基于改进Stacking集成学习的高强度钢柱屈曲能力预测

由于屈曲强度的形成机制复杂,影响屈曲强度的因素较多,目前对屈曲强度的认识还不全面。近年来,机器学习已初步应用于预测结构屈曲强度等力学性能,然而基于实验测试的样本数据量小容易造成过拟合,导致其预测精度低。本文提出一种基于改进Stacking算法的GSSA(Grid Search-Stacking Algorithm)模型,并对某型号高强度钢柱屈曲强度进行预测,提升了屈曲强度的预测精度。首先,基于标准Stacking算法通过使用网格搜索算法选择最优基模型组合,并采用留一交叉验证(LOOCV)法训练基模型,实现了GSSA模型的构建,有效解决了小样本集训练带来的预测精度低问题;然后,为了进一步验证GSSA模型的可靠性,本文采用Bland-Altman法对GSSA模型进行一致性评价,结果表明,GSSA模型具有很好的可靠性;最后,采用SHAP模型对GSSA模型预测的屈曲强度进行了可解释性分析,实现了其影响因素评价。

Prediction of buckling capacity of high-strength steel columns based on improved stacking ensemble learning

The existing knowledge of buckling strength is insufficient due to the complicated formation mechanism of buckling strength,which is influenced by various factors.Machine learning has been initially applied to predict mechanical properties of structures such as buckling strength in recent years.However,the small samples based on experimental tests easily cause overfitting,resulting in low prediction accuracy.In this paper,a Grid Search-Stacking Algorithm (GSSA) model based on the improved Stacking Algorithm was proposed to predict the buckling strength of high-strength steel columns,and the prediction accuracy of buckling strength is improved.First,based on the standard Stacking algorithm,the GSSA model was constructed by employing the grid search algorithm to select the optimal combination of base models and the leave-one-out cross-validation (LOOCV) method to train the base models,which effectively solves the problem of low prediction accuracy caused by the training of small samples;then,in order to further verify the reliability of the GSSA model,this paper adopted the Bland-Altman method to evaluate the consistency of the GSSA model,and the results show that the GSSA model has high reliability.Finally,the SHAP model was introduced to perform an interpretability analysis of buckling strength predicted by the GSSA model and realizes the evaluation of its influence factors.