基于神经网络的黄石市区岩土力学参数预测研究  

Prediction of rock and soil mechanical parameters in Huangshi Citybased on neural network

作　　者：王雨萌 吴佳 张潮 章小龙 李国敏 宋建邦 Wang Yumeng;Wu Jia;Zhang Chao;Zhang Xiaolong;Li Guomin;Song Jianbang(School of Environmental Study,China University of Geosciences(Wuhan),Wuhan 430078,China;Yejin Geological Team of Hubei Geological Bureau,Shiyan 435004,China)

机构地区：[1]中国地质大学(武汉)环境学院,湖北武汉430078 [2]湖北省地质局冶金地质勘探大队,湖北十堰435004

出　　处：《资源环境与工程》2025年第2期224-233,共10页Resources Environment & Engineering

基　　金：湖北省地质局冶金地质勘探大队科技基金项目(YJDKY2022-03)。

摘　　要：由于黄石市区新时期经济发展对城市地质工作的要求进一步提高,传统方法确定岩土力学参数存在的人为因素对试验结果有一定影响、时效性低等不足仍需进一步改进。随着机器学习不断向各个领域的深入,其方法也逐渐应用于岩土工程领域。以黄石市区主要分布的第四系全新统湖冲积层(Qh^(l+al))硬塑粉质黏土和第四系全新统冲洪积层(Qh^(al+pl))淤泥质土为例,通过计算其物理参数与力学参数的灰色关联度,选取相关性较大的物理参数,根据3σ原则排除原始数据的异常值,最后运用BP与GA-BP神经网络模型预测土体力学参数,从而确定了压缩系数、压缩模量、凝聚力、内摩擦角4项力学参数通过2种神经网络模型计算得到的预测值,以及4项力学参数的误差指标RMSE值。研究结果表明,机器学习不仅明显改善了传统方法耗时长、人为因素影响较大的缺陷,还提升了土体力学参数的预测精度,在一定程度上排除了偶然误差的干扰。检测结果验证了该方法的可行性与可靠性,可为不同类型岩土力学参数的预测提供参考。Due to the economic development of Huangshi City in the new period,the requirements of urban geological work are further improved.The human factors existing in the traditional method to determine the rock and soil mechanical parameters have a certain influence on the experimental results,and some deficiencies such as low timeliness need to be further improved.With the deepening of machine learning in various fields,its methods are gradually applied to the field of geotechnical engineering.The paper takes Qh^(l+al) hard plastic clay and Qh^(al+pl) muddy soil that are mainly distributed in the downtown area of Huangshi City as examples.By calculating the gray correlation degree between physical parameters and mechanical parameters,the physical parameters with greater correlation are selected,and outliers of original data are excluded according to the 3σprinciple.Finally,BP and GA-BP neural network models are used to predict soil mechanical parameters.The predicted values of four mechanical parameters,such as compression coefficient,compression modulus,cohesion and internal friction angle,calculated by two neural network models,and the error index RMSE values of four mechanical parameters are determined.The research results show that machine learning not only greatly improves the defects of traditional methods,which are time-consuming and greatly influenced by human factors,but also greatly improves the prediction accuracy of soil mechanical parameters,eliminating the interference of accidental errors to a certain extent.The detection results verify the feasibility and reliability of this method,which can provide references for the prediction of different types of rock and soil mechanical parameters.

关 键 词：岩土力学参数 灰色关联度分析 神经网络模型 误差分析 

分 类 号：TU452[建筑科学—岩土工程]