Study On The Ice Regime And Ice Prediction Model Of Freeze-up And Break-up Date In Inner Mongolia Reach Of The Yellow River

Ice jam is a common hydrological phenomenon in rivers in alpine regions.In China,ice jam mainly occurs in the Yellow River Basin and Songhua River Basin.Due to the special geographical location,river characteristics and hydrometeorology,the Inner Mongolia reach of the Yellow River has different degrees of ice flood almost every year.The loss caused by ice flood disasters is serious,and it is difficult to prevent and control.Therefore,it is of great significance to study the characteristics and changes of ice conditions and to do a good job in ice forecasting for the prevention and control of ice flood disasters.Firstly,based on the historical hydrological,meteorological and ice data of the Inner Mongolia reach of the Yellow River from 1958 to 2021,this thesis analyzes the ice characteristics of the reach,and analyzes the dynamic factors,thermal factors,river factors and human factors affecting the ice conditions of the reach.Secondly,the prediction factors are selected by rescaled range analysis,gray correlation method and recursive feature elimination method.The multi-layer perceptron(MLP),long short-term memory network(LSTM)and gated recurrent unit neural network(GRU)are used to predict the freeze-up date and brrak-up date of Bayangaole,Sanhuhekou and Toudaoguai hydrological stations.Finally,the prediction results are compared and analyzed with the error days and accuracy evaluation indexes.The main research results are as follows :(1)From 1958 to 2021,the date of ice-run and freeze-up was postponed 25 d,the date of break-up was advanced 16 d.Compared with 1958-1968,the average number of ice days in 2014-2021 decreased by 5 days,the number of frozen days decreased by 15 days,the river closure increased by 212m~3/s,and the peak flow increased by 96m~3/s.(2)Affected by the upstream reservoir,the upstream water of the Inner Mongolia reach of the Yellow River increased,the water level rose obviously during the freezing period,and the water storage capacity of the channel increased by 74 million m~3.(3)The prediction of MLP,LSTM and GRU for the freeze-up date and break-up date are the first-class scheme.The prediction accuracy of the gated recurrent unit neural network prediction model is higher than that of the other two models,which is the optimal model.(4)When using the gated recurrent unit neural network model to predict,the maximum error days of the freeze-up date are 3 days,and the maximum error days of the break-up date are 2 days.