基于人工智能方法的隧道塌方风险预测研究  

Research on Tunnel Collapse Risk Prediction Based on Artificial Intelligence Method

作　　者：刘志锋[1,2] 陈名煜 吴修梅 魏振华 LIU Zhifeng;CHEN Mingyu;WU Xiumei;WEI Zhenhua(East China University of Technology,Nanchang 330013,Jiangxi,China;Jiangxi Province Geological Environment and Underground Space Engineering Research Center,Nanchang 330013,Jiangxi,China;Jiangxi Province Key Laboratory of Water Information Cooperative Sensing and Intelligent Processing,Nanchang 330099,Jiangxi,China)

机构地区：[1]东华理工大学,江西南昌330013 [2]江西省地质环境与地下空间工程研究中心,江西南昌330013 [3]江西省水信息协同感知与智能处理重点实验室,江西南昌330099

出　　处：《水力发电》2024年第3期31-38,共8页Water Power

基　　金：第3批江西省“双千计划”项目(DHSQT32022001);江西省水信息协同感知与智能处理重点实验室开放基金项目(2016WICSIP008)。

摘　　要：为了对隧道塌方风险展开研究,整理246起隧道塌方事故案例,通过建立塌方风险评估指标体系,基于人工智能预测方法,分别采用随机森林算法、径向基函数神经网络、BP神经网络模型、粒子群算法优化BP神经网络模型,对塌方风险进行预测。结果表明,随机森林算法、径向基函数神经网络、BP神经网络模型、粒子群算法优化BP神经网络模型的塌方预测准确率分别为81.67%、83.33%、86.67%、93.33%,F_(1)值分别为0.645、0.642、0.5、0.833。粒子群算法优化BP神经网络模型预测准确率和F_(1)值均大幅提高,预测效果最好,大大减少了评估结果的主观性,为隧道塌方风险研究提供了新的研究思路。In order to research the tunnel collapse risk,246 tunnel collapse accident cases are sorted out.By establishing the collapse risk assessment index system,the random forest algorithm(RF),radial basis function neural network(RBFNN),BP neural network(BPNN)and particle swarm optimization algorithm optimized BP neural network model(PSO-BPNN)are used to build models based on artificial intelligence prediction method,respectively,for predicting collapse risk.The results show that the collapse prediction accuracy of RF,RBFNN,BPNN and PSO-BPNN are 81.67%,83.33%,86.67%and 93.33%,respectively,and the F_(1) score values are 0.645,0.642,0.5 and 0.833,respectively.The prediction accuracy and F_(1) score of PSO-BPNN are greatly improved,and the prediction effect is the best,which greatly reduces the subjectivity of evaluation results,and provides a new idea for the research of tunnel collapse risk.

关 键 词：隧道工程 塌方 风险预测 随机森林算法 径向基函数神经网络 BP神经网络 粒子群算法 

分 类 号：U458[建筑科学—桥梁与隧道工程]