基于BP神经网络对隧道初衬砼应力的评价分析对比预测
Evaluation Analysis and Comparison Prediction of Initial Lining Concrete Stress in Tunnels Based on BP Neural Network
郭剑锋 1刘少凯 2刘秀 2吴勇 1刘伟 2寿凌超 3王立峰4
作者信息
- 1. 中国电建集团华东勘测设计研究院有限公司,杭州 310012
- 2. 合肥市轨道交通集团有限公司,合肥 230001
- 3. 浙江明燧科技有限公司,杭州 311400
- 4. 浙江科技学院,杭州 310023
- 折叠
摘要
为确定杨家山特大断面隧道特定区段的最佳预测引起的初衬砼应力模型,本文采用BP(back propaga-tion)神经网络算法,以该区段初衬砼应力监测数据为输入值,使用5种方法训练网络,分析应力预测值和真实值的差异,给出预测误差的分布情况,以及网络训练过程中性能、验证和测试曲线.同时基于多目标优化问题的分析方法,对5种训练方式进行综合性评价.结果表明:从优到差的顺序依次为:Ploak-Ribiere共轭梯度法＞自适应动量梯度下降法＞拟牛顿算法＞Powell-Beale共轭梯度法＞Levenberg-Marquardt,Ploak-Ribiere共轭梯度法最优,预测准确度达到98%以上,故在后续隧道开挖过程中可通过Ploak-Ribiere共轭梯度法训练BP神经网络,对特定区段隧道所产生的初支与围岩应力进行有效预测,保证施工安全.
Abstract
In order to determine the best prediction induced initial lining concrete stress model for a specific section of Yangjiashan mega section tunnel.In this paper,the BP neural network method is adopted to train the network using five methods with the initial lining concrete stress monitoring data of the section as the input value,and to analyse the differ-ence between the predicted and real values of the stress,giving the distribution of the prediction error,as well as the performance,validation and testing curves of the network during the training process.A comprehensive evaluation of the five training methods is also carried out based on the analysis method of multi-objective optimisation problem.The re-sults show that the order from the best to the worst is as follows:Traincgp>Traingdx>Trainbfg>Traincgb>Trainlm,and Ploak-Ribiere conjugate gradient method is the best,and the prediction accuracy reaches more than 98%,so it can be used by Ploak-Ribiere conjugate gradient method during the subsequent tunnel Therefore,the BP neural net-work can be trained by the Ploak-Ribiere conjugate gradient method in the excavation process to effectively predict the initial support and peripheral rock stresses generated in a specific section of the tunnel to ensure the construction safety.
关键词
特大隧道/初衬砼应力/BP神经网络/训练方式对比Key words
super large tunnel/initial lining concrete stress/BP neural network/comparison of training methods引用本文复制引用
基金项目
浙江省重点研究计划(2021C01131)
出版年
2024
NETL
NSTL
万方数据