Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Power Electronic Converter Control Framework Based on Graph Neural Networks -- An Early Proof-of-Concept

Darius Jakobeit, Oliver Wallscheid

TL;DR

This paper tackles topology-dependent PEC control by introducing a topology-agnostic framework that encodes netlists as typed bipartite graphs and processes them with a task-conditioned GNN backbone. A differentiable predictive control objective trains a shared meta-controller, with per-switch local control heads enabling scalable distributed actuation across diverse converter graphs. Early validation on 100 buck configurations shows near-optimal voltage tracking compared to online nonlinear programming baselines, with a median gap around 16.7% in the reported scenarios. The results support generalization to broader converter families and tasks and point to future work on sim-to-real deployment, real-time constraints, and robustness to model mismatch.

Abstract

Power electronic converter control is typically tuned per topology, limiting transfer across heterogeneous designs. This letter proposes a topology-agnostic meta-control framework that encodes converter netlists as typed bipartite graphs and uses a task-conditioned graph neural network backbone with distributed control heads. The policy is trained end-to-end via differentiable predictive control to amortize constrained optimal control over a distribution of converter parameters and reference-tracking tasks. In simulation on randomly sampled buck converters, the learned controller achieves near-optimal tracking performance relative to an online optimal-control baseline, motivating future extension to broader topologies, objectives, and real-time deployment.

A Power Electronic Converter Control Framework Based on Graph Neural Networks -- An Early Proof-of-Concept

TL;DR

This paper tackles topology-dependent PEC control by introducing a topology-agnostic framework that encodes netlists as typed bipartite graphs and processes them with a task-conditioned GNN backbone. A differentiable predictive control objective trains a shared meta-controller, with per-switch local control heads enabling scalable distributed actuation across diverse converter graphs. Early validation on 100 buck configurations shows near-optimal voltage tracking compared to online nonlinear programming baselines, with a median gap around 16.7% in the reported scenarios. The results support generalization to broader converter families and tasks and point to future work on sim-to-real deployment, real-time constraints, and robustness to model mismatch.

Abstract

Power electronic converter control is typically tuned per topology, limiting transfer across heterogeneous designs. This letter proposes a topology-agnostic meta-control framework that encodes converter netlists as typed bipartite graphs and uses a task-conditioned graph neural network backbone with distributed control heads. The policy is trained end-to-end via differentiable predictive control to amortize constrained optimal control over a distribution of converter parameters and reference-tracking tasks. In simulation on randomly sampled buck converters, the learned controller achieves near-optimal tracking performance relative to an online optimal-control baseline, motivating future extension to broader topologies, objectives, and real-time deployment.
Paper Structure (15 sections, 27 equations, 3 figures)

This paper contains 15 sections, 27 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Graph Representations of PECs
  3. From netlists to typed bipartite graphs
  4. Graph snapshots
  5. Dynamical state $\bm{y}$ and a measurement-to-graph embedding
  6. Permutation consistency and variable-size topologies
  7. Task-conditioned message passing backbone
  8. Phase A (component $\rightarrow$ net)
  9. Phase B (net $\rightarrow$ component)
  10. Optimal Control and Learning a Topology-Agnostic Meta-Controller
  11. General constrained finite-horizon optimal control problem
  12. Topology-agnostic meta-control policy
  13. End-to-end DPC training
  14. Early validation
  15. Conclusion and outlook

Figures (3)

  • Figure 1: Abstracted high-level view of the proposed GNN-based PEC control framework.
  • Figure 2: Time series of two exemplary buck converters controlled by the same GNN-based meta-controller and comparison to the achievable optimal control (OC) performance.
  • Figure 3: Closed-loop control performance comparison of 100 randomly sampled buck converters considering 10 step-response reference tracking cases each.