深圳市岩溶地层标准贯入击数神经网络模型  

An artificial neural network for standard penetration blow counts of karst strata in Shenzhen

作　　者：严辉 林沛元[2,3] YAN Hui;LIN Peiyuan(Guangdong Pearl River Delta Intercity Railway Co.LTD,Guangzhou 510000,China;a.State Key Laboratory for Tunnel Engineering,Sun Yat-Sen University,Guangzhou 510275,China;School of Civil Engineering,Sun Yat-Sen University,Guangzhou 510275,China)

机构地区：[1]广东珠三角城际轨道交通有限公司,广州510000 [2]中山大学隧道工程灾变防控与智能建养全国重点实验室,广州510275 [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2期305-321,共17页Bulletin of Geological Science and Technology

基　　金：国家自然科学基金项目(52008408)。

摘　　要：潜在岩溶地质灾害威胁粤港澳大湾区广州、深圳等核心城市安全及其地下空间开发与利用。标准贯入试验是岩溶地层勘察的必备手段之一,为土层划分、承载力评估、基础选型等提供重要依据。针对传统的标准贯入试验提高工程成本并受操作人员技能水平影响较大的问题,本研究提出了一种快速且准确获取岩溶区土层标贯击数的新方法。以深圳市岩溶区为例,收集了1 006组土层标贯数据,建立了一个11-5-1结构的单隐藏层神经网络模型,该模型仅拥有5个神经元,具有解析解,易于计算。研究结果显示,该神经网络模型的决定系数达到了0.93,表明模型拥有高度的准确性;模型因子平均值为1.04,变异系数介于9%~23%。总体上,模型精度高,预测偏差离散性低。讨论了影响模型稳定性和预测性能的多种因素,如隐藏层神经元数量、数据标准化方法、激活函数选择、数据分割比例和随机抽样效应等。通过在深圳市龙岗区2个独立工程案例的应用,验证了该神经网络模型在工程实践中的实用价值。本研究为未来岩溶区工程勘察方法的发展提供了重要参考。[Objective]This study addresses the threat of potential karst geological disasters to the core cities of the Guangdong-Hong Kong-Macao Greater Bay Area,such as Guangzhou and Shenzhen,impacting the safety and development of their underground spaces.Standard penetration testing,a crucial method for investigating karst strata,plays a vital role in soil layer classification,load-bearing capacity evaluation,and foundation selection.However,traditional standard penetration tests can escalate project costs and are significantly influenced by the skill level of the operators.[Methods]To deal with these challenges,this paper introduces a new method for rapidly and accurately obtaining standard penetration test data in karst areas.Focusing on the karst regions of Shenzhen,we collected 1006 sets of soil penetration data and developed a single hidden layer neural network model with an 11-5-1 structure;this model is featured by only five neurons and has an analytical form that enables easy computation.[Results]The research findings reveal that this neural network model has a high determination coefficient of 0.93,indicating its high accuracy in prediction.The model factor has a mean value of 1.04,with a coefficient of variation(COV)ranging between 9%and 23%.Overall,the model demonstrates high precision and low predictive dispersion.The paper thoroughly examines various factors affecting the model's stability and predictive performance,including the number of neurons in the hidden layer,data normalization methods,choice of activation functions,data splitting ratios,and the impact of random sampling.The practical applicability of this neural network model has been validated through its implementation in two independent engineering projects in the Longgang District of Shenzhen.[Conclusion]This study offers significant insights for the advancement of engineering survey methods in karst regions.

关 键 词：岩溶地层 粤港澳大湾区 深圳市 标准贯入试验 人工神经网络 模型评价 

分 类 号：TU413.3[建筑科学—岩土工程]