Scholarlink

<p>As an economical and practical material, carbon steel is widely used in various equipment in atmospheric environment, especially in power energy facilities. The corrosion process in the atmosphere is slow and complex, and is affected by many factors, which makes it difficult to explore the underlying relationships of atmospheric corrosion. The parameters are highly nonlinear and unbalanced, which makes the traditional machine learning algorithm perform poorly in prediction accuracy. In this study, machine learning is combined with data balance technology and optimization algorithm, and synthetic minority oversampling technology is adopted to solve the problem of data imbalance, which generates synthetic data to reduce overfitting. Then the processed data are brought into the machine learning model for training verification, and different models are compared. The empirical results show that the prediction model proposed in this paper is superior to single model and other integrated models in prediction accuracy and generalization ability.</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

Prediction of Industrial Atmospheric Corrosion and Evaluation of Influencing Factors Based on Machine Learning

  • 期刊论文

  • 发表日期：

    2025年4月28日

查看全文
  • Hongxia Wan,
  • Sijia Liu,
  • Dongdong Song,
  • Wenli Cai,
  • Yong Wang,
  • Lin Geng,
  • Changfeng Chen

摘要

As an economical and practical material, carbon steel is widely used in various equipment in atmospheric environment, especially in power energy facilities. The corrosion process in the atmosphere is slow and complex, and is affected by many factors, which makes it difficult to explore the underlying relationships of atmospheric corrosion. The parameters are highly nonlinear and unbalanced, which makes the traditional machine learning algorithm perform poorly in prediction accuracy. In this study, machine learning is combined with data balance technology and optimization algorithm, and synthetic minority oversampling technology is adopted to solve the problem of data imbalance, which generates synthetic data to reduce overfitting. Then the processed data are brought into the machine learning model for training verification, and different models are compared. The empirical results show that the prediction model proposed in this paper is superior to single model and other integrated models in prediction accuracy and generalization ability.