理化检验-物理分册2026，Vol.62Issue(8)：20-28,9.DOI:10.11973/lhjy-wl260033
基于机器学习优化的Larson-Miller参数法预测316H核电不锈钢持久强度
Prediction of the persistent strength of 316H nuclear power stainless steel using the machine learning-optimized Larson-Miller parameter method
张天宇 1尹嵬 2李忠基 1李建春 1张峰 1贾元伟1
作者信息
- 1. 山西太钢不锈钢股份有限公司,太原 030003
- 2. 太原钢铁(集团)有限公司 先进不锈钢全国重点实验室,太原 030003
- 折叠
摘要
Abstract
The Larson-Miller parameter method was commonly used to extrapolate the persistent strength of materials.The fracture times of 316H nuclear power stainless steel under various temperature-stress conditions were used as research data,calculations reveal that this method exhibited systematic deviations when handling multiple datasets under identical conditions.It was also found that uncleaned outliers in the raw data affect the reliability of the fitting model.To improve the accuracy of extrapolation results,the Larson-Miller parameter method was combined with various machine learning algorithms-including decision trees,AdaBoost,elastic net,random forests,K-nearest neighbors,and gradient boosting-to establish new models and conduct systematic analysis.The differences between the Larson-Miller parameter method and these new models were compared in terms of fitting processes,predictive performance,and model evaluation.The persistent strength of 316H nuclear power stainless steel at 550 ℃ and 600 ℃ after holding for 105 hours was also extrapolated.The results show that most of the new models exhibited excellent predictive performance,with the ensemble learning algorithms of random forests and gradient boosting demonstrating the best overall performance,significantly outperforming the traditional Larson-Miller parameter method across all performance metrics.关键词
316H核电不锈钢/持久强度/Larson-Miller参数法/机器学习/模型预测性能Key words
316H nuclear power stainless steel/persistent strength/Larson-Miller parameter method/machine learning/model predictive performance分类
通用工业技术引用本文复制引用
张天宇,尹嵬,李忠基,李建春,张峰,贾元伟..基于机器学习优化的Larson-Miller参数法预测316H核电不锈钢持久强度[J].理化检验-物理分册,2026,62(8):20-28,9.
