基于贝叶斯网络的堰塞坝稳定性快速评价模型

Rapid evaluation method of landslide dam stability based on Bayesian network

为了科学、合理地评价堰塞坝稳定性,收集了国内外313个已溃和未溃堰塞坝的基础信息,建立了案例数据库;考虑堰塞坝形态特征、物质组成以及堰塞湖水动力条件包含的8个影响参数,基于贝叶斯网络提出了两种新的堰塞坝稳定性快速评价模型,分别为节点缺失条件下的简化模型和节点完整条件下的精细化模型。这两种新模型与国内外典型的堰塞坝稳定性快速评价模型的对比结果表明:基于贝叶斯网络的简化模型的绝对准确率和保守准确率分别为83.33%和94.44%,精细化模型的绝对准确率和保守准确率分别为80.77%和96.15%,验证了贝叶斯网络模型的可靠性和科学性,且贝叶斯网络模型可以实时输入和更新样本集,从而不断提高模型预测精度。

To perform scientific and rational evaluation on landslide dam stability,essential information was undertaken from 313 collapsed and intact landslide dams worldwide to establish a case database.Eight influential factors related to the morphological characteristics and material composition of landslide dams,hydrodynamic conditions of barrier lakes were considered.Based on the analysis of Bayesian networks,two novel and expedited methods for assessing landslide dam stability were introduced,including a simplified model with missing nodes and a detailed model with intact nodes.The comparison between typical models of rapid evaluation method of landslide dam stability from both home and aboard shows that the simplified Bayesian network model can achieve an absolute accuracy rate of 83.33%and a conservative accuracy rate of 94.44%.Similarly,the detailed model demonstrates absolute and conservative accuracy rates of 80.77%and 96.15%,respectively.The reliability and scientificity of the Bayesian network model are verified.The Bayesian network models can incorporate and update real-time sample sets,and thereby the predictive accuracy of the models can be increased perpetually.