Scholarlink

<p>The learning performance of physics-informed neural networks (PINNs) was compared based on the data, and the effects of learning time reduction through transfer learning were analyzed. Various datasets were used to evaluate the performance of PINNs and how the quality of data affects the learning efficiency of the model. A method was proposed to reduce the learning time required to analyze transient phenomena, including eddy currents, by applying transfer learning techniques and using pre-trained models. The experimental results revealed an 80.2% maximum reduction in learning time by transfer learning, which is expected to help expand the applicability of PINNs along with the performance variations based on data. The foundational material for optimizing the neural network models for solving physics-based problems in the future is provided.</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

Enhancing Learning Efficiency in Physics Informed Neural Network through Data Comparison and Transfer Learning

  • 期刊论文

  • 发表日期：

    2025年4月15日

查看全文
  • Ji-Hoon Han,
  • Jong-Hoon Park,
  • Seung-Min Song,
  • Sun-Ki Hong

摘要

The learning performance of physics-informed neural networks (PINNs) was compared based on the data, and the effects of learning time reduction through transfer learning were analyzed. Various datasets were used to evaluate the performance of PINNs and how the quality of data affects the learning efficiency of the model. A method was proposed to reduce the learning time required to analyze transient phenomena, including eddy currents, by applying transfer learning techniques and using pre-trained models. The experimental results revealed an 80.2% maximum reduction in learning time by transfer learning, which is expected to help expand the applicability of PINNs along with the performance variations based on data. The foundational material for optimizing the neural network models for solving physics-based problems in the future is provided.