中南大学学报（自然科学版）2025，Vol.56Issue(5)：1981-1992,12.DOI:10.11817/j.issn.1672-7207.2025.05.024
基于集成学习的混凝土抗压强度预测模型研究
Research on prediction model of compressive strength of concrete based on ensemble learning
摘要
Abstract
To accurately predict the compressive strength of concrete,the grey wolf optimization(GWO)algorithm was employed to optimize the hyperparameters of the light gradient boosting machine(LGBM).Firstly,the ratio of water to binder,blast furnace slag percentage in relation to the total binder,fly ash percentage in relation to the total binder,superplasticizer percentage in relation to the total binder,sand ratio,and age were used as inputs,the compressive strength was used as the output,and a GWO-LGBM prediction model was established.Secondly,the model's performance was evaluated on the training and testing sets to verify the effectiveness of GWO in optimizing LGBM hyperparameters.Thirdly,the model was applied to new data to test its robustness.Finally,the impact of input parameters on compressive strength was analyzed by using the GWO-LGBM model to further validate its predictive reliability.The results show that the GWO-LGBM model achieves root mean square errors(ERMSE)of the compressive strength of concrete,1.68 MPa and 3.49 MPa on the training and testing sets,respectively,with R2 of 0.99 and 0.95.This optimization effectively mitigates the issue of LGBM getting trapped in local optima.When the model is applied to a new dataset,83％of the predictions have a relative error below 10％,demonstrating a strong generalization ability.The increase of the ratio of water to binder significantly reduces compressive strength.The addition of blast furnace slag and fly ash decreases early strength but has minimal impact on later strength.When the ratio of water to binder ratio is fixed,there exists an optimal sand ratio that maximizes the strength.The patterns captured by the model is consistent with the influencing compressive strength theory,validating the rationality of its predictive results.关键词
混凝土/抗压强度/集成学习/轻量级梯度提升机/灰狼优化算法Key words
concrete/compressive strength/ensemble learning/light gradient boosting machine/grey wolf optimization algorithm分类
计算机与自动化引用本文复制引用
周继发,郭宏,潘自立,曾晓辉,郑振华,涂金根,郭桃明,孙晗凌,谢友均,龙广成,唐卓..基于集成学习的混凝土抗压强度预测模型研究[J].中南大学学报（自然科学版）,2025,56(5):1981-1992,12.基金项目
国家重点研发计划项目(2023YFB2604304-2,2021YFB37037)(Projects(2023YFB2604304-2,2021YFB37037)supported by the National Key Research and Development Program of China) （2023YFB2604304-2,2021YFB37037）
