Scholarlink

<p>To enhance the control accuracy of wireless power transfer (WPT) systems in scenarios without bilateral communication, this study optimizes the parameter prediction method for WPT systems. A transfer learning optimization is applied to the backpropagation neural network to compare traditional parameter prediction approaches. This study focuses on an S-S-compensated wireless charging system. The influence of secondary-side parameters on the primary-side current is analyzed using input–output characteristics, and a system-level black-box model is established. Subsequently, a neural network model is employed to perform nonlinear fitting of the system characteristics. Through transfer learning, the neural network model trained on simulation data is adapted into a parameter-prediction model based on real-world data. Finally, the trained algorithm is implemented on an ARM chip, and a 400&#xa0;W wireless charging experimental platform is constructed to validate its effectiveness and accuracy. Offline and online experimental results indicate that the model trained with traditional methods achieves a prediction error rate of 13.38%. By contrast, after optimization with a transfer learning algorithm trained on real-world datasets, the error rate is reduced to 2.43%.</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

Transfer learning-based parameter identification method for wireless power transfer systems

  • 期刊论文

  • 发表日期：

    2026年3月13日

查看全文
  • Zhenjie Li,
  • Yusong Zhang,
  • Yu Liu,
  • Yuhong Bai,
  • Mingfei Ban

摘要

To enhance the control accuracy of wireless power transfer (WPT) systems in scenarios without bilateral communication, this study optimizes the parameter prediction method for WPT systems. A transfer learning optimization is applied to the backpropagation neural network to compare traditional parameter prediction approaches. This study focuses on an S-S-compensated wireless charging system. The influence of secondary-side parameters on the primary-side current is analyzed using input–output characteristics, and a system-level black-box model is established. Subsequently, a neural network model is employed to perform nonlinear fitting of the system characteristics. Through transfer learning, the neural network model trained on simulation data is adapted into a parameter-prediction model based on real-world data. Finally, the trained algorithm is implemented on an ARM chip, and a 400 W wireless charging experimental platform is constructed to validate its effectiveness and accuracy. Offline and online experimental results indicate that the model trained with traditional methods achieves a prediction error rate of 13.38%. By contrast, after optimization with a transfer learning algorithm trained on real-world datasets, the error rate is reduced to 2.43%.