岩体抗剪强度参数神经网络反分析方法改进  

Improvement of the Neural Network Inverse Analysis Method of the Rock Mass Shear Strength Parameters

作　　者：姜照容[1] 于志伟[1] 

机构地区：[1]三峡大学土木与建筑学院,443002

出　　处：《灾害与防治工程》2013年第2期42-47,共6页Disaster and Control Engineering

摘　　要：采用神经网络反演岩体抗剪强度参数，当把安全系数作为网络输入时，由于网络输入参数个数小于输出个数，神经网络无法建立输入与输出之间的映射关系，导致反演结果误差偏大。针对这一问题，提出一种用于岩体抗剪强度参数神经网络反分析的新方法：首先利用神经网络反分析确定反演抗剪强度参数的范围，然后利用神经网络正分析在上述范围内进行仿真，预测其对应的安全系数，最后用最优化方法，把安全系数的预测值与反演值差值的绝对值作为目标函数，找出与反演工况下安全系数差值最小的安全系数相对应的粘聚力与内摩擦角组合，认为该组合即为最终反演的岩体抗剪强度参数。工程实例表明，由该方法得到的岩体抗剪强度参数计算出的安全系数与反演工况下安全系数相等，说明该方法能有效解决上述问题，是可行的。The neural network is used for inversion of rock mass shear strength parame- ters, but when the numbers of the network input parameters are fewer than those of the output parameters, it results in that the neural network can not establish the mapping be- tween the input and output, with large errors. In response to this problem, this paper pro- poses a new method of neural network inverse analysis. Firstly, it makes use of neural net- work inverse analysis to determine the range of inversion parameter, then simulates in the above range to predict the safety factor. Finally, with the optimization method, it takes the absolute difference value between predictive and inverse of the safety factor as the objective function, and finds out the cohesion and internal friction angle combinations sponding to minimal safety factor difference in the inversion conditions. The combina the rock shear strength parameters of inversion. The actual projects show that the factor obtained by the method is equal to that in the inversion conditions, indicatin this method is effective to solve the above problems, and it is feasible.

关 键 词：神经网络 抗剪强度参数 安全系数 反分析 最优化 

分 类 号：TU459.2[建筑科学—岩土工程]