Hybrid PINNs cut electricity system simulation times by orders of magnitude

Joseph Nyangon's state-of-the-art review, published on arXiv, dives deep into hybrid physics-informed machine learning (PIML) for electricity systems. The paper systematically evaluates architectures including physics-informed neural networks (PINNs), Deep Operator Networks (DeepONets), Fourier Neural Operators, Extreme Learning Machine-enhanced PINNs, graph-based PINNs (PIGNNs), and domain-decomposition PINNs. Each is analyzed through case studies on field analysis, fault detection, digital twins, surrogate modeling, and control optimization. Key finding: embedding Maxwell's equations and other first-principles constraints significantly boosts predictive accuracy under sparse and noisy data, while cutting simulation time by orders of magnitude relative to traditional finite element methods.

Beyond performance gains, the review highlights PIML's ability to enhance generalization across operating regimes and support real-time digital-twin calibration with uncertainty quantification. Persistent challenges remain, such as training instability for stiff multi-scale problems and the lack of standardized benchmarks. Nonetheless, the findings position PIML as a paradigm shift from opaque data-driven methods to transparent, physics-constrained approaches, promising more resilient and intelligent electricity grids. This 59-page review with 6 figures serves as a critical resource for researchers and engineers building next-generation energy systems.