The use of advanced computational methods in reliable river flow prediction

Abstract

Research on the Yangtze and Narmada Rivers has focused on predicting river flow and preventing
flooding.The minimum prediction period of numerical computing methods is about a month ago,which
is also achieved because it is calculated too short for use in hydrological applications by combining a
deep neural network - empirical mode degradation (EMD) algorithm and a encoder-decoder Long-Short-
Term Memory (En-de-LSTM) architecture to predict river flow in a study conducted on the Yangtze
River. Monthly flow data from the Hankou hydrological station on the Yangtze River (January 1952 to
December 2008) was selected to teach the model, and results were obtained for two years and ten years
of continuous predictions of floods. To achieve a reliable and accurate result,modern programs
CatBoost,LGBM, Random Forest and XGBoost are used.When checking the consistency of the
calculation results, the mean quadrad error (MSE),mean absolute error(MAE),root mean quadrad
error(RMSE),root mean quadrad percentage error(RMSPE), normalized root mean quadrad
error(NRMSE), and quadrad error (R2) indicators are used. Among the above models, Random Forest
has shown to be able to produce effective results even in complex conditions of hydrological prediction.