Tools to Explain Neural Networks for Power System Dynamics

2026-08-08

AI Summary

This study addresses the limited engineering applicability of machine learning surrogate models in power system dynamic simulation due to their lack of physical interpretability. For the first time, neural tangent kernel (NTK) theory is introduced into this domain, integrated with small-signal eigenvalue analysis to establish a formal connection between system physical stiffness and neural network optimization stiffness. This linkage reveals the evolution mechanism of error modes during training, enabling the development of an adaptive loss weighting strategy. The proposed approach not only provides modal-level physical explanations for performance differences across network architectures—such as ActNet—but also significantly enhances training convergence and reliability. Consequently, this work lays a theoretical foundation for designing interpretable, structure-aware surrogate models tailored to power system dynamics.

Abstract

This paper presents, for the first time in power systems literature to our knowledge, analytical tools to explain the training performance of machine learning surrogate models for power system dynamics. Power system simulations are increasingly challenged by stiff and multi-timescale dynamics arising from converter-interfaced resources and fast control loops. Machine learning surrogates emerge as promising tools to handle this complexity and accelerate dynamic simulations. However, their performance remains difficult to interpret, which limits their adoption. Building on the small-signal eigenvalue analysis in power systems, this paper uses the Neural Tangent Kernel (NTK) method. NTK delivers a modal interpretation of the learning performance, identifying error modes that decay rapidly versus others that converge slowly. This connection explains how physical stiffness and timescale separation in power system dynamic models appear as optimization stiffness during Neural Network (NN) training. Based on this analysis, we develop adaptive loss-weighting strategies to improve and explain why structure-aware neural architectures, such as ActNet, perform better than vanilla NNs. We assess the proposed approach on physics-informed machine learning surrogate models of \acp{SM} and power electronic converters. The methods introduced in this paper can deliver the necessary analytical tools to interpret and improve the performance of machine learning surrogates, paving the way for the systematic, physics-aware design of NN architectures and training strategies. By moving beyond trial-and-error development, these tools reveal training dynamics and failure modes, support more reliable design decisions, and strengthen confidence in machine-learning surrogates for engineering applications.

Problem

• machine learning surrogates
• power system dynamics
• training interpretability
• neural networks
• simulation acceleration

Innovation

• Neural Tangent Kernel
• power system dynamics
• machine learning surrogates
• adaptive loss weighting
• physics-informed neural networks