基于机器学习的太湖流域多层次防洪调度方案综合评价  

Comprehensive evaluation of multi-level flood control operation schemes in the Taihu Lake Basin based on machine learning

作　　者：高玮志 高华勇 王兆礼[1,2] 赖成光 GAO Weizhi;GAO Huayong;WANG Zhaoli;LAI Chengguang(School of Civil Engineering&Transportation,South China University of Technology,Guangzhou 510641,China;Guangdong Artificial Intelligence and Digital Economy Laboratory(Guangzhou),Guangzhou 510330,China)

机构地区：[1]华南理工大学土木与交通学院,广东广州510641 [2]人工智能与数字经济广东省实验室(广州),广东广州510330

出　　处：《水资源保护》2023年第3期118-125,236,共9页Water Resources Protection

基　　金：国家重点研发计划(2018YFC1508200,2021YFC3001000);国家自然科学基金(51879107);广东省科技计划(2020A0505100009)。

摘　　要：为解决太湖流域多层次防洪调度方案在不同防洪层次目标下的评价问题,构建了流域、区域、城镇多层次防洪排涝调度方案综合评价指标体系,并基于K近邻(KNN)和随机森林(RF)算法构建调度方案综合评价模型。结果表明,联合KNN模型和RF模型实现了KNN-RF组合模型评价,其针对流域、区域与城镇3个层次防洪目标进行调度方案评价的平均相对误差和平均绝对误差分别降低至1.25%、0.82%、2.43%和0.511、0.342、1.380,最大相对误差和最大绝对误差得到改善,等级划分总体正确率高于95%;KNN-RF组合模型能筛选出各层次防洪目标下较优的调度方案,减少单一算法不确定性导致的异常评价误差,评价精度显著提高。In order to solve the evaluation problem of multi-level flood control operation schemes under different flood control level objectives,taking the the Taihu Lake Basin as study area,a comprehensive evaluation index system of multi-level flood control and waterlogging drainage operation schemes for basins,regions and cities was constructed.And a comprehensive evaluation model of operation schemes was constructed based on K-nearest neighbor(KNN)and random forest(RF)algorithms.The results show that the combination evaluation of KNN-RF is achieved by combining KNN model and RF model.The average relative error and average absolute error of KNN-RF combining model in the operation scheme evaluation for the three levels of flood control objectives,namely,basin,region and town,are reduced to 1.25%,0.82%,2.43%and 0.511,0.342,1.38,respectively.The maximum relative error and maximum absolute error have been improved,and the overall accuracy rate of grading is higher than 95%.The KNN-RF combining model can screen out optimal operation schemes for various levels of flood control objectives,reduce abnormal evaluation errors caused by the uncertainty of a single algorithm,and significantly improve the evaluation accuracy.

关 键 词：多层次防洪调度方案 K近邻算法 随机森林算法 KNN-RF组合模型 太湖流域 

分 类 号：TV87[水利工程—水利水电工程]