基于改进非线性自回归网络的洪水预测算法  

Flood forecasting algorithm based on improved nonlinear auto-regressive network

作　　者：崔雅博 罗清元 刘丽娜[3] CUI Ya-bo;LUO Qing-yuan;LIU Li-na(School of Information Engineering,Kaifeng University,Kaifeng 475004,China;Division of Station Network Monitoring,Hydrology and Water Resources Bureau of Henan Province,Zhengzhou 450003,China;Supercomputing Center of Henan,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]开封大学信息工程学院,河南开封475004 [2]河南省水文水资源局站网监测处,郑州450003 [3]郑州大学河南省超算中心,郑州450001

出　　处：《沈阳工业大学学报》2023年第1期84-89,共6页Journal of Shenyang University of Technology

基　　金：河南省高等学校重点科研项目(22B520020);开封大学青年创新人才培育基金项目(KDQN-2020-GK007)。

摘　　要：针对流域的洪水预测具有高度非线性和随机性的问题,提出了一种混合预测模型用于流域的洪水预测.该模型是一个集成了数据预处理模块的具有外部输入的非线性自回归神经网络,采用小波变换进行时间序列分解,利用多基因遗传编程进行细节缩放,以提高时域和频域特性的提取能力,进一步捕获时间序列的非平稳性,与NARX结合可以大幅提高洪水预测的准确性,利用栾川水文站15年中所测水文数据对所提模型进行验证和测试.实验结果表明,相比较于传统算法和其他预测算法,所提出的算法具有更高的预测准确度和性能,可广泛应用在洪水预测等领域.Aiming at the high non-linearity and randomness of flood forecasting,a mixed forecasting model was proposed.Integrated with data preprocessing module,it was a nonlinear auto-regressive model with exogenous inputs(NARX).Wavelet transform(WT)was used to decompose the time series,and multi-gene genetic programming(MGGP)was used to scale the details to improve the extraction ability of time domain and frequency domain characteristics and to further capture the non-stationarity of the time series.NARX was incorporated to greatly improve the accuracy of flood forecasting.In addition,the hydrological data measured in Luanchuan hydrological station in the past 15 years were used to verify and test the as-proposed model.Experimental results show that the as-proposed algorithm has higher prediction accuracy and performance compared with the traditional and other prediction algorithms,and can be widely applied in flood forecasting and other fields.

关 键 词：洪水预测 非线性自回归网络 混合预测模型 小波变换 多基因遗传编程 数据预处理 机器学习 神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]