基于DNN和属性数学的岩溶隧道突涌水风险预测

Risk Prediction of Water Inrush in Karst Tunnels Based on DNN and Attribute Mathematics

突涌水是岩溶隧道施工最常见的灾害,不仅会造成大面积涌泥塌方,支护设备等损毁破坏,甚至会造成大量人员伤亡,开展岩溶隧道突涌水灾害后果预测评估对保证隧道施工安全具有重要意义。为此,提出一种基于深度神经网络(DNN)和属性数学理论的岩溶隧道突涌水灾害预测方法,首先基于对国内外103条岩溶突涌水隧道的调查和总结,筛选了7个岩溶致灾因素和3个灾害后果特征指标,建立了包含489组岩溶隧道突水灾害信息数据库;其次,引入K均值聚类算法,根据致灾因素聚类结果和灾害后果将岩溶突涌水灾害划分为4个等级,并进一步分析了不同岩溶突涌水灾害等级对应的各致灾因素的分布区间;在此基础上,采用深度神经网络(DNN)、随机森林(RF)、K近邻(KNN)和AdaBoost四种算法进行岩溶隧道突涌水灾害等级预测,并根据属性数学理论实现了岩溶隧道突涌水灾害等级的概率预测;最后,对所提方法在湖北鸡公岭隧道等工程中进行了应用验证,将预测结果与实际情况进行对比分析,验证了所提方法的准确性和可靠性。相比以往岩溶突涌水灾害等级预测方法,所提出的结合深度神经网络和属性数学理论的岩溶隧道突涌水灾害预测方法不仅可实现对岩溶隧道突涌水灾害等级的自动划分,还可同时获得灾害等级发生的概率,对实际岩溶隧道突涌水灾害预测更为客观全面。

Water inrush is the most common disaster during karst tunnel construction,and it causes tunnel mud collapse,support,equipment damage,and substantial construction personnel casualties;therefore,predicting tunnel water inrush disasters is necessary.In this study,a method for predicting tunnel water inrush disasters based on a deep neural network(DNN)and attribute mathematical theory was proposed.First,based on an investigation of 103 karst water inrush tunnels worldwide,seven karst disaster indices and three disaster consequence indices were screened,and an extensive database comprising 489 datasets was established.Subsequently,the K-means clustering algorithm was introduced to cluster the tunnel water inrush data.Based on the clustering results and parameters of disaster consequences,the risk levels were classified into four classes,and reasonable distribution intervals of the calamitous parameters corresponding to each class were presented.Based on this,four algorithms,which were DNN,random forest,K-nearest neighbors,and AdaBoost,were employed to predict the levels of karst tunnel water inrush disaster.Furthermore,based on the attribute mathematical theory,the probability prediction of the karst tunnel water inrush disaster levels was achieved.Finally,the proposed method was applied and validated in engineering projects,such as the Jigongling Tunnel in Hubei Province.Compared with previous methods for predicting karst water inrush disasters,this study combines DNN and attribute mathematics theory to realize the automatic classification of karst tunnel water inrush disaster levels and obtain the probability,which is more objective and comprehensive for predicting actual karst tunnel water inrush disasters.