基于GA-BP神经网络的隧道围岩力学参数反演

Inversion of Mechanical Parameters of Tunnel Surrounding Rock Based on GA-BP Neural Network

为在岩土工程设计和施工中获得更加合理、可靠的岩体力学参数,改善BP神经网络算法存在的收敛速度慢、依赖初值等不足,采用GA-BP神经网络方法,对隧道围岩力学参数反演进行了研究。依托江西省萍莲高速公路莲花隧道工程,选取右洞YK35+095~YK35+135段作为模拟对象,利用FLAC3D有限差分法,按照微台阶法开挖,构建三维数值计算模型;设计了25组围岩力学参数的正交试验,代入已建立的数值模型,得到系列拱顶沉降、周边位移以及地表沉降值,由此构造了神经网络的样本;采用遗传算法和神经网络相结合,其中遗传算法种群规模取值30,最大遗传代数取值25,交叉概率取值0.8,变异概率取值0.01,通过网络训练,得到了训练成熟的GA-BP神经网络,建立了待反演的围岩力学参数与位移之间的映射关系;将莲花隧道YK35+115断面实测的拱顶沉降、周边位移和地表沉降值,输入到已训练成熟的GA-BP神经网络模型中,输出得到围岩的弹性模量、泊松比、黏聚力、内摩擦角等参数。结果表明:采用GA-BP神经网络反演获得的围岩力学参数,代入到FLAC3D数值模型中正演计算,得到拱顶沉降、周边位移和地表沉降与现场实测值仅相差2.94%,3.16%和4.86%,误差较小;基于GA-BP神经网络的隧道围岩力学参数反演方法精度较高。

In order to obtain more reasonable and reliable mechanical parameters of rock mass in geotechnical engineering design and construction,and overcome the shortcomings of BP neural network algorithm such as slow convergence speed and dependence on initial values,the inversion of the mechanical parameters of tunnel surrounding rock is studied.Relying on Lianhua Tunnel Project of Pingxiang-Lianhua expressway in Jiangxi Province,selecting the right tunnel of YK35+095 toYK35+135 section as the simulation object,the 3 D numerical model of micro-step excavation is established by FLAC3 D finite difference method.The orthogonal experiment of 25 groups of surrounding rock mechanical parameters is designed and substituted into the established numerical model to obtain a series of vault settlement,peripheral displacement and surface settlement,thus the neural network sample is constructed.Combining genetic algorithm and neural network,in which the population size of genetic algorithm is 30,the maximum genetic generation is 25,the cross probability is 0.8,and the mutation probability is 0.01.Through network training,a mature GA-BP neural network is obtained,and the mapping relationship of the surrounding rock mechanical parameters to be inverted and the displacement is established.By inputting the measured values of vault settlement,peripheral displacement and surface settlement of the YK35+115 section of Lianhua Tunnel into the trained GA-BP neural network model,the elastic modulus,Poisson’s ratio,cohesion and internal friction angle of the surrounding rock are output.The result shows that(1)The mechanical parameters of the surrounding rock obtained by inversion using GA-BP neural network are substituted into the FLAC3 D numerical model.The vault settlement,periphery displacement and surface settlement are only 2.94%,3.16%and 4.86%different from the actual measured values,and the errors are small.(2)The inversion method based on GA-BP neural network has high accuracy for calculating the mechanical parameters of tunnel surrounding rock.