基于决策树的地下工程透水事故发生风险定量评估

Quantitative Risk Evaluation of Water Inrush Hazard Based on Decision Tree of Underground Engineering

地下工程建设深度和规模的不断扩大导致透水事故频发,严重制约着地下工程建设的安全作业。以国内外107个透水事故案例为基础,以瞬时最大涌水量和事故累计涌水量为指标,采用k均值聚类算法实现了透水事故灾害等级划分,并考虑水文地质条件、地层岩性和过程监测信息,采用决策树方法建立了透水发生风险评估模型。研究结果表明:透水事故发生风险评估指标与事故灾害等级具有较好的相关性,其中水文条件加权值、地下水位高程差、埋深、涌水量平均增长速率和累积涌水量与灾害等级呈正相关,而岩石饱和单轴抗压强度与灾害等级呈负相关;剪枝策略能够有效提高透水事故发生风险评估决策树模型的泛化性能,相对于初始决策树模型,该模型针对验证集和测试集的评估准确率分别从63.6%和90.5%提高至90.9%和100%,针对所有透水样本的综合评估准确率为95.3%,表明该模型具有较好的适用性,可为地下工程透水灾害的救灾决策提供重要的理论和技术支撑。

The continuous expansion of the depth and scale of underground engineering construction leads to frequent water inrush hazard,which seriously threats to production safety. In this paper,based on 107 cases of water inrush hazard occurred domestic and overseas,k-means clustering algorithm was used to divide the water inrush level by considering the instantaneous maximum water inflow and accumulated water inflow. Moreover,the decision tree method was applied to established a risk evaluation model of water inrush hazard by considering the hydrogeological conditions,stratum lithology and monitoring information. The results show that there is a good correlation between evaluation indexes and water inrush level,in which the weighted value of hydrological conditions,elevation difference of groundwater level,buried depth,average growth rate of water inflow and cumulative water inflow are positively correlated with water inrush level,and the saturated uniaxial compressive strength of rock is negatively correlated with water inrush level. The pruning strategy can effectively improve the generalization performance of decision tree model for risk assessment of water inrush hazard. Compared to the initial tree model,the evaluation accuracy of the improved model for the verification set and test set promoted from 63. 6% and 90. 5% to 90. 9% and 100% respectively.The comprehensive evaluation accuracy for all samples was 95. 3%,which indicated that the model established in this paper has better applicability and can provide important theoretical and technical support for water inrush hazard relief decision-making in underground engineering.