东北大学学报(自然科学版) ›› 2024, Vol. 45 ›› Issue (5): 738-744.DOI: 10.12068/j.issn.1005-3026.2024.05.017

• 资源与土木工程 • 上一篇    

混凝土抗压强度的可解释深度学习预测模型

章伟琪, 王辉明   

  1. 新疆大学 建筑工程学院,新疆 乌鲁木齐 830017
  • 收稿日期:2022-12-26 出版日期:2024-05-15 发布日期:2024-07-31
  • 作者简介:章伟琪(1995-),女,安徽池州人,新疆大学硕士研究生
    王辉明(1967-),男,江苏镇江人,新疆大学教授,博士生导师.
  • 基金资助:
    新疆建筑结构与抗震重点实验室开放课题(600120004)

Interpretable Deep Learning Prediction Model for Compressive Strength of Concrete

Wei-qi ZHANG, Hui-ming WANG   

  1. College of Civil Engineering and Architecture,Xinjiang University,Urumqi 830017,China. Corresponding author: WANG Hui-ming,E-mail: wanghmxj@126. com
  • Received:2022-12-26 Online:2024-05-15 Published:2024-07-31

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.

关键词: 混凝土, 抗压强度, 深度学习, SHAP方法, 可解释性

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.

Key words: concrete, compressive strength, deep learning, SHAP method, interpretation

中图分类号: