基于BR-BP神经网络的高液限黏土改性剂掺量预估方法  

Prediction method of high liquid limit clay modifier dosage based on BR-BP neural network

作　　者：李臣光 陈亮胜 梁松林 张寒冰 韦秉旭[3] LI Chenguang;CHEN Liangsheng;LIANG Songlin;ZHANG Hanbing;WEI Bingxu(Guangxi Road Construction Engineering Group Co.,Ltd.,Nanning 650500,Guangxi,China;Luoyang Branch,Henan Provincial Communications Planning&Design Institute Co.Ltd.,Luoyang 471000,Henan,China;School of Transportation Engineering,Changsha University of Science and Technology,Changsha 410114,Hunan,China)

机构地区：[1]广西路建工程集团有限公司,广西南宁650500 [2]河南省交通规划设计研究院股份有限公司洛阳分公司,河南洛阳471000 [3]长沙理工大学交通运输工程学院,湖南长沙410114

出　　处：《水利水电技术（中英文）》2023年第10期190-202,共13页Water Resources and Hydropower Engineering

基　　金：广西交通运输厅重点科技项目(桂交便函[2022]第二批174号);国家自然科学基金(52178413)。

摘　　要：【目的】为预测在不同湿度、温度、初始含水率环境下高液限黏土降水至稳定含水率所需的水泥和石灰改性剂掺量,【方法】以广西来都高速的高液限黏土为研究对象,结合室内湿法击实试验、液塑限试验及改进CBR试验,设计考虑初始含水率、改性剂掺量、温度、湿度4因素5水平的正交试验对稳定含水率的影响,基于贝叶斯正则化(BR)算法的改进BP神经网络,建立了考虑多因素的高液限黏土改性剂掺量预估模型,将该模型与传统人工神经网络的预测结果进行比较,并采用现场试验验证了该模型预测结果的可靠性。【结果】结果显示:以湿法最佳含水率21%作为降水目标值是可行的,所提出基于改进BP神经网络的改性剂掺量预估模型较传统的人工神经网络预测结果更为精确。【结论】结果表明:通过该方法预测施工现场不同工况下所需石灰、水泥掺量,现场焖料后含水率和模型计算的稳定含水率均低于21%,石灰、水泥改性土填筑碾压后路堤压实度均可达到94%以上。研究成果对于高液限黏土改性剂掺量预测及路堤填筑技术具有借鉴意义。[Objective]In order to predict the amount of cement and lime modifier required for dewatering high liquid limit clay to stable water content under different humidity,temperature and initial water content.[Methods]Taking the high liquid limit clay of Laidu Expressway in Guangxi as the research object,combined with the indoor wet compaction test,liquid-plastic limit test and improved CBR test,the orthogonal test considering the influence of 4 factors and 5 levels of initial water content,modifier content,temperature and humidity on the stable water content was designed.Based on the improved BP neural network of Bayesian regularization(BR) algorithm,a prediction model of high liquid limit clay modifier content considering multiple factors was established,the prediction result of the model were compared with those of the traditional artificial neural network,and the reliability of the prediction result of the model was verified by field tests.[Results]It was feasible to take the optimum moisture content of wet method as the precipitation target value,and the modified BP neural network based modifier dosage prediction model was more accurate than the traditional artificial neural network prediction result.[Conclusion]This method is used to predict the amount of lime and cement required at the construction site under different working conditions.The water content after the site soaking and the stable water content calculated by the model are lower than 21%,and the compactness of the embankment after the lime and cement modified soil filling and rolling can reach more than 94%.The research result have reference significance for the prediction of the content of high liquid limit clay modifier and the embankment filling technology.

关 键 词：高液限黏土 改性剂 正交试验 BR-BP神经网络 预估模型 

分 类 号：U416[交通运输工程—道路与铁道工程]