基于随机森林的大坝潜在风险预测方法  

Prediction method of dam potential risk based on random forest

作　　者：丁炜 徐毅[2] 金有杰[1] 张日 陈建宁 DING Wei;XU Yi;JIN Youjie;ZHANG Ri;CHEN Jianning(Nanjing Research Institute of Hydrology and Water Conservation Automation,Ministry of Water Resources,Nanjing 210012,China;Jingjiang Water Resources Bureau Planning and Construction Division,Jingjiang 214500,China)

机构地区：[1]水利部南京水利水文自动化研究所,江苏南京210012 [2]靖江市水利局规划建设科,江苏靖江214500

出　　处：《水利信息化》2023年第1期46-50,共5页Water Resources Informatization

基　　金：水利部南京水利水文自动化研究所揭榜攻关项目(NSZS062201);江苏南水科技有限公司自立项目(NSZS0818001)。

摘　　要：大坝潜在风险预测在降低大坝溃坝概率和减少水库大坝失事方面发挥着重要作用。利用随机森林方法构建基于数据驱动的大坝潜在风险预测模型,可以减少建模过程中的人工干预,实现风险预测高效化、智能化。首先预处理大坝基础数据资料,构建训练和测试数据集,然后构建大坝风险预测模型并用训练数据训练模型,利用网格搜索和交叉验证确定模型最优参数,最后通过模型评价指标和多种算法对比结果全面评估模型性能。实验结果表明:基于随机森林的风险预测模型在测试数据上的准确率为90.54%,相较于ANN (人工神经网络),KNN (最邻近),SVM (支持向量机)算法分别高出4.87%,18.59%,37.93%,满足实际应用的需求。Dam potential risk prediction plays an important role in reducing dam failure probability and reservoir dam failure events. In this paper,a data-driven model for potential risk prediction of reservoir dams is proposed to reduce manual intervention and realize efficient and intelligent risk prediction based on random forest algorithm. Firstly,the basic data of reservoir dams are preprocessed to develop training and test datasets. Then,the dam risk prediction model is set up and the training data is used to train the model. Grid search and cross validation are applied to determine the optimal parameters. We can comprehensively evaluate model performance through model evaluation indicators and comparison results of multiple algorithms. The experimental results show that the accuracy of the random forest-based risk prediction model on the test data is 90.54%,and the model accuracy are 4.87%,18.59%,and 37.93% higher than that of ANN,KNN,and SVM algorithms,respectively. The model can meet the demands of practical applications.

关 键 词：大坝 风险预测 随机森林 网格搜索 交叉验证 

分 类 号：TV698.1[水利工程—水利水电工程]