基于Inter.iamb-Tabu算法的区域滑坡敏感性评价  

Regional Landslide Susceptibility Assessment Based on Inter.Iamb-Tabu Algorithm

作　　者：尹超 李仲波 张凯 王绍平 郭兵 YIN Chao;LI Zhongbo;ZHANG Kai;WANG Shaoping;GUO Bing(School of Civil Engineering and Geomatics,Shandong University of Technology,Zibo 255049,Shandong,China;Shandong Jinding Zhida Group Co.,Ltd.,Zibo 255049,Shandong,China;Rizhao City Construction Investment Group Co.,Ltd.,Rizhao 276800,Shandong,China)

机构地区：[1]山东理工大学建筑工程与空间信息学院,山东淄博255049 [2]山东金鼎智达集团有限公司,山东淄博255049 [3]日照城投建设集团有限公司,山东日照276800

出　　处：《山地学报》2023年第4期608-620,共13页Mountain Research

基　　金：国家自然科学基金青年科学基金(51808327);山东省自然科学基金(ZR2019PEE016,ZR2021MD011)。

摘　　要：确定滑坡敏感性概率的空间分布,可为滑坡防治政策制定和土地利用规划提供科学依据。由于成灾环境的空间差异和滑坡机理的复杂性,基于不同逻辑架构(物理模型、理论模型等)的滑坡敏感性评价针对特定孕灾环境,无法提前确定最优数学评价模型,而基于深度学习法的、融合多种模型的混合算法能较好地解决这个问题。本文以山东省淄博市博山区为研究区域,基于单因素Logistic回归法筛选了滑坡致灾因子,通过频率比法对致灾因子进行了分级;对4种基于贝叶斯网络的改进算法(MMHC、MMPC-Tabu、Fast.iamb-Tabu、Inter.iamb-Tabu)进行验证,引入错误指数确定了建模效果最佳的算法;计算了博山区774570个栅格的滑坡敏感性概率,绘制了博山区滑坡敏感性分布图;基于GIS的空间叠加和栅格计算器功能,对比了各模型的滑坡敏感性评价结果。研究结果表明:(1)基于Inter.iamb-Tabu开展博山区滑坡敏感性建模时效果最佳。该模型包含10个节点(9个致灾因子节点和1个结局节点)、14条有向边。研究区域可划分为极高敏感区、高敏感区、中敏感区、低敏感区和极低敏感区,分别占总面积的7.30%(49.80km^(2))、16.50%(112.56km^(2))、26.10%(178.05km^(2))、33.20%(226.49km^(2))和16.90%(115.29km^(2)),分别有67处、22处、7处、2处和1处滑坡位于上述敏感区。(2)MMHC、MMPC-Tabu和Fast.iamb-Tabu模型在训练时易舍弃样本中部分因子的特征信息,以达到模型整体精度最优的目的。研究成果可为山东省淄博市博山区滑坡防治提供理论依据,采用的研究方法同样适用于类似地区。Determining the spatial distribution of landslide susceptibility probability can provide scientific basis for landslide prevention policy making and land use planning.Due to the spatial differentiation of geo-disaster-prone environment and the complexity of landslide formation,landslide susceptibility evaluation based on different logical frameworks(physical model,theoretical model,etc.)cannot determine the optimal mathematical evaluation model in advance for specific geo-environment,while the hybrid algorithm based on deep learning method and integrating of multiple models can better solve this problem.In this study,it took Boshan district of Zibo city,Shandong province of China for a case study.It selected landslide-triggering factors by using univariate logistic regression method,and weighed them by frequency ratio method.Based on Bayesian Network,four improved algorithms,i.e.,MMHC,MMPC-Tabu,Fast.iamb-Tabu and Inter.iamb-Tabu were verified for suitability,followed by the best algorithm to be determined for landslide susceptibility assessment by introducing the error exponent into all models.Landslide susceptibility probabilities of 774570 grids in Boshan district were calculated and the resulting zoning map was plotted.Based on the spatial overlay and raster calculator function of GIS,the landslide susceptibility assessment results of each model were compared.This research found(1)the Inter.iamb-Tabu model had the best performance in landslide susceptibility assessment.It contained 10 nodes(9 landslide-triggering factor nodes and 1 ending node)and 14 directed edges.Accordingly,Boshan district could be zoned into five areas,i.e.,extremely high susceptible areas,high susceptible areas,medium susceptible areas,low susceptible areas and extremely low susceptible areas,accounting for 7.30%(49.80 km 2),16.50%(112.56 km 2),26.10%(178.05 km 2),33.20%(226.49 km 2)and 16.90%(115.29 km 2)of the whole areas of Boshan district,respectively.There were 67,22,7,2 and 1 landslide identified in aforesaid five respective areas.(

关 键 词：滑坡敏感性评价 致灾因子 贝叶斯网络 Inter.iamb-Tabu算法 

分 类 号：P694[天文地球—地质学]