Scholarlink

<p>The modular axial flux PMSG (MAFG) is a promising future wind generator with advantages like convenient manufacturing, installation and maintenance. However, modularization causes new issues—reduced flux density and increased harmonics—from inter-module magnetic contact reluctance. To address these problems, this paper proposes an optimization design method of the MAFG for wind turbine. The proposed method integrates a deep neural network (DNN), Bayesian hyperparameter tuning and non-dominated sorting genetic algorithm II (NSGA-II) multi-objective algorithm. This hybrid approach addresses key challenges: inter-module magnetic field distortion, harmonic enhancement, and nonlinear parameter coupling. Three structural parameters are optimized using Latin Hypercube Sampling (LHS) combined with Bayesian hyperparameter tuning to train the DNN surrogate model. The Bayesian-optimized DNN surrogate model accelerates the NSGA-II search for optimal structural parameters. The average torque <i>T</i><sub><i>a</i></sub> and torque ripple <i>T</i><sub><i>ri</i></sub> are set as optimization objectives, reflecting the demand for high power output and reliable operation. Simulation results validate the effectiveness of the proposed method: the optimized MAFG achieves an increase by 10.4% in <i>T</i><sub><i>a</i></sub> and a reduction by 25.8% to 4.6% in torque ripple <i>T</i><sub><i>ri</i></sub>, which satisfies the low-ripple requirement. The proposed methodology balances performance, efficiency and cost, providing a systematic guidance for optimization high-power-density renewable energy generators.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimization Design of Modular Axial Flux PMSG for Wind Turbine Based on DNN-Bayesian-NSGA-II Algorithm

  • 期刊论文

  • 发表日期：

    2025年12月12日

查看全文
  • Zhizheng Zhang,
  • Ying Liu,
  • Yixin Zhang,
  • Xiangzheng Kong,
  • Qinghai Qin,
  • Bin Cai

摘要

The modular axial flux PMSG (MAFG) is a promising future wind generator with advantages like convenient manufacturing, installation and maintenance. However, modularization causes new issues—reduced flux density and increased harmonics—from inter-module magnetic contact reluctance. To address these problems, this paper proposes an optimization design method of the MAFG for wind turbine. The proposed method integrates a deep neural network (DNN), Bayesian hyperparameter tuning and non-dominated sorting genetic algorithm II (NSGA-II) multi-objective algorithm. This hybrid approach addresses key challenges: inter-module magnetic field distortion, harmonic enhancement, and nonlinear parameter coupling. Three structural parameters are optimized using Latin Hypercube Sampling (LHS) combined with Bayesian hyperparameter tuning to train the DNN surrogate model. The Bayesian-optimized DNN surrogate model accelerates the NSGA-II search for optimal structural parameters. The average torque Ta and torque ripple Tri are set as optimization objectives, reflecting the demand for high power output and reliable operation. Simulation results validate the effectiveness of the proposed method: the optimized MAFG achieves an increase by 10.4% in Ta and a reduction by 25.8% to 4.6% in torque ripple Tri, which satisfies the low-ripple requirement. The proposed methodology balances performance, efficiency and cost, providing a systematic guidance for optimization high-power-density renewable energy generators.