摘要
建立智能位移反分析系统,用其确定隧道围岩的力学参数.针对BP神经网络易陷入局部极小值和训练时间过长等缺点,利用遗传算法全局寻优能力优化BP神经网络的权值和阈值.结合均匀设计法在围岩力学参数初始域范围内设计实验方案,这样不仅减少了迭代时间和次数,还提高了预测精度.通过对绿春坝隧道围岩力学参数的反演,验证了该方法的可靠性及适用性.将反演得出的围岩力学参数代入到数值模型中进行计算,结果表明,数值计算值与现场实际监测值的误差分别为-8.9%和4.5%.
A displacement back analysis algorithm was developed for deriving the mechanical parameters of surrounding rocks. Given the drawbacks of BP neural network （BPNN） such as easily getting stuck in local minima and over long training time, a genetic algorithm with global optimization ability was used to optimize the weights and thresholds of the BPNN. The parameters of surrounding rocks were designed in the initial domain by the uniform method, which reduced the iterative time and improved the forecast accuracy. Applying the method to the back analysis of the mechanical parameters of Ltichunba railway tunnel, we introduced the parameters as obtained into the numerical model for computing. The results show that the errors of the calculated and measured values were - 8.9 % and 4.5 %, respectively, which illustrates the reliability and applicability of the method.
出处
《东北大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2012年第2期276-278,283,共4页
Journal of Northeastern University(Natural Science)
基金
国家自然科学基金资助项目(51179031
51074042)
辽宁省自然科学基金资助项目(20092011)
中央高校基本科研业务费专项资金资助项目(N090401008)
深部岩土力学与地下工程国家重点实验室开放基金资助项目(SKLGDUEK1009)
关键词
围岩
力学参数
反分析
均匀设计
BP神经网络
遗传算法
surrounding rock
mechanical parameter
back analysis
uniform design
BP neuralnetwork
genetic algorithm
