Physics-Informed Neural Network Models for EMT Simulators
Nadal, Darii, Aristidou, Bakhshizadeh, Nellikkath, Vilmann · eess.SY,cs.SY · 2026-07-25 · 原文 · 证据态 未学
This is the first paper, to the best of our knowledge, to propose a framework that integrates Physics-Informed Neural Network (PINN) models in Electromagnetic Transient (EMT) simulations. Both industry and EMT simulation tools face emerging challenges, as the share of inverter-based resources in power grids rapidly grows. Vendors need to protect the Intellectual Property (IP) of their control design when sharing their component models with customers and system operators. EMT simulation tools need to increase computation speed while maintaining numerical stability. NN surrogates offer numerically stable models with higher degrees of IP protection, as they do not expose the underlying model structure and control parameters. We show that NN surrogates can accurately capture the dynamic behavior of both open-loop and closed-loop controllers, producing identical results in both a validated in-house EMT solver and commercial tools, such as PSCAD. Compared with the artificial delay methods used by commercial solvers, we show that NN surrogates are more numerically robust. Compared with iterative numerical solvers, we show that NN surrogates replacing components with algebraic loops (e.g.,
讲义
讲义·推断 依据「原文」自动生成的结构化摘要（推断），非原文表述；以原文为准。
1. 人话版
This is the first paper, to the best of our knowledge, to propose a framework that integrates Physics-Informed Neural Network (PINN) models in Electromagnetic Transient (EMT) simulations.
Both industry and EMT simulation tools face emerging challenges, as the share of inverter-based resources in power grids rapidly grows.
2. 领域脉络
本文类目：eess.SY、cs.SY，属于其所在研究脉络的最新进展。
3. 机制拆解
Vendors need to protect the Intellectual Property (IP) of their control design when sharing their component models with customers and system operators.
EMT simulation tools need to increase computation speed while maintaining numerical stability.
NN surrogates offer numerically stable models with higher degrees of IP protection, as they do not expose the underlying model structure and control parameters.
4. 证据与数字
摘要未给出量化结果——留意原文的实验与数据。
5. 反例与边界
摘要未声明局限与反例——这是需要警惕的信号，精读时先问边界。
6. 跨领域连接与意外收获
横跨 2 个类目（eess.SY、cs.SY），关注其在你兴趣板块间的迁移面。
7. 可复用方法
把本文机制与你手头项目对照，找一个两周内能验证的最小实验。
8. 术语表
精读时把不熟的术语记入此处，作为下次回忆的锚点。
主动回忆
先合上内容自己复述，再点「显示」核对，然后如实自评。评分即时进 FSRS 排程。
讲义·推断 依据「原文」自动生成的结构化摘要（推断），非原文表述；以原文为准。
1. 人话版
This is the first paper, to the best of our knowledge, to propose a framework that integrates Physics-Informed Neural Network (PINN) models in Electromagnetic Transient (EMT) simulations.
Both industry and EMT simulation tools face emerging challenges, as the share of inverter-based resources in power grids rapidly grows.
2. 领域脉络
本文类目：eess.SY、cs.SY，属于其所在研究脉络的最新进展。
3. 机制拆解
Vendors need to protect the Intellectual Property (IP) of their control design when sharing their component models with customers and system operators.
EMT simulation tools need to increase computation speed while maintaining numerical stability.
NN surrogates offer numerically stable models with higher degrees of IP protection, as they do not expose the underlying model structure and control parameters.
4. 证据与数字
摘要未给出量化结果——留意原文的实验与数据。
5. 反例与边界
摘要未声明局限与反例——这是需要警惕的信号，精读时先问边界。
6. 跨领域连接与意外收获
横跨 2 个类目（eess.SY、cs.SY），关注其在你兴趣板块间的迁移面。
7. 可复用方法
把本文机制与你手头项目对照，找一个两周内能验证的最小实验。
8. 术语表
精读时把不熟的术语记入此处，作为下次回忆的锚点。