面向小样本的铁尾矿砂钢管混凝土柱承载力预测OA
Prediction on bearing capacity of concrete-filled steel with iron tailing sand replacement under small samples condition
为提升小样本条件下铁尾矿砂钢管混凝土柱承载力的预测精度,提出一种基于双学习器与Boosting机制的迁移学习算法.以决策树(DT)、随机森林(RF)、极端随机树(ET)、AdaBoost和CatBoost五种传统机器学习模型为基学习器,分别构建迁移学习模型,用于预测铁尾矿砂钢管混凝土柱的极限承载力,并对最优模型进行SHAP可解释性分析.研究结果表明:该迁移学习算法可提升各基学习器在小样本条件下的预测精度,其中,Tr-AdaBoost模型综合性能最佳,其决定系数R2、均方根误差(RMSE)、平均绝对误差(MAE)和平均绝对百分比误差(MAPE)分别为 0.979、113.890 kN、65.896 kN 和 4.5%.与原始 AdaBoost 模型相比,其 R2 提升约4.93%,RMSE降低约44.67%.可解释性分析显示,铁尾矿砂取代率对特征重要性排名有显著影响,并基于Tr-AdaBoost模型开发图形用户界面预测工具.研究结果为铁尾矿砂的工程化利用与快速设计评估提供支撑.
To improve the prediction accuracy of bearing capacity of concrete-filled steel tubular columns containing iron tailings sand under small sample conditions,a transfer learning algorithm based on double learners and Boosting mechanism is proposed.Five traditional machine learning models,decision tree(DT),random forest(RF),extreme randomized tree(ET),AdaBoost and CatBoost,were used as base learners to construct transfer learning models for predicting the ultimate bearing capacity of concrete-filled steel tubular columns with iron tailings sand,and SHAP interpretability analysis was performed on the optimal model.The results show that the transfer learning algorithm improves the prediction accuracy of each base learner under small sample conditions.The Tr-AdaBoost model has the best comprehensive performance,and its determination coefficient(R2),root mean square error(RMSE),mean absolute error(MAE)and mean absolute percentage error(MAPE)are 0.979,113.890 kN,65.896 kN and 4.5%,respectively.Compared with the original AdaBoost model,its R2 increased by approximately 4.93%and RMSE decreased by approximately 44.67%.The interpretability analysis shows that the replacement rate of iron tailings has a significant impact on the ranking of feature importance.Based on the Tr-AdaBoost model,a graphical user interface prediction tool was developed.The research results provide data support for the engineering utilization and rapid design evaluation of iron tailings.
李九阳;罗程升;王炳鑫;朱岳鹏;陈立;邵明亮
长春工程学院 土木工程学院,吉林 长春 130012长春工程学院 土木工程学院,吉林 长春 130012长春工程学院 土木工程学院,吉林 长春 130012长春工程学院 土木工程学院,吉林 长春 130012同济大学 土木工程学院,上海 200092长春国信奥辉装配式建筑发展有限公司,吉林 长春 130600
建筑与水利
迁移学习机器学习小样本铁尾矿砂可解释性分析
transfer learningmachine learningsmall sampleiron tailings sandinterpretability analysis
《辽宁工程技术大学学报(自然科学版)》 2026 (5)
563-573,11
国家自然科学基金项目(51778063)吉林省科技厅科技发展计划项目(YDZJ202302CXJD052)吉林省自然科学基金项目(YDZJ202401370ZYTS)
评论