Study On Flood Forecasting In Ankang Section Of Hanjiang River Basin Based On Big Data Analysis Method

With the rapid development of computer technology,the application of big data has attracted more and more attention.Based on the original calculation method and mathematical model,It is a new technology and method that can be continuously improved and updated.This method can change the problem of cumbersome data processing and basic parameter calibration in traditional hydrological methods.At the same time,artificial intelligence and other technical means are used to make the calculation speed faster,and the results obtained are closer to the measured values.It is of great benefit to guide practical work.China is a flood-prone country.In order to reduce disaster losses,it is particularly important to forecast flood effectively and accurately.Flood forecasting is a quantitative and timely scientific forecasting of the occurrence and change of floods based on the hydrometeorological factors that have appeared in the earlier stage and at present.Flood forecasting is usually based on the rainfall-runoff relationship or the corresponding relationship between water level and discharge at upper and lower stations.Its forecast period is generally not long,but its accuracy is relatively high.Therefore,in hydrological work,flood forecasting is mainly combined with rainfall to forecast runoff.The Hanjiang River Basin belongs to the subtropical monsoon climate zone,and the precipitation distribution is very uneven during the year.Summer and autumn are the flood season.The precipitation can account for 80% of the total annual precipitation,especially from June to September,which accounts for about 60% of the total annual precipitation.Especially in June-September,the precipitation is the largest,accounting for about 60% of the annual precipitation.The runoff in flood season has the characteristics of double peaks.The floods in the Hanjiang River Basin caused by heavy rains are mainly related to the activities of monsoon.Floods occur every year from May to September,especially in July and August.Ankang belongs to the upper reaches of the Hanjiang River.Because of the particularity of geographical environment and climatic conditions,rainstorm centers are often concentrated inAnkang section,which has the characteristics of "ten years and nine floods".Flood disaster has become one of the biggest natural disasters in Ankang area.Engineering measures are the basis of flood control,while non-engineering measures are the guarantee of safe flood season.Therefore,in-depth study on flood forecasting in Ankang section of Hanjiang River Basin is of great significance for accurate forecasting of flood peak,flood volume and flood process.Taking Ankang section of Hanjiang River Basin as the research background,this paper collects and collates the basic hydrological data in this area.According to the actual situation of the data,the calculation period is divided into three periods:1991-2005,2006-2012 and 2013-2017.Tyson polygon method was used to collate daily rainfall data.The natural inflow runoff is calculated by sending outflow runoff data.For the trend analysis of time series,the Mann-Kendall method of nonparametric test is used.The abrupt change test of annual rainfall in Ankang section of Hanjiang River Basin is carried out by programming calculation based on R language.Calculate the statistical value of M-K catastrophe analysis of annual average inflow in Ankang section.Statistical analysis shows that the rainfall runoff of Ankang Reservoir has a clear linear relationship.It shows that the runoff of Ankang Reservoir originates from rainfall,with large annual rainfall and large annual runoff coefficient.That is to say,the flood process of Ankang Reservoir is closely related to precipitation.In this paper,the Long-Short Term Memory(LSTM)model based on deep learning is selected to simulate the daily runoff process in Ankang section of the Hanjiang River Basin.The inflow runoff and rainfall of Ankang Reservoir belong to non-stationary time series and belong to random events,but in the continuous time series,certain regularity will still be found.This paper is based on the 27-year hydrological data from 1991 to 2017.Because the inflow runoff is deduced from the water level,data loss or abnormal data value often occur in the collected data.In order to improve the accuracy of prediction,Pandas is used to process data,including eliminating abnormal data and filling blank data.LSTM algorithm based on deep learning is introduced into flood forecasting.By compiling Python language and adjusting parameters many times.After thesuccessful construction of the calculation model,the runoff into the reservoir was simulated using the 27-year hydrological data of Ankang section.Seven representative years were selected for daily flood simulation and 10 floods were further studied.Daily data are used to simulate the runoff of representative years,and time-interval data are selected to simulate the runoff of field floods.Compared with the measured runoff,the accuracy of the results obtained by the two simulations is higher.In order to verify the results of simulation calculation using LSTM model,as a comparison,the Xin'anjiang model is used to simulate the research area.Xin'anjiang model is a well-known and widely accepted hydrological model in China.It has good application effect in wet season and wet area in semi-humid area.From the point of view of application conditions,the research area of this paper is applicable.In this paper,the Xin'anjiang model is used to simulate the daily runoff in Ankang section of the Hanjiang River Basin.The parameters of Xin'anjiang model were calibrated and simulated by C language program based on the collected data.The hydrological data from 1991 to 1995 were used for parameter calibration,and the flood process and peak discharge were simulated from 1996 to 2017.After comparing the results with the measured data,it is found that the simulation results of Xin'anjiang model and the measured runoff are not satisfactory.It can be seen that LSTM model method has obvious advantages in the research area.The accuracy of hydrological forecasting is improved,the forecasting time is shortened,and the research of simulation calculation goes deep into the stage of flood events.Problems were found in the study: it was difficult to obtain hydrological data.In the calculation of on-site floods,there are very few data to meet the requirements.However,there are few data available for training in simulation calculation,which affects the learning effect.With the development of big data,it will promote the acquisition of information more convenient,and will greatly improve the accuracy of hydrological forecasting results.