基于回弹法预测岩石单轴抗压强度的MLP-ANN模型  

MLP-ANN model for predicting uniaxial compressive strength of rocks based on the rebound method

作　　者：李明 窦斌[1] 朴昇昊 马云龙 王帅 孙左帅 王祥 LI Ming;DOU Bin;PIAO Shenghao;MA Yunlong;WANG Shuai;SUN Zuoshuai;WANG Xiang(Faculty of Engineering,China University of Geosciences(Wuhan),Wuhan 430074,China;XCMG Foundation Engineering Machinery Co.,Ltd.,Xuzhou Jiangsu 221000,China;School of Civil Engineering,Sun Yat-sen University,Guangzhou 510275,China)

机构地区：[1]中国地质大学(武汉)工程学院,武汉430074 [2]徐工基础工程机械有限公司,江苏徐州221000 [3]中山大学土木工程学院,广州510275

出　　处：《地质科技通报》2025年第1期164-174,共11页Bulletin of Geological Science and Technology

基　　金：国家重点研发计划(2022YFC3005600);徐工基础工程机械有限公司项目“超长距离水平定向钻进智能地质勘察技术研究”(SKSU-76140-20220623-0001)。

摘　　要：岩石单轴抗压强度是岩土工程中的重要参数,合理确定其数值对工程设计至关重要。本文提出了一种基于多层感知机的人工神经网络(MLP-ANN)模型,用于预测岩石单轴抗压强度。该模型以岩性、节理面、施密特锤回弹高度和纵波波速为输入参数,采用最大最小归一化进行参数标准化,并通过k折交叉验证提高模型的泛化能力。为优化模型性能,文章探讨了神经元数量、数据分割比例和激活函数对预测结果的影响。经对比验证,研究确定了最优模型配置:神经元数量为8,训练集与测试集比例为8∶2,激活函数选用Tanh函数。模型预测值与实际值对比分析结果表明,最优模型的平均绝对误差为3.500 MPa,均方根误差为5.836 MPa。结果表明,该模型预测误差较小,预测准确率较高,具有较好的实用性。[Objective]The uniaxial compressive strength(UCS)of rock is an important parameter in geotechnical engineering,as well as accurately determining its value is crucial for engineering design.[Methods]This study proposed a machine learning model based on a multi-layer perceptron-Artificial Neural Network(MLP-ANN)to predict the UCS of rock.The model takes lithology,joint surfaces,Schmidt hammer rebound height,and P-wave velocity as input parameters,and applies min-max normalization to standardize these parameters.Additionally,k-fold cross-validation is used to improve the model’s generalization ability.To further optimize model performance,the paper explores the impact of the number of neurons,data splitting ratio,and activation function on prediction results.[Results]Through comparative validation,the study determines the optimal model configuration:8 neurons,a training-to-testing ratio of 8∶2,and the Tanh activation function.The comparison between predicted and actual values shows that the optimal model achieves an average absolute error of 3.500 MPa and a root mean square error of 5.836 MPa.[Conclusion]These results indicate that the model has a small prediction error and high accuracy,illustrating good practicality.

关 键 词：单轴抗压强度 施密特锤实验 人工神经网络 模型评价 回弹法 

分 类 号：TU411.7[建筑科学—岩土工程]