地质灾害危险性评价中不同机器学习方法优劣对比:以宁强县大安镇为例  

Comparison of the advantages and disadvantages of different machine learning methods in geohazard risk assessment:Taking Da'an Town,Ningqiang County as an example

作　　者：冯旻譞 毛伊敏 贾俊[1] 齐琦[1] 孟晓捷 刘港 高波[1] 高满新 FENG Minxuan;MAO Yimin;JIA Jun;QI Qi;MENG Xiaojie;LIU Gang;GAO Bo;GAO Manxin(Key Laboratory for Geo-hazards in Loess Area,MNR,Xi’an Center of China Geologic Survey,Xi’an 710054,Shaanxi,China;College of Geological Engineering and Geomatics,Chang’an University,Xi’an 710064,Shaanxi,China;School of Information Engineering,Shaoguan University,Shaoguan 512005,Guangdong,China)

机构地区：[1]自然资源部黄土地质灾害重点实验室/中国地质调查局西安地质调查中心,陕西西安710054 [2]长安大学地质工程与测绘学院,陕西西安710064 [3]韶关学院信息工程学院,广东韶关512005

出　　处：《中国地质》2025年第1期205-214,共10页Geology in China

基　　金：中国地质调查局项目(DD20230436、DD20221739);陕西省卫星应用技术中心项目(SCZK2022–CS–1645/001)联合资助。

摘　　要：【研究目的】地质灾害的孕育和发生受多种因素的影响,具有不确定性和复杂性,给地质灾害的危险性评价带来一定困难。随着AI技术的发展,智能算法能更准确地计算地质灾害孕育与诱发因素之间的多元复杂非线性关系,大大提高了地质灾害危险性模型的准确性,在区域地质灾害危险性评价中逐步得到应用。【研究方法】本文结合宁强县大安镇野外地质调查数据,挑选与地质灾害发生密切相关的12种致灾因子,即高程、坡度、坡高、坡向、坡型、工程地质岩组、断裂距离、水系距离、道路距离、植被覆盖、降雨及地震动峰值等作为危险性分区评价因子。通过构建样本集,运用贝叶斯、随机森林、策略梯度神经网络、KNN和神经网络算法这5种模型进行宁强县大安镇地质灾害危险性建模并进行比较。【研究结果】贝叶斯模型(AUC 0.894)表现最好,绝大多数已发生的地质灾害点位于评价的极高和高危险区,且贝叶斯模型计算结果达到预测精度评价要求。【结论】在宁强县大安镇地质灾害样本数目很少的情况下选择贝叶斯算法模型进行地质灾害危险性评价,是具有可行性的。[Objective]The occurrence of geohazards are influenced by various factors,which have uncertainty and complexity,making it difficult to assess the risk of geohazards.With the development of AI technology,intelligent algorithms can more accurately calculate the complex and nonlinear relationships between geohazard triggering indexes,greatly improving the accuracy of geological hazard risk assessment models.[Methods]Based on the field geological survey data of Da'an Town,Ningqiang County,12 indexes closely related to the occurrence of geohazards were selected,namely elevation,slope,slope height,slope direction,slope type,engineering geological rock formations,fault distance,water system distance,road distance,vegetation coverage,rainfall,and seismic ground motion,as risk zoning evaluation factors.By constructing a sample set,Bayesian,strategy gradient neural network,random forest,KNN and neural network algorithm are used to model and compare the geohazard risk assessment result in Da'an Town,Ningqiang County.[Results]The experimental results show that the Bayesian model(AUC 0.894)performs the best,with the vast majority of geohazards located in the extremely high and high-risk evaluated areas,and meets the requirements for prediction accuracy evaluation.[Conclusions]It is feasible to choose Bayesian algorithm models for geological hazard risk assessment when the number of geohazard samples is small.

关 键 词：地质灾害 危险性评价 机器学习 灾害地质调查工程 大安镇 宁强县 陕西 

分 类 号：P694[天文地球—地质学]