摘要
除室内试验和现场试验之外,反演分析是一种可以利用现有变形监测数据获取滑坡体等效力学参数的方法。以清江杨家槽滑坡体为例进行研究,提出了将均匀设计、遗传算法与BP神经网络结合起来应用于滑坡体反分析的新方法。先将具有很好全局寻优能力的改进遗传算法作为BP神经网络的学习算法,形成遗传神经网络;然后利用均匀设计方法设计网络学习样本,训练遗传神经网络映射滑坡体变形与滑坡体力学参数的非线性关系;最后将实测位移值作为网络输入,网络输出即为参数的反演值。该方法克服了优化反分析方法反演时间过长,解不易收敛等缺陷,实现了多参数的同时反演。通过对反分析结果进行检验与评价,证明其结果符合实际工程要求。
Back analysis is another way to gain equivalent mechanical parameters by actual deformation monitoring data except laboratory test and field test. As an example, this article applied uniform design, genetic algorithm and BP neural network to back analysing the mechanical parameters of Yangjiachao landslide, Qingjiang River. Firstly, the genetic neural network using the improved genetic algorithm as its learning algorithm was established. Secondly, network learning sample was generated by uniform design method, and the genetic neural network was trained for describing the sophisticated nonlinear relationship of dam displacements and material parameters. When the actual dam displacements were input into the trained network, the real material parameters could be obtained lastly. The above method may overcome the limitation of optimization back analysis method, i. e. , analyzing time is too long and the convergence of solution is difficult to insure, and realize the back analysis of many parameters at one time. The results show that this method is feasible and can be used in engineering practice.
出处
《长江科学院院报》
CSCD
北大核心
2005年第6期44-48,共5页
Journal of Changjiang River Scientific Research Institute
基金
水利部科技创新项目(SCX2003-21)
长江科学院资助项目(院基金监2002-1)
关键词
反分析
滑坡
均匀设计
遗传算法
BP网络
杨家槽
back analysis
landslide
uniform design
genetic algorithm
BP neural network
Yangjiachao