基于M-LSTM法的隧道围岩地质信息动态智能预测研究  

Dynamic Intelligent Prediction of Tunnel Surrounding Rock Geological Information Based on M-LSTM Method

作　　者：杨存斌 任洋[1,2] 吴岳华 何万超 李天斌 YANG Cunbin;REN Yang;WU Yuehua;HE Wanchao;LI Tianbin(State Key Laboratory of Geohazard Prevention and Geoenvironment Protection,Chengdu University of Technology,Chengdu 610059;College of Environmental and Civil Engineering,Chengdu University of Technology,Chengdu 610059)

机构地区：[1]成都理工大学地质灾害防治与地质环境保护国家重点实验室,成都610059 [2]成都理工大学环境与土木工程学院,成都610059

出　　处：《现代隧道技术》2025年第1期74-82,共9页Modern Tunnelling Technology

基　　金：国家自然科学基金(U19A20111,42130719);四川省自然科学基金(2022NSFSC0411);地质灾害防治与地质环境保护国家重点实验室自主研究课题(SKLGP2021Z011).

摘　　要：为提高隧道围岩地质信息智能化预测精度,以岩体完整性、岩石坚硬程度、富水情况、岩石风化程度和地应力状态作为地质信息的定量指标,通过收集隧道已开挖段的各类地质指标数据,并采用K-means聚类算法对数据进行清洗,得到隧道围岩各地质信息指标的高关联数据库。依据数据库样本,构建基于改进长短期记忆神经网络(M-LSTM)的隧道施工期围岩地质信息动态智能预测模型,根据对隧道已开挖段地质信息的智能学习实现对前方未开挖段地质信息时间序列数据的动态智能预测。结果表明,该方法得到的岩体完整性预测精度为91.6%,岩石坚硬程度预测精度为93.8%,富水情况预测精度为85.4%,岩石风化程度预测精度为85.4%,地应力状态预测精度为87.5%;M-LSTM法比LSTM法和普通神经网络(ANN)法具有更高的计算效率和精度。To improve the accuracy of intelligent prediction for tunnel surrounding rock geological information,quantitative indicators such as rock integrity,rock hardness,water abundance condition,rock weathering degree,and geostress state were used as geological information parameters.By collecting various geological indicator data from the excavated sections of the tunnel and using the K-means clustering algorithm to clean the data,a highly cor-related database for geological information indicators of tunnel surrounding rocks was established.Based on the sam-ple database,a dynamic intelligent prediction model for surrounding rock geological information during tunnel con-struction,based on an improved long short-term memory neural network(M-LSTM),was developed.This model en-ables dynamic intelligent prediction of the time-series geological information data for unexcavated sections based on intelligent learning from the geological information of the excavated tunnel sections.The results show that the pre-diction accuracy for rock integrity is 91.6%,rock hardness is 93.8%,water abundance condition is 85.4%,rock weathering degree is 85.4%,and geostress state is 87.5%.Meanwhile,the M-LSTM method demonstrates higher computational efficiency and accuracy compared to the LSTM method and the ordinary neural network(ANN)method.

关 键 词：隧道工程 地质信息预测 改进LSTM法 K-MEANS聚类算法 

分 类 号：U452.12[建筑科学—桥梁与隧道工程]