基于IA-BP智能算法的初始地应力场反演研究  

In-situ stress field inversion via IA-BP intelligent algorithm

作　　者：孙港 王军祥 郭连军 寇海军 徐景龙 SUN Gang;WANG Junxiang;GUO Lianjun;KOU Haijun;XU Jinglong(School of Architecture and Civil Engineering,Shenyang University of Technology,Shenyang 110870,P.R.China;China Railway 19th Bureau Group 5th Engineering Co.,Ltd.,Dalian 116100,Liaoning,P.R.China)

机构地区：[1]沈阳工业大学建筑与土木工程学院,沈阳110870 [2]中铁十九局集团第五工程有限公司,辽宁大连116100

出　　处：《土木与环境工程学报（中英文）》2023年第2期89-99,共11页Journal of Civil and Environmental Engineering

基　　金：国家自然科学基金(51774066);辽宁省自然科学基金(2019-MS-242);辽宁省教育厅重点攻关项目(LZGD2020004);中国博士后科学基金(2018M630293)。

摘　　要：初始地应力场是地下工程设计与施工的重要依据,在实际工程中难以精准测得,为了能较准确地获得初始地应力场的分布规律,提出将免疫算法与BP神经网络相结合(IA-BP)的算法对初始地应力场进行反演研究。免疫算法优化BP神经网络就是将BP神经网络的连接权值和阈值作为免疫算法中的抗体进行编码。该混合算法既可以利用免疫算法全局寻优的特点快速搜索到全局最优解或次优解附近,又可以采用BP算法避免在最优解和次优解附近发生震荡,对其进行局部优化,从而达到快速收敛全局最优解的目的。通过COMSOL分别构建平面边坡模型及三维立体模型,对其进行正分析计算,将计算的结果作为“实测值”,对地应力进行反演分析,并将IA-BP算法反演的结果与PSO-BP算法及多元线性回归算法的反演结果进行对比。结果表明:二维边坡模型下,IA-BP算法反演结果误差更小。三维模型下,IA-BP算法所得实测值与反演值之间的相对误差的绝对值为0%~10.64%(平均为3.39%),PSO-BP算法所得实测值与反演值之间相对误差的绝对值为0%~48.39%(平均为6.93%),多元线性回归算法所得实测值与反演值之间相对误差的绝对值为0.55%~121.95%(平均为21.87%),通过对比可知,IA-BP算法整体反演结果精度最高。无论是平面模型还是三维模型,利用IA-BP算法反演出的结果与其他两种算法反演的结果相比,误差更小。将IA-BP智能算法运用到地应力场的反演研究中,可以为地下工程的建设提供依据。Initial in-situ stress field is an important basis for design and construction of underground engineering,while it is difficult to accurately measure the initial in-situ stress in engineering practice.In order to accurately obtain distribution law pattem of initial geostress field,the immune algorithm combined with BP neural network(IA-BP algorithm)for inversion of initial in-situ stress field is studied.The optimization of BP neural network by immune algorithm is to encode the connection weights and thresholds of BP neural network as antibodies in the immune algorithm.The hybrid algorithm can not only take advantage of the characteristics of immune algorithm as well as the global optimization quick search for the global optimal solution or near optimal solution,and can also adopt BP algorithm to avoid the near optimal and sub-optimal solutions,oscillation on the local optimization,realizing the aim of fast converge of the global optimal solution.The plane slope model and three-dimensional model were constructed by COMSOL respectively to carry out forward analysis and calculation,and the calculated results were taken as“measured values”to conduct inversion analysis of in-situ stress,but the inversion results of IA-BP algorithm were compared with those of PSO-BP and multiple linear regression algorithms.The results show that the inversion error of IA-BP algorithm is smaller under the two-dimensional(2D)slope model.Under the three-dimensional(3D)model,IA-BP algorithm from the measured values and the inversion of the absolute value of relative error between 0% and 10.64%(3.39%on average),for PSO-BP,it is between 1% and 48.39% with average value of 6.93%,for MLR,it is between 0.55% and 121.95% with average of 21.87%.By comparison,it can be known that the overall inversion results of IA-BP algorithm have the highest accuracy.In both 2D and 3D models,the error of the inversion results is smaller than those using the other two.The application of IA-BP intelligent algorithm to the inversion of in-situ stress field ca

关 键 词：地应力场 免疫智能算法 粒子群智能算法 多元线性回归 

分 类 号：TU431[建筑科学—岩土工程]