基于机器学习算法的滑坡土壤含水率预测方法研究  

Research on Prediction Method of Landslide Soil Moisture Content Based on Machine Learning Algorithm

作　　者：杨小平[1,2] 段生锐 蒋力 刘光辉 YANG Xiao-ping;DUAN Sheng-rui;JIANG Li;LIU Guang-hui(School of Information Science and Engineering,Guilin University of Technology,Guilin 541004,China;Guangxi Key Laboratory of Embedded Technology and Intelligent System,Guilin University of Technology,Guilin 541004,China;Guangxi Zhuang Autonomous Region Geological Environment Monitoring Station,Nanning 530029,China;Guilin Saipu Electronic Technology Co.,Ltd.,Guilin 541004,China)

机构地区：[1]桂林理工大学信息科学与工程学院,广西桂林541004 [2]桂林理工大学广西嵌入式技术与智能系统重点实验室,广西桂林541004 [3]广西壮族自治区地质环境监测站,广西南宁530029 [4]桂林赛普电子科技有限公司,广西桂林541004

出　　处：《水电能源科学》2024年第3期73-77,共5页Water Resources and Power

基　　金：国家高新技术研发计划(863计划)(2013AA12210504);广西壮族自治区科技攻关项目(AC1638012);广西壮族自治区南宁市青秀区科技局科技计划(RZ19100041)。

摘　　要：土壤含水率是影响坡体稳定性的决定因素之一。针对滑坡体内部土壤水分信息难以准确感知的问题,建立了一种基于机器学习算法树突神经网络的土壤含水率预测模型(DDNN),通过分析土壤水分垂向变化特征和数据相关性确定关键的影响因子后,将水分预测模型DDNN与GA-BP、RF、RBFNN三种算法进行对比试验。发现DDNN预测模型的拟合优度R2最高为0.998,均方根误差和平均绝对误差均最小,分别为0.091、0.059,其预测精度明显高于其他三种算法。并采用关系谱探究了相关影响因素对土壤含水率的敏感程度。结果表明,敏感度由高到低依次为气温、降水、初始水分、风速、地温,研究结果可为滑坡体稳定性分析提供技术方法支撑。Soil moisture content is one of the decisive factors affecting slope stability.It is difficult to accurately perceive the soil moisture information inside the landslide.A soil moisture content prediction model(DDNN)based on machine learning algorithm dendritic neural network was established.After determining the key influencing factors by analyzing the vertical variation characteristics of soil moisture and data correlation,the water prediction model was compared with GA-BP,RF and RBFNN algorithms.The results show that the goodness of fit R2 of the DDNN prediction model was 0.998,and the root mean square error and mean absolute error were the smallest,which were 0.091 and 0.059,respectively.The prediction accuracy was significantly higher than the other three algorithms.The relationship spectrum was used to explore the sensitivity of related influencing factors to soil moisture content.The results show that the sensitivity from high to low is temperature,precipitation,initial moisture,wind speed and ground temperature.The research results can provide technical support for the stability analysis of landslide.

关 键 词：机器学习算法 树突神经网络 滑坡体 土壤含水率预测 相关性 敏感性 

分 类 号：P642.22[天文地球—工程地质学]