摘要
堆石坝变形包括瞬时弹塑性变形及时间相关的流变变形,从实际的监测变形中精确区分这两种变形有一定技术难度。本文将瞬时及流变变形参数的反演问题转化为一个组合优化问题,采用智能优化算法寻找最佳的堆石变形参数。研究中,首先拟定了多种变形参数样本,采用有限元法计算坝体变形;然后采用径向基神经网络训练上述样本,建立堆石变形参数与坝体变形之间的映射关系;最后根据坝体实际变形测量值,采用多种群遗传算法优化得到坝体瞬时及流变变形参数。采用径向基神经网络替代有限元可节省计算时间,提高计算效率;而多种群遗传优化算法可避免传统遗传算法早熟问题。用反演参数再次计算得到的水布垭坝体沉降与实测值吻合较好。
Settlement of rockfill dams include instantaneous elastic-plastic deformation during constructionand time-dependent rheological deformation after water impoundment,but it is difficult to distinguish thesetwo kinds of deformation accurately from actual monitoring deformation.In this paper,a back analysis meth-od was proposed to obtain the instantaneous and rheological deformation parameters successively by the com-binatorial intelligent optimization algorithm.Firstly,dam deformation was calculated by the finite elementmethod using some prepared parameter samples.Then a RBF neural network has been trained using thesesamples to establish a mapping relationship between the parameters and dam deformation.Thirdly,the in-stantaneous and rheological deformation parameters of the dam have been determined by multiple populationgenetic optimization algorithm according to the actual dam deformation measurements.The computing timeof dam deformation has been saved greatly by RBF neural network instead of finite element method,andthe precocious problem can be avoided by the multiple population genetic algorithm.The recalculated settle-ment values of Shuibuya concrete faced rockfill dam using the inversion parameters are well agreed withthe actual measured values.
出处
《水利学报》
EI
CSCD
北大核心
2016年第1期18-27,共10页
Journal of Hydraulic Engineering
基金
国家自然科学基金项目(51179024
51379029)
关键词
堆石流变
参数反演
RBF神经网络
多种群遗传算法
rockfill dams
creep
parameter inversion
RBF neural network
multiple population genetic algorithm.
