摘要
针对传统BP神经网络、遗传算法在反分析应用过程中存在的问题,将基于免疫算法优化的遗传算法与BP神经网络结合起来,构建了具有更快的收敛速度和更强的全局搜索性能的GA-BP网络,根据某抽水蓄能电站地下洞室的开挖和埋深特点,选取弹性模量和侧压力系数为待反演参数并设定取值范围,以设定的反演参数值和有限元计算得出的洞室理论位移为训练样本,利用GA-BP网络训练此样本,得到洞室位移值与洞室物理力学参数之间的关系,将实测位移值输入训练好的GA-BP网络中获得参数的反演值,通过反演值计算出不同监测断面的位移值,从而验证了GA-BP网络在参数反分析中应用的准确性。
Focusing on the problems of applying traditional BP neural network and genetic algorithm in back analysis, combined BP neural network with the genetic algorithm based on the optimization of immune algorithm, this paper built a new GA-BP network with a higher convergence speed and better global searching performance. According to the charac teristics of burial depth and excavation of underground chamber of a pumped storage power station, elastic modulus and lateral pressure coefficient are chosen as inversion parameters and set their value bounds. Using the GA BP network training samples which were made up of the given inversion parameter values and displacement values based on finite ele ment calculation, it got the relationship between physical mechanics parameters and displacement values of underground chamber. Then the actual displacement was input into the trained GA-BP network and it obtained the real material param eters. By means of inversion value, the displacement values of different monitoring sections are calculated. Thus, it verifies the accuracy of GA-BP network in application of parameter back analysis.
出处
《水电能源科学》
北大核心
2014年第3期141-144,共4页
Water Resources and Power
基金
国家自然科学基金项目(51139001)
关键词
BP神经网络
遗传算法
免疫算法
反分析
岩体力学参数
BP neural network
genetic algorithm
immune algorithm
back analysis
rock mass mechanics parameters