Research On Multi-model Hydrological Runoff Prediction Method Based On LSTM

The Long Short-Term Memory network is suitable for dealing with the problem of long sequence dependence,and has a good ability to deal with and simulate hydrological runoff,which is affected by meteorological and other factors at long intervals.In this study,four models of Jinghe River Basin are established based on LSTM,and the monthly average runoff is predicted by direct prediction model,difference prediction model,average prediction model and composite calibration model.The composite calibration model and average model also have a good ability to simulate and predict on the Xin'an River,proving the effectiveness,availability and generalizability of the model.Under the condition of mastering the accurate hydrological data,the direct prediction model has a good prediction effect on the daily average runoff,but the accuracy of the monthly average runoff prediction will decrease slightly due to the accumulation of daily errors.The difference prediction monthly model eliminates the error accumulation in the direct prediction monthly model and improves the prediction accuracy of the monthly average runoff,but still depends on the hydrological data in the pre-sequence time period.In the absence of mastering the recent daily hydrological data,the average model is particularly practical and important.The prediction of the model completely abandons the dependence on the pre-sequence hydrological data,and can independently predict the daily average runoff and the monthly average runoff.The composite calibration model is based on the complementary performance advantages of the direct prediction model and the difference prediction model in the flat and peak periods.The results of the two models are calibrated twice to improve the accuracy and stability of prediction.This study provides different types of models for predicting the monthly average runoff in a watershed.It is worth pointing out that the average model abandons the dependence on the hydrological data in the recent pre-period time period,which proves that the near-term pre-hydrological data is not necessary in the hydrological forecast simulation.