基于互信息及人工神经网络的降雨-径流预报方法
Rainfall-runoff Forecasting Based on Mutual Information and Artificial Neural Network
王加红 1陆颖 2袁旭 3袁嫄 4张丽梅 4张珂瑶 5郭子璞 5晏翠玲5
作者信息
- 1. 华能澜沧江水电股份有限公司小湾水电厂,云南 大理 675702;云南大学国际河流与生态安全研究院,云南 昆明 650500
- 2. 云南大学国际河流与生态安全研究院,云南 昆明 650500;云南省国际河流与跨境生态安全重点实验室,云南 昆明 650500
- 3. 华能澜沧江水电股份有限公司,云南 昆明 650214
- 4. 中国电建集团昆明勘测设计研究院有限公司,云南 昆明 650033
- 5. 云南大学国际河流与生态安全研究院,云南 昆明 650500
- 折叠
摘要
输入变量空间信息的提取、筛选与输入优化是提高数据驱动水文模型性能的关键环节.为提高数据驱动短期径流预报模型效果,以全国第二届水科学数值模拟创新大赛题目为算例,基于互信息及人工神经网络方法构建降雨-径流水文模型,进行小时尺度径流预报,探讨互信息对人工神经网络模型在降雨-径流预报精度提高上的作用.结果表明,基于互信息及人工神经网络的降雨-径流预报模型模拟精度高,适用性较好,验证期纳什效率系数和相关系数分别为0.94、0.96.互信息方法能够实现径流预报模型的输入优化,避免数据冗余,可为空间信息缺乏流域径流预报因子选择提供新思路.
Abstract
The spatial information extraction,screening and input optimization of input variables are the key steps to improve the performance of data-driven hydrological models.In order to improve the effect of data-driven short-term run-off forecasting model,the second National Water Science Numerical Simulation Innovation Competition is taken as an ex-ample.The mutual information and artificial neural network are used to construct a rainfall-runoff model to forecast hour-ly scale runoff.Furthermore,we explored the effect of mutual information on the improvement of rainfall-runoff predic-tion accuracy by artificial neural networks model.The results show that the rainfall-runoff forecasting model based on mutual information and artificial neural network has high simulation accuracy and good applicability,the Nash-Sutcliffe efficiency coefficient is 0.94 and the correlation coefficient square is 0.96 in the validation period.The mutual information method can optimize the input of runoff forecasting model,avoid data redundancy,which can provide a new idea for the selection of runoff forecasting factors in the basin lacking spatial information.
关键词
数值模拟/径流预报/互信息/数据驱动模型/人工神经网络Key words
numerical simulation/runoff forecast/mutual information/data-driven model/artificial neural network引用本文复制引用
基金项目
国家自然科学基金项目(32060831)
华能澜沧江水电股份公司科研业务(HY2020/S9)
云南省万人计划项目(QNBJ2018166)
出版年
2024
NETL
NSTL
万方数据