摘要
注浆量是反映灌浆施工质量的重要指标之一。目前基于机器学习的注浆量预测方法缺乏对裂隙倾向、倾角等参数的全面考虑。裂隙灌浆模拟具有能够综合考虑地质、设计、施工等因素影响的优势,然而面临裂隙参数小样本、计算效率低下的不足。针对上述问题,提出基于改进混合核极限学习机(ICSO-MKELM)的注浆量预测代理模型。主要包括:(1)提出基于改进bootstrap方法的三维随机裂隙网络模型建模方法,解决裂隙数据小样本问题,并结合离散元方法开展灌浆数值模拟研究;(2)建立基于改进混合核极限学习机的注浆量预测代理模型,采用改进的鸡群算法优化混合核极限学习机的参数选择,克服混合核极限学习机参数选择效率不高、且难以有效选择全局最优参数的不足。通过将建立的代理模型应用于某工程坝基帷幕灌浆的注浆量预测,并与基于RBF-KELM单核极限学习机模型、Poly-KELM单核极限学习机模型、BP神经网络模型的注浆量预测结果对比,验证了本文所提方法的优越性。
Grouting volume is one of the important indicators reflecting the quality of grouting construction.The current grouting volume prediction method based on machine learning lacks a comprehensive consideration of fracture dip angel,fracture dip direction and other parameters.Grouting numerical simulation has the advantage of comprehensively considering the influence of geology,design,construction and other factors.However,it still faces the lack of fracture parameter samples and low calculation efficiency.In response to the above problems,a surrogate model for grouting volume prediction based on the improved multiple kernel extreme learning machine(ICSO-MKELM)is proposed,which mainly includes:(1)A three-dimensional stochastic fracture network modeling method based on the improved bootstrap method is proposed to solve the problem of small samples of fracture data.And based on the established stochastic fracture network model,numerical simulation of fracture grouting using the three-dimensional discrete element method is proceeded.(2)A surrogate model of grouting volume prediction based on the improved multiple kernel extreme learning machine is established.The improved chicken swarm optimization is used to optimize the parameter selection of the multiple kernel extreme learning machine,which overcomes the inefficiency of parameter selection,and improves the efficiency of the global optimal parameter selection.At last,the proposed surrogate model is applied to predict the grouting volume of a dam foundation curtain grouting construction.Compared with the prediction results of grouting volume based on RBF-KELM single-core extreme learning machine model,Poly-KELM single-core extreme learning machine model,and BP neural network model,the superiority of the proposed method is proved.
作者
石祖智
常峻
吴斌平
佟大威
郭辉
乔天诚
SHI Zuzhi;CHANG Jun;WU Binping;TONG Dawei;GUO Hui;QIAO Tiancheng(State Key Laboratory of Hydraulic Engineering Simulation and Safety,Tianjin University,Tianjin 300350,China;China Water Resources Beifang Investigation Design&Research Co.,Ltd.,Tianjin 300222,China)
出处
《水利水电技术（中英文）》
北大核心
2021年第9期57-66,共10页
Water Resources and Hydropower Engineering
基金
国家重点研发计划(2018YFC0406704)
国家自然科学基金(51839007)
天津市自然科学基金(19JCQNJC06800)。
关键词
注浆量预测
代理模型
改进bootstrap方法
三维随机裂隙网络
离散元数值模拟
混合核极限学习机
改进的鸡群算法
grouting volume prediction
surrogate model
improved bootstrap method
three-dimension stochastic fracture network model
three-dimension discrete element numerical
multiple kernel extreme learning machine
improved chicken swarm optimization
