基于粒子群优化BP神经网络的隧道围岩位移反演分析

Research on Displacement Back-analysis of Tunnel Surrounding Rock Based on BP Neural Network Optimized by Particle Swarm

针对无锡惠山隧道岩体破碎、围岩稳定性差等特点,基于长期现场监测变形位移数据,借助粒子群算法的参数优化功能,利用Matlab神经网络工具箱编制了优化PSO—BP隧道位移反分析系统。PSO—BP系统利用正交试验设计和有限元方法获得学习样本,再通过粒子群算法搜索最优的神经网络模型参数。用BP神经网络模型建立待反参数与实测位移之间的非线性映射关系,最后用粒子群算法从全局空间上搜索最优反演参数。克服了普通智能优化算法收敛速度慢、正分析计算量大等缺陷,具有全局优化特性。将模型应用于惠山隧道Ⅳ级围岩断面ZK6＋485的反分析中,计算结果与实测值对比表明采用PSO—BP预测模型进行隧道位移预测是可行的。

According to such characteristics as the crushed rock and the inferior stability of surrounding rock of Huishan Tunnel in Wuxi,and based on the displacement data monitored in site and by meams of particle swarm optimization,the PSO-BP displacement back-analysis system is drawn up by using the Matlap neural network tool box.The system uses the orthgonal test design and finite element method to obtain the learning sample,and uses the particle swarm algorithm to search out the optimum neural network model parameters.At the same time,the non-linear reflecting relationship between the stand-by inverse parameter and measured displacement is set up by using the BP neural network model.At last,the particle swarm algorithm is used to search out the optimum inversion parameters in the global space.The method overcomes some defects of normal algorithms such as the slow convergence speed and the large calculation discharge of positive analysis,and has the global optimization characteristics.The applied results in Huishan Tunnel show that it is feasible to use the PSO-BP model in displacement detection of tunnels.