基于粗糙集-神经网络的IBURI地震滑坡易发性研究

Susceptibility of landslides caused by IBURI earthquake based on rough set-neural network

强震引发的滑坡范围广、数量多,呈现明显的区域性特征,并受强震动外的降雨等环境因素以及区域性水文地质等多种复杂因素的影响。在地震滑坡易发性评价中制定合理的指标选择策略能够从根本上提升评价精度和评价效率。以2018年9月6日北海道IBURI地区的Mj 6.7级地震引发的3 307处滑坡为研究对象,详细研究地震滑坡评价指标体系的构建。首先,结合地震灾害现场调查及资料分析,在充分考虑地震特性的基础上选择17项原始评价指标,然后,引入粗糙集理论作为指标选择策略,利用粗糙集对不确定数据的约简能力删除9项对评价结果影响较小的因子。最后,用优化后的指标构建BP神经网络的输入层,采用粗糙集-BP神经网络模型对IBURI地震滑坡易发性进行评价。结果显示,模型的预测精度从63.8%提升至94.4%,说明以粗糙集理论作为指标选择策略的粗糙集-BP神经网络模型能够有效提高地震滑坡易发性评价的准确性。

The landslides induced by strong earthquakes are wide in distribution and large in quantity,showing obvious regional characteristic. Besides intense ground motion,earthquake-triggered landslides are also affected by environmental conditions such as rainfall,regional hydrogeology and other complex factors. Reasonable index selection strategy can fundamentally improve the accuracy and efficiency of the susceptibility evaluation of seismic landslides. The present research takes 3307 landslides triggered by Mj 6.7 Hokkaido Eastern Iburi earthquake on September 6,2018,as the study object to analyze the construction of evaluation indicator system. First,based on the comprehensive consideration of seismic characteristics,17 original evaluation indicators are selected according to the earthquake disaster investigation and data analysis. Then,rough set theory is introduced as the index selection strategy,and the reduction ability of the rough set is adopted to delete 9 factors with less relevance to landslides. Finally,the input layer of BP neural network is constructed with the optimized indicator system,and a rough set-BP neural network model for evaluating the susceptibility of seismic landslides is established. The results show that the prediction accuracy of the model is improved from 63.8% to 94.4%,indicating that the rough set-BP neural network model can effectively improve the accuracy of seismic landslide susceptibility evaluation.