基于AutoML-SHAP的超高性能混凝土抗压强度可解释预测

硅酸盐通报 ›› 2024, Vol. 43 ›› Issue (10): 3634-3644.

基于AutoML-SHAP的超高性能混凝土抗压强度可解释预测

李硕¹, 艾丽菲拉·艾尔肯², 罗文波^2,3, 陈锦杰¹

1.湘潭大学土木工程学院,湘潭 411105;
2.长沙学院土木工程学院,长沙 410022;
3.湘潭大学岩土力学与工程安全湖南省重点实验室,湘潭 411105

收稿日期:2024-03-18 修订日期:2024-06-17 出版日期:2024-10-15 发布日期:2024-10-16
通信作者: 罗文波,博士,教授。E-mail:luowenbo@ccsu.edu.cn
作者简介:李硕(2000—),男,硕士研究生。主要从事土木工程材料性能与人工智能算法的研究。E-mail:1439179459@qq.com
基金资助:
国家自然科学基金(12072308)

Interpretable Prediction of Compressive Strength of Ultra-High Performance Concrete Based on AutoML-SHAP

LI Shuo¹, AILIFEILA Aierken², LUO Wenbo^2,3, CHEN Jinjie¹

1. School of Civil Engineering, Xiangtan University, Xiangtan 411105, China;
2. School of Civil Engineering, Changsha University, Changsha 410022, China;
3. Hunan Key Laboratory of Geomechanics and Engineering Safety, Xiangtan University, Xiangtan 411105, China

Received:2024-03-18 Revised:2024-06-17 Published:2024-10-15 Online:2024-10-16

摘要/Abstract

摘要： 超高性能混凝土(UHPC)的抗压强度与其配比成分之间存在高度非线性的复杂关系,利用传统的统计方法难以准确预测抗压强度。为解决这一问题,本文提出一种基于自动机器学习(AutoML)技术的UHPC抗压强度预测办法,同时引入沙普利加和解释(SHAP)增加其可解释性。AutoML和SHAP的集成有助于构建精确、高效且可解释的模型。结果表明,AutoML模型可自动建立,其准确性、稳健性优于基础模型。SHAP通过全局解释性分析、单样本解释分析以及特征依赖性解释分析,阐明了各个特征因素对抗压强度的影响机理,有助于UHPC抗压强度发展机制以及影响参数重要性的理解,可为UHPC的设计与应用提供参考。

关键词: 超高性能混凝土, 抗压强度, 机器学习, AutoML, SHAP

Abstract: The correlations between compressive strength of UHPC and its mixture composition exhibit pronounced nonlinearity, presenting a challenge for analysis through conventional statistical approaches. In this study, an automatic machine learning (AutoML) technology was proposed to predict compressive strength of UHPC, and shapley additiveex planations (SHAP) was introduced to explain the AutoML model. The integration of AutoML and SHAP offered synergistic benefits, facilitating the development of a precise, efficient, and comprehensively interpretable model. Results demonstrate that AutoML model is automatically built with better accuracy and robustness than the base model. SHAP provides a global explanation, a single sample explanation, and a feature dependence explanation of characterization factors, which explains mechanism of the effect of each characterization factor on compressive strength. SHAP contributes to the understanding of mechanism of UHPC compressive strength development and the importance of characteristic factors, and can provide assistance in the design and application of UHPC.

Key words: ultra-high performance concrete, compressive strength, machine learning, AutoML, SHAP

中图分类号:

TP181

李硕, 艾丽菲拉·艾尔肯, 罗文波, 陈锦杰. 基于AutoML-SHAP的超高性能混凝土抗压强度可解释预测[J]. 硅酸盐通报, 2024, 43(10): 3634-3644.

LI Shuo, AILIFEILA Aierken, LUO Wenbo, CHEN Jinjie. Interpretable Prediction of Compressive Strength of Ultra-High Performance Concrete Based on AutoML-SHAP[J]. BULLETIN OF THE CHINESE CERAMIC SOCIETY, 2024, 43(10): 3634-3644.

参考文献

[1] ABELLÁN-GARCÍA J. Four-layer perceptron approach for strength prediction of UHPC[J]. Construction and Building Materials, 2020, 256: 119465.
[2] KUMAR M, BISWAS R, KUMAR D R, et al. Soft computing-based prediction models for compressive strength of concrete[J]. Case Studies in Construction Materials, 2023, 19: e02321.
[3] KUMAR T R, PANCHAL V R, PANDYA K S. An ensemble approach to improve BPNN model precision for predicting compressive strength of high-performance concrete[J]. Structures, 2022, 45: 500-508.
[4] ALYASEEN A, PODDAR A, KUMAR N, et al. Assessing the compressive and splitting tensile strength of self-compacting recycled coarse aggregate concrete using machine learning and statistical techniques[J]. Materials Today Communications, 2024, 38: 107970.
[5] KHAN S A, KHAN M A, AMIN M N, et al. Sustainable alternate binding material for concrete using waste materials: a testing and computational study for the strength evaluation[J]. Journal of Building Engineering, 2023, 80: 107932.
[6] LI Q F, SONG Z M. Prediction of compressive strength of rice husk ash concrete based on stacking ensemble learning model[J]. Journal of Cleaner Production, 2023, 382: 135279.
[7] ZHAO Y P, GOULIAS D, SAREMI S. Enhancing prediction accuracy of concrete compressive strength using stacking ensemble machine learning[J]. Comput Concrete, 2023, 32(3): 233-46.
[8] ZHENG J Y, WANG M H, YAO T C, et al. Dynamic mechanical strength prediction of BFRC based on stacking ensemble learning and genetic algorithm optimization[J]. Buildings, 2023, 13(5): 1155.
[9] CHEN H G, YANG J J, CHEN X H. A convolution-based deep learning approach for estimating compressive strength of fiber reinforced concrete at elevated temperatures[J]. Construction and Building Materials, 2021, 313: 125437.
[10] ZHANG C B, TIAN X N, ZHAO Y, et al. Automated machine learning-based building energy load prediction method[J]. Journal of Building Engineering, 2023, 80: 108071.
[11] CHEN W Y, ZHANG L M. An automated machine learning approach for earthquake casualty rate and economic loss prediction[J]. Reliability Engineering & System Safety, 2022, 225: 108645.
[12] HARIRI A M A, POURKAMALI A F. An automated machine learning engine with inverse analysis for seismic design of dams[J]. Water, 2022, 14(23): 3898.
[13] LUNDBERG S, LEE S I. A unified approach to interpreting model predictions[C]//Conference and Workshop on Neural Information Processing Systems. California, USA, 2017.
[14] ZHAO W J, FENG S Y, LIU J X, et al. An explainable intelligent algorithm for the multiple performance prediction of cement-based grouting materials[J]. Construction and Building Materials, 2023, 366: 130146.
[15] ERICKSON N, MUELLER J, SHIRKOV A, et al. AutoGluon-tabular: robust and accurate AutoML for structured data[J]. ArXiv e-Prints, 2020: 06505.
[16] PHOEUK M, KWON M. Accuracy prediction of compressive strength of concrete incorporating recycled aggregate using ensemble learning algorithms: multinational dataset[J]. Advances in Civil Engineering, 2023, 2023: 5076429.
[17] HUU NGUYEN M, NGUYEN T A, LY H B. Ensemble XGBoost schemes for improved compressive strength prediction of UHPC[J]. Structures, 2023, 57: 105062.
[18] ABELLÁN G J. Study of nonlinear relationships between dosage mixture design and the compressive strength of UHPC[J]. Case Studies in Construction Materials, 2022, 17: e01228.
[19] KIM K, KIM W, SEO J, et al. Prediction of concrete fragments amount and travel distance under impact loading using deep neural network and gradient boosting method[J]. Materials, 2022, 15(3): 1045.
[20] WU X P, ZHU F, ZHOU M M, et al. Intelligent design of construction materials: a comparative study of AI approaches for predicting the strength of concrete with blast furnace slag[J]. Materials, 2022, 15(13): 4582.
[21] MARANI A, JAMALI A, NEHDI M L. Predicting ultra-high-performance concrete compressive strength using tabular generative adversarial networks[J]. Materials, 2020, 13(21): 4757.
[22] ALIMI O A, OUAHADA K, ABU-MAHFOUZ A M. A review of machine learning approaches to power system security and stability[J]. IEEE Access, 2020, 8: 113512-113531.
[23] KASHEM A, KARIM R, MALO S C, et al. Hybrid data-driven approaches to predicting the compressive strength of ultra-high-performance concrete using SHAP and PDP analyses[J]. Case Studies in Construction Materials, 2024, 20: e02991.
[24] ALABDULJABBAR H, KHAN M, AWAN H H, et al. Predicting ultra-high-performance concrete compressive strength using gene expression programming method[J]. Case Studies in Construction Materials, 2023, 18: e02074.
[25] QIAN Y F, SUFIAN M, ACCOUCHE O, et al. Advanced machine learning algorithms to evaluate the effects of the raw ingredients on flowability and compressive strength of ultra-high-performance concrete[J]. PLoS One, 2022, 17(12): e0278161.
[26] WONG H S, BUENFELD N R. Determining the water-cement ratio, cement content, water content and degree of hydration of hardened cement paste: method development and validation on paste samples[J]. Cement and Concrete Research, 2009, 39(10): 957-965.
[27] ABBASS W, KHAN M I, MOURAD S. Evaluation of mechanical properties of steel fiber reinforced concrete with different strengths of concrete[J]. Construction and Building Materials, 2018, 168: 556-569.
[28] JI Y J, CAHYADI J H. Effects of densified silica fume on microstructure and compressive strength of blended cement pastes[J]. Cement and Concrete Research, 2003, 33(10): 1543-1548.
[29] YIP C K, LUKEY G C, VAN-DEVENTER J S J. The coexistence of geopolymeric gel and calcium silicate hydrate at the early stage of alkaline activation[J]. Cement and Concrete Research, 2005, 35(9): 1688-1697.
[30] NOCHAIYA T, WONGKEO W, PIMRAKSA K, et al. Microstructural, physical, and thermal analyses of Portland cement-fly ash-calcium hydroxide blended pastes[J]. Journal of Thermal Analysis and Calorimetry, 2010, 100(1): 101-108.