赵忠明, 王帅彬, 江东平,等.基于集成学习的多工况条件下花岗岩单轴抗压强度预测[J].河南理工大学学报（自然科学版）,2026,45(4):114-122.

ZHAO Z M, WANG S B, JIANG D P,et al.Prediction of granite uniaxial compressive strength under multiple working conditions based on ensemble learning[J].Journal of Henan Polytechnic University(Natural Science) ,2026,45(4):114-122.

基于集成学习的多工况条件下花岗岩单轴抗压强度预测

赵忠明1, 王帅彬1, 江东平2, 王春1,2, 付天予3, 程博1

1.河南理工大学 能源科学与工程学院，河南 焦作  454003;2.中钢集团马鞍山矿山研究总院有限公司，安徽 马鞍山 243000;3.中国矿业大学（北京） 应急管理与安全工程学院，北京  100083

摘要: 目的 针对传统岩石抗压强度参数测试方法耗时长、成本高等问题，提出一种基于集成学习的花岗岩单轴抗压强度预测方法。  方法 利用电子万能试验机对不同工况条件下花岗岩试样进行轴向压缩试验，结合试验数据和前人研究，构建多工况条件下花岗岩单轴抗压强度数据集，选用6种不同的算法模型（随机森林、梯度提升树、自适应提升、K近邻、神经网络、支持向量回归），基于集成学习框架建立多工况条件下花岗岩单轴抗压强度预测模型，采用多种评价指标对模型的性能进行综合对比分析，最后对模型的输入特征进行特征重要性分析。  结果 结果表明，花岗岩的单轴抗压强度受多种工况条件影响变化显著，构建的集成学习模型均表现出良好的预测效果，集成学习模型的综合性能由大到小排序为：随机森林算法模型、梯度提升树算法模型、自适应提升算法模型。最后通过特征重要性分析，各特征相对重要性由大到小排序为：加热温度、不同地质条件、加热-浸水循环次数、浸水温度、浸水时间、冷却方式、传热溶液pH。  结论 集成学习算法模型中RF模型的综合性能优于其他2种模型，特征重要性分析得出高温对花岗岩的单轴抗压强度影响最显著，该模型为多工况条件下花岗岩的单轴抗压强度预测提供了一种新思路。

关键词:花岗岩;集成学习;随机森林;抗压强度预测;特征重要性分析

doi:10.16186/j.cnki.1673-9787.2024120030

基金项目:国家自然科学基金资助项目（52274076）；河南省优秀青年科学基金资助项目（242300421070）；“双一流”创建学科项目（GCCRC202403）；中国博士后科学基金第73批面上项目（2023M733279）

收稿日期:2024/12/13

修回日期:2025/04/21

出版日期:2026/06/17

Prediction of granite uniaxial compressive strength under multiple working conditions based on ensemble learning

Zhao Zhongming1, Wang Shuaibin1, Jiang Dongping2, Wang Chun1,2, Fu Tianyu3, Cheng Bo1

1.School of Energy Science and Engineering， Henan Polytechnic University， Jiaozuo  454003， Henan， China;2.Sinosteel Maanshan General Institute of Mining Research Co.， Ltd.， Maanshan  243000， Anhui， China;3.School of Emergency Management and Safety Engineering， China University of Mining & Technology， Beijing  100083， China

Abstract: Objectives To address the problems of long testing periods and high costs associated with traditional methods for measuring rock compressive strength， a prediction method for granite uniaxial compressive strength based on ensemble learning was proposed. Methods Axial compression tests were conducted on granite specimens under different working conditions using an electronic universal testing machine. Based on the experimental results and published data， a dataset for granite the uniaxial compressive strength under multiple working conditions was established. Six different machine learning algorithms were selected to develop prediction models within an ensemble learning framework. The performance of the models was comprehensively evaluated using multiple evaluation metrics， and feature importance analysis was subsequently performed on the input variables.  Results The results indicate that the uniaxial compressive strength of granite is significantly affected by different working conditions. All established ensemble learning models exhibited good predictive performance. The overall model performance was ranked as follows： Random Forest， Gradient Boosting， Adaptive Boosting. Feature importance analysis further revealed that the relative importance of the influencing factors followed the order： heating temperature， geological conditions， heating-immersion cycle， immersion temperature， immersion time， cooling method， heat-transfer solution.  Conclusions Among the ensemble learning models， the RF model achieved the best overall predictive performance. The analysis also showed that high temperature has the most significant influence on granite uniaxial compressive strength. The proposed model provides a new approach for predicting the uniaxial compressive strength of granite under multiple working conditions.

Key words:granite;ensemble learning;random forest;uniaxial compressive strength prediction;feature importance analysis