Scholarlink

<p>Low-alloy steel, widely utilized in marine construction and engineering due to its exceptional mechanical properties, requires performance optimization to meet the demands of complex service environments. To address the critical challenges of high-dimensional feature spaces and limited sample sizes, we propose a novel two-step feature selection algorithm integrating Differential Evolution and Feature Importance Scores. Comparative analysis demonstrates the algorithm’s superior performance in reducing feature numbers while enhancing model accuracy. The developed predictive models for tensile strength and elongation achieve <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> values of 0.97 and 0.90, respectively, using optimized descriptors. SHAP analysis reveals critical influencing features, including average covalent radius, sum of fusion enthalpy, variance of bulk modulus, tempering temperature, and tempering time. Symbolic regression further elucidates quantitative relationships between these features and target properties. The proposed framework offers an interpretable and efficient approach for optimizing low-alloy steel performance, providing valuable insights for material design and processing parameter selection through its advanced feature selection and modeling methodologies.</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

Differential Evolution-Enhanced Descriptor Selection for Low-Alloy Steel Performance Prediction

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Siqian Liu,
  • Guipeng Liu,
  • Kehong Zhong,
  • Juan Qi,
  • Longwei Cheng,
  • Dongmei Ai

摘要

Low-alloy steel, widely utilized in marine construction and engineering due to its exceptional mechanical properties, requires performance optimization to meet the demands of complex service environments. To address the critical challenges of high-dimensional feature spaces and limited sample sizes, we propose a novel two-step feature selection algorithm integrating Differential Evolution and Feature Importance Scores. Comparative analysis demonstrates the algorithm’s superior performance in reducing feature numbers while enhancing model accuracy. The developed predictive models for tensile strength and elongation achieve \(R^{2}\) R 2 values of 0.97 and 0.90, respectively, using optimized descriptors. SHAP analysis reveals critical influencing features, including average covalent radius, sum of fusion enthalpy, variance of bulk modulus, tempering temperature, and tempering time. Symbolic regression further elucidates quantitative relationships between these features and target properties. The proposed framework offers an interpretable and efficient approach for optimizing low-alloy steel performance, providing valuable insights for material design and processing parameter selection through its advanced feature selection and modeling methodologies.