基于AE-RCNN的洪水分级智能预报方法研究

Research on flood classified intelligent forecasting method based on AE-RCNN

复杂产汇流特性地区使用洪水分级预报方法可提高预报精度,本文提出一种基于自编码器(Autoencoder,AE)和残差卷积神经网络(Residual Convolutional Neural Network,RCNN)的洪水分级智能预报方法,使用自编码器和K均值聚类算法实现对原始水文数据的特征提取和洪水分级,通过RCNN模型提升卷积神经网络的有效训练深度,以山东省小清河流域黄台桥水文站为例开展洪水分级智能预报研究。结果表明应用降维数据聚类的AE-RCNN模型MAE指标、RMSE指标、NSE指标分别为5.04、7.91、0.92,优于CNN模型、RCNN模型和降雨聚类RCNN模型。该方法能够有效提取水文数据特征、提高洪水预报精度。

Hierarchical flood forecasting method in areas with complex flow generation and confluence characteristics can improve forecast accuracy.This paper proposes a hierarchical intelligent flood forecasting method based on autoencoder(AE)and residual convolutional neural network(RCNN),using autoencoder and K-means clustering algorithm to realize feature extraction and flood classification of hydrological data,using the RCNN model to improve the effective training depth of the convolutional neural network.Taking the Huangtaiqiao Hydrological Station in the Xiaoqing River Basin in Shandong Province as an example,the research on flood classification intelligent forecasting was carried out.The results show that the MAE index,RMSE index,and NSE index of the AE-RCNN model applying the clustering of downscaled data are 5.04,7.91 and 0.92,respectively,which are better than the CNN model,RCNN model,and rainfall clustering RCNN model.This method can effectively extract the characteristics of hydrological data and improve the accuracy of flood forecasting.