T20.2 - Graph Neural Network based Performance Modeling for the Dual Active Bridge Converter with Operational Generalization

Wednesday, March 19, 2025

1:50 PM – 2:10 PM ET

Location: Level Three, A312

Technical Lecture Presenter(s)

Fanfan Lin, PhD (she/her/hers)

Zhejiang University

Presentation Paper

Co-Author(s)

Weihao Lei, MA (he/him/his)

Zhejiang University
XB

Xiaokun Bao

Zhejiang University
XL

Xinze Li

University of Arkansas
XZ

Xin Zhang

College of Electrical Engineering, Zhejiang University

Existing Artificial Intelligence (AI) assisted performance modeling approach for the power converter generally has limited generalization capability, restricted by the training data. To address this challenge, this paper proposes an innovative graph-neural-network (GNN)-based modeling approach, leveraging on both GNN and Multi-Layer Perceptron. Taking the dual active bridge converter as a design case, the proposed performance modeling approach can be generalized to different operational conditions and modulation strategies. And the mean absolute percentage error for generalizing to out-of-domain cases is no higher than 1% compared to the models specifically trained for those domains. The feasibility of the proposed approach is also validated by 1kW hardware experiments.