摘要
近年来,珠江三角洲磨刀门水道咸潮频发,严重威胁周边地区的供水安全。分别应用随机森林算法(SVM)、支持向量机(SVM)以及Elman神经网络(ENN)建立回归模型,并应用贝叶斯模型平均算法实现咸潮月尺度集成预测。研究结果显示:①随机森林、支持向量机以及Elman神经网络算法在小样本集上表现出不同的不确定性特征;②贝叶斯模型平均能够显著提高模拟精度,纳什效率系数(NSE)达到0.67,相比于3个子模型在测试集上分别提高了22%、24%、33%。
The increase of saltwater intrusion in recent years in Modaomen waterway,Pearl River Delta in China,has threatened the freshwater supply in the surrounding regions.This paper builds the regression model by Random Forest(RF)algorithm,Support Vector Machine(SVM)and Elman Neural Network(ENN),and conducts a monthly integrated forecast through Bayesian Model Averaging(BMA)method.The results indicate that:①RF,SVM and ENN show different extent of uncertainty on small sample sets;②The simulation accuracy of BMA is significantly improved,with NSE of 0.67,which is 22%,24%and 33%higher than that of RF,SVM and ENN,respectively.
作者
卢鹏宇
林凯荣
杨裕桂
袁菲
何用
LU Pengyu;LIN Kairong;YANG Yugui;YUAN Fei;HE Yong(Key Laboratory of Dynamics and Associated Process Regulation of Pearl River Estuary,Ministry of Water Resources,Guangzhou 510611,China;Research Center of Water Resource and Environment,Sun Yat-sen University,Guangzhou 510275,China;Guangdong Key Laboratory of Marine Civil Engineering,Zhuhai 519082,China;Pearl River Water Resources Research Institute,Guangzhou 510611,China)
出处
《人民珠江》
2020年第10期1-5,29,共6页
Pearl River
基金
国家重点研发计划项目(2017YFC0405900)
水利部珠江河口动力学及伴生过程调控重点实验室开放研究基金资助项目(〔2017〕KJ12)。
关键词
月尺度咸潮预测
随机森林
支持向量机
ELMAN神经网络
磨刀门水道
monthly forecast of saltwater intrusion
Random Forest
Support Vector Machine
Elman Neural Network
Modaomen waterway
作者简介
卢鹏宇,女,从事水文水资源研究。E-mail:lupy5@mail2.sysu.edu.cn;通讯作者:林凯荣,男,从事水文水资源研究。E-mail:linkr@mail.sysu.edu.cn。
