Graph Neural Ordinary Differential Equations for Power System Identification

TL;DR

This paper introduces message-passing graph neural ODEs (MPG-NODEs) for power system identification, enabling flexible, data-driven modeling of complex, heterogeneous power networks with transfer learning capabilities.

Contribution

The work extends graph NODEs with message-passing and novel features to improve power system modeling, especially for heterogeneous and evolving networks.

Findings

MPG-NODEs outperform monolith NODEs in power system identification.

The method enables transfer learning for network modifications with minimal retraining.

Case study on IEEE 9-bus system demonstrates improved flexibility and accuracy.

Abstract

With the shift towards decentralized energy generation, the increasing complexity of power systems renders physics-based modeling challenging. At the same time the growing amount of available measurement data opens the door for obtaining models in a data-driven manner. A modern method to do so are neural ordinary differential equations (NODEs), offering a framework for continuous time system identification. Recent extensions, so called graph NODEs impose a structural inductive bias that has the potential to improve generalization of the learned representation. In this work, we employ graph NODEs and extend them with novel ideas to develop message-passing graph NODEs (MPG-NODEs) for identification of coupled systems with heterogeneous node dynamics and edge couplings. This encompasses state-of-the-art machine learning architectures to infer latent representations of unmeasured states from past measurements, local node and edge embeddings to account for heterogeneity as well as an autoregressive scheme to allow for piecewise constant control inputs. We apply MPG-NODEs to identify voltage and frequency dynamics of power systems and compare them to a monolith NODE under identical measurement assumptions. Our case study on the IEEE 9-bus system indicates that the proposed MPG-NODE offers a much more flexible framework with transfer learning options that allow to add or remove powerlines and units with little to no retraining.