Scholarlink

<p>Geopolymers prepared from industrial by-products (IBPs) significantly enhance the compressive strength (CS) of cohesive soils. However, existing machine learning (ML) models for predicting the CS of IBP--geopolymer stabilized cohesive soils (IBP-GCS) are limited by their consideration of few influencing factors and narrow applicability. To address these limitations, a hybrid ML model was constructed that leverages the contents of key chemical components to predict the CS of IBP-GCS. Firstly, a database of 787 samples was compiled from the literature. Secondly, eight ML models were trained and tested, and their generalization performance was evaluated using six performance metrics. Finally, Shapley additive explanation method was employed to assess the importance of the feature variable. The results indicate that the extreme gradient boosting model tuned with the zebra optimization algorithm (ZOA-XGB) achieved the best performance, with a coefficient of determination of 0.91 on the independent test set. The contents of calcium oxide, silicon dioxide, and the curing age were identified as the key variables affecting CS. Further optimization strategies were proposed to improve the effectiveness of IBP-GCS. When the total water content is less than 50%, specific recommendations are made: the silicon dioxide content should be below 1.87%, the aluminium oxide content below 2.47%, and the calcium oxide content above 5.50%. Thus, the established ZOA-XGB model provides a reliable tool for predicting the CS of IBP-GCS based on the contents of key chemical components, offering scientific and practical guidance for the design and construction of IBP-GCS in soft foundation engineering.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Indirect estimation of compressive strength of industrial byproduct-geopolymer stabilized cohesive soils: a novel hybrid extreme gradient boosting model

  • 期刊论文

  • 发表日期：

    2025年10月6日

查看全文
  • Qianglong Yao,
  • Yiliang Tu,
  • Jiahui Yang

摘要

Geopolymers prepared from industrial by-products (IBPs) significantly enhance the compressive strength (CS) of cohesive soils. However, existing machine learning (ML) models for predicting the CS of IBP--geopolymer stabilized cohesive soils (IBP-GCS) are limited by their consideration of few influencing factors and narrow applicability. To address these limitations, a hybrid ML model was constructed that leverages the contents of key chemical components to predict the CS of IBP-GCS. Firstly, a database of 787 samples was compiled from the literature. Secondly, eight ML models were trained and tested, and their generalization performance was evaluated using six performance metrics. Finally, Shapley additive explanation method was employed to assess the importance of the feature variable. The results indicate that the extreme gradient boosting model tuned with the zebra optimization algorithm (ZOA-XGB) achieved the best performance, with a coefficient of determination of 0.91 on the independent test set. The contents of calcium oxide, silicon dioxide, and the curing age were identified as the key variables affecting CS. Further optimization strategies were proposed to improve the effectiveness of IBP-GCS. When the total water content is less than 50%, specific recommendations are made: the silicon dioxide content should be below 1.87%, the aluminium oxide content below 2.47%, and the calcium oxide content above 5.50%. Thus, the established ZOA-XGB model provides a reliable tool for predicting the CS of IBP-GCS based on the contents of key chemical components, offering scientific and practical guidance for the design and construction of IBP-GCS in soft foundation engineering.