern
EMT Simulations (limited set)
↓ step-response data
PINN Training
↓ poles + residues prediction
Whole-system Impedance/Admittance Model
↓ across full operating space
Oscillation Risk Detection → Root Cause Analysis → Optimal Generation Distribution
Workflow
Step 1: Data Generation
- Run limited EMT simulations at selected operating points
- Extract step-response data (voltage, current, frequency transients)
- Label with operating condition parameters (power flow, loading)
Step 2: PINN Training
- Input: Operating conditions + time/frequency
- Output: Poles and residues of transfer function
- Physics constraint: Impedance model must satisfy circuit laws
- Loss = data fit + physics residual
Step 3: Analysis & Deployment
- Predict impedance across full operating space
- Identify oscillation modes (poles near imaginary axis)
- Trace root causes (which inverter/parameter drives instability)
- Compute safe operating regions
Activation Keywords
- PINN power system stability
- small-signal stability multi-inverter
- impedance model neural network
- oscillation risk prediction
- power system transfer function
- 多逆变器小信号稳定性
- 物理信息神经网络电力系统
Related Skills
physics-guided-neural-network - PINN fundamentalshybrid-quantum-classical-pinn - quantum-enhanced PINNsdata-driven-nonlinear-optimal-control-robustness - robustness in data-driven control
Pitfalls
- Limited training data: PINN quality depends on EMT simulation coverage — ensure diverse operating points
- High-dimensional systems: Scalability to large multi-inverter networks requires careful architecture design
- Validation requirement: Always validate PINN predictions against independent EMT simulations at unseen operating points
