混凝土抗压强度的可解释深度学习预测模型
Interpretable Deep Learning Prediction Model for Compressive Strength of Concrete
章伟琪 1王辉明1
作者信息
- 1. 新疆大学 建筑工程学院,新疆 乌鲁木齐 830017
- 折叠
摘要
为快速、准确地预测混凝土抗压强度,采用深度学习技术建立预测模型,使用贝叶斯优化算法进行模型自动优化调节,并结合SHapley Additive exPlanations(SHAP)可解释性方法对预测结果进行分析,以克服预测模型的"黑盒子"问题.利用深度学习模型挖掘各输入特征参数与抗压强度之间潜在的规律;通过可视化输入特征参数的SHAP值分析参数对抗压强度预测结果的重要性及影响规律.结果表明,所建深度学习模型相比其他传统模型具有更好的性能;SHAP分析结果与试验结果一致,该模型较好地反映了各特征参数之间复杂的非线性关系,可为混凝土材料的工程设计提供依据和参考.
Abstract
To quickly and accurately predict the compressive strength of concrete,a prediction model is established using deep learning technology.The model is automatically optimized and adjusted using the Bayesian optimization algorithm,and the prediction results are analyzed by combining with the SHapley Additive exPlanations(SHAP)interpretable method,which overcomes the problem of the"black box"of the prediction model.The deep learning model is used to mine the potential law between each input feature parameter and compressive strength,the importance of the parameters on the compressive strength prediction results and the influence law is analyzed by visualizing the SHAP values of the input feature parameters.The results show that the constructed deep learning model outforms other traditional models.The SHAP analysis results are consistent with the experimental results,and the model better reflects the complex nonlinear relationship among the characteristic parameters,which can provide the basis and reference for the engineering design of concrete materials.
关键词
混凝土/抗压强度/深度学习/SHAP方法/可解释性Key words
concrete/compressive strength/deep learning/SHAP method/interpretation引用本文复制引用
基金项目
新疆建筑结构与抗震重点实验室开放课题(600120004)
出版年
2024
NETL
NSTL
万方数据