基于全连接神经网络的弥勒市域滑坡风险测度与分级评价  

Quantitative Risk Assessment of Landslides in Mile City Based on Fully Connected Neural Networks

作　　者：朱习松 常鸣[1] 周贤熙 赵伯驹 刘洋[1] ZHU Xisong;CHANG Ming;ZHOU Xianxi;ZHAO Boju;LIU Yang(State Key Laboratory of Geological Disaster Prevention and Geoenvironmental Protection,Chengdu University of Technology,Sichuan Chengdu 610059,China)

机构地区：[1]成都理工大学地质灾害防治与地质环境保护国家重点实验室,四川成都610059

出　　处：《防灾减灾学报》2025年第1期8-16,共9页Journal of Disaster Prevention And Reduction

基　　金：四川省自然科学基金(2024NSFSC0071);四川省中央引导地方科技发展专项项目(2024ZYD0121)。

摘　　要：弥勒市位于云南省东南部,作为滇中与两广地区的重要交通枢纽,农业与旅游资源丰富,但易受降雨引发的滑坡灾害威胁。为评估该区域滑坡风险,采用全连接神经网络模型,结合斜坡结构、断层距离和工程地质岩组等11个指标,进行滑坡易发性评价。引入降雨作为时间概率因子进行危险性评价。通过人口、建筑、道路及夜间灯光等数据,评价承灾体易损性,耦合滑坡危险性与易损性,构建弥勒市滑坡定量风险评价模型。结果表明:高和极高风险区主要集中在弥阳-西三镇、西二镇和东山镇;中风险区主要位于弥勒市东北部;低风险区则集中于西一镇、五山乡、巡检司镇和江边乡南部。通过ROC曲线验证,滑坡易发性评价模型的AUC值为0.8923,模型具有较高的精度。野外调查验证了两处新增滑坡,均位于高和极高风险区,证明风险评价结果的可靠性。研究为弥勒市的基础设施建设和滑坡灾害的早期预警提供了科学依据。Mile City,located in the southeastern part of Yunnan Province,serves as a key transportation hub between central Yunnan and the Guangdong-Guangxi region,with abundant agricultural and tourism resources.However,it is highly susceptible to landslide hazards triggered by rainfall.To assess landslide risk in this region,a fully connected neural network model was employed,integrating 11 indicators,including slope structure,fault proximity,and geological formations,to evaluate landslide susceptibility.Rainfall was introduced as a temporal probability factor for hazard assessment.Additionally,data on population,buildings,roads,and nighttime lighting were used to assess the vulnerability of exposed assets,combining landslide hazard and vulnerability to establish a quantitative landslide risk model for Mile City.Results indi⁃cate that high and very high-risk zones are concentrated in Miyang-Xisan Town,Xier Town,and Dongshan Town,with moderate risk primarily in the city’s northeast,and low-risk zones in Xiyi Town,Wushan Township,Xunjian Town,and southern Jiangbian Township.The ROC curve analysis verified the model’s accuracy,achieving an AUC value of 0.8923,while field surveys confirmed that two newly identified landslides lie within high and very high-risk areas,further validating the model’s reliability.This research provides scientific support for infrastructure planning and early landslide hazard warnings in Mile City.

关 键 词：弥勒市 滑坡 全连接神经网络 易发性评价 风险评价 

分 类 号：P694[天文地球—地质学]