A Novel A2C-Driven Reinforcement Learning Framework for Magnetic Flux Path Estimation in Ferrite-Core Transformers
Year: 2026
Type: article
Abstract: This paper proposes a novel A2C-driven reinforcement learning framework for estimating magnetic flux paths in ferrite-core transformers. Accurate characterization of leakage and fringing flux is essential for high-frequency magnetic-component design; however, conventional magnetic-equivalent-circuit (MEC) approaches rely on simplified reluctance approximations that often fail to capture localized nonlinear behavior. Finite-element-method (FEM) analysis offers higher fidelity but at the expense of substantial computational overhead, particularly for frequency-dependent ferrite materials. To overcome these limitations, the proposed method formulates magnetic-field estimation as a sequential decision-making problem and embeds the Advantage Actor–Critic (A2C) algorithm into a magnetic-resistance-matrix (MRM) environment that emulates the spatial reluctance landscape of the core. Without requiring any pre-labeled datasets or FEM-trained surrogate models, the A2C agent autonomously learns energy-consistent flux trajectories by minimizing cumulative reluctance along the magnetic path. The framework is validated on an N87 ferrite EE-core transformer using FEM as the reference. The estimated flux-distribution map achieves an average local relative error of 5.98% and attains a structural similarity index (SSIM) of 0.82 with respect to the FEM result, successfully reproducing key electromagnetic features such as core-corner flux crowding and stray-flux dispersion. These results demonstrate that the proposed reinforcement-learning-based approach provides a computationally efficient and physically meaningful alternative to traditional magnetic-field analysis, offering a promising tool for next-generation high-frequency ferrite-component modeling and optimization.
Source: IEEE Access
Authors: Gangseok Lee, Daeung Jeong, Seonghyeon Go
Institutions: Bucheon University, Hanyang University
Language: English
Topic: Advanced DC-DC Converters
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,075
OpenAlex ID: W7155397285