Scholarlink

<p>Precise prediction and mechanistic elucidation of level fluctuation in continuous casting mold, coupled with effective feature extraction from limited industrial data samples, are essential for optimizing process control, enhancing production stability in modern steel manufacturing. To tackle the aforementioned issues, a novel feedforward neural network model that integrates Bayesian optimization and nested cross-validation (BO-NCV-FNN) is proposed in this work. The overfitting problem caused by limited data are addressed by the proposed model, with adaptive hyperparameter optimization employed to enhance robustness. Compared to established machine learning (ML) benchmarks—Linear Regression, Support Vector Machine (SVM), Random Forest, and k-Nearest Neighbor (k-NN)—superior performance is achieved by the BO-NCV-FNN, with a coefficient of determination (R<sup>2</sup>) of 0.93, mean square error (MSE) of 0.01, reflecting exceptional predictive accuracy and minimal systematic deviation. Generalization capability is validated through random subset evaluations, with R<sup>2</sup> values maintained above 0.86. An interpretable analysis of the interactions between process parameters controlling level fluctuation was conducted based on the BO-NCV-FNN. Overfitting risks are minimized in this work while data features are accurately extracted, providing industrially applicable prediction accuracy.</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

A Novel Feedforward Neural Network Model for Predicting the Level Fluctuation in Continuous Casting Mold

  • 期刊论文

  • 发表日期：

    2025年7月3日

查看全文
  • Yue Sun,
  • Zhongqiu Liu,
  • Yongtao Xiong,
  • Jun Yang,
  • Guodong Xu,
  • Baokuan Li

摘要

Precise prediction and mechanistic elucidation of level fluctuation in continuous casting mold, coupled with effective feature extraction from limited industrial data samples, are essential for optimizing process control, enhancing production stability in modern steel manufacturing. To tackle the aforementioned issues, a novel feedforward neural network model that integrates Bayesian optimization and nested cross-validation (BO-NCV-FNN) is proposed in this work. The overfitting problem caused by limited data are addressed by the proposed model, with adaptive hyperparameter optimization employed to enhance robustness. Compared to established machine learning (ML) benchmarks—Linear Regression, Support Vector Machine (SVM), Random Forest, and k-Nearest Neighbor (k-NN)—superior performance is achieved by the BO-NCV-FNN, with a coefficient of determination (R2) of 0.93, mean square error (MSE) of 0.01, reflecting exceptional predictive accuracy and minimal systematic deviation. Generalization capability is validated through random subset evaluations, with R2 values maintained above 0.86. An interpretable analysis of the interactions between process parameters controlling level fluctuation was conducted based on the BO-NCV-FNN. Overfitting risks are minimized in this work while data features are accurately extracted, providing industrially applicable prediction accuracy.