Projected Neural Differential Equations for Power Grid Modeling With Constraints

MCML Authors

Niki Kilbertus

Prof. Dr.

Core PI

Ethics in Systems Design and Machine Learning

Abstract

Neural differential equations offer a powerful approach for data-driven simulation. However, many applications in science and engineering possess known constraints that should be obeyed by the learned model. We introduce projected neural differential equations (PNDEs), a new method for constraining neural differential equations based on projection of the learned vector field to the tangent space of the constraint manifold. In tests on two challenging examples from power grid modeling, PNDEs outperform existing methods while requiring fewer hyperparameters. Our approach demonstrates significant potential for enhancing the modeling of constrained dynamical systems, particularly in complex domains like power grid dynamics where accuracy and reliability are essential.

inproceedings WBG+24b

---

D3S3 @NeurIPS 2024

Workshop on Data-driven and Differentiable Simulations, Surrogates, and Solvers at the 38th Conference on Neural Information Processing Systems. Vancouver, Canada, Dec 10-15, 2024.

---

Authors

A. White • A. Büttner • M. Gelbrecht • N. Kilbertus • F. Hellmann • N. Boers

---

Links

URL

---

Research Area

 A3 | Computational Models