| Under the background of climate change,sudden hydrological event is a kind of small probability event with strong suddenness,low frequency and wide influence.Under the influence of global climate change,the frequent sudden hydrological events in river basins have a growing trend,which has a great impact on the development of river basin ecology,environment and social economy.In this paper,by analyzing the environmental characteristics of climate change in the Hutou Lake Basin of Hubei Province,and considering the problems of flood prediction and drought identification calculation accuracy,a hydrological model of the Hutou Lake Basin was established.To remedy the problem of insufficient convergence accuracy of the traditional algorithm(SCE,Shuffled Complex Evolution)in calibrating hydrological model parameters,an improved differential mixed complex evolution algorithm(DSCE,Differential shuffled complex evolution)was proposed to calibrate the parameters in the model.Based on the calibration results of the algorithm,the advantages and disadvantages of the model in flood forecasting are analyzed.Combined with hydrological model and machine learning model,it also makes classification prediction of sudden drought in river basin,and achieves the effect of drought early warning and prediction.The main tasks are as follows.(1)Under the background of global climate change,by analyzing the environmental characteristics of water environment,water resources,climate,geography,etc.,and applying the methods of hydrological statistical parameters and trend test,this paper reveals the interannual variation trend of temperature and precipitation in the basin from 1980 to 2016.The results show that the temperature in the studied period has obvious abrupt change characteristics and obvious rising trend,that is,the average temperature,the highest temperature and the lowest temperature increase at the rate of 0.34℃/10 a,0.48℃/10 a and 0.33℃/10 a,respectively,while the precipitation has neither obvious trend change nor obvious abrupt change characteristics.(2)In order to make up for the lack of accuracy in calibrating hydrological model parameters by traditional algorithm,an improved DSCE algorithm is proposed to calibrate hydrological model parameters.Firstly,the parameter calibration process of SCE and the improved DSCE algorithm are compared.Taking the calculation time,calculation times,nash efficiency coefficient and the convergence speed of sensitive parameters in the model as evaluation indexes,it is found that DSCE algorithm is obviously superior to SCE algorithm in all aspects.After selecting DSCE algorithm,the influence of complex number(p value)in the algorithm on parameter calibration process is also discussed.When p = 2 is determined,it can have both calculation efficiency and accuracy.(3)Based on the calibration results of the improved DSCE algorithm,the flood peak error Ep,annual runoff error EQ and Nash coefficient NSE are used as indicators to verify the model.From the perspective of the qualified rate of prediction,the prediction accuracy of the model is Grade A;Judging from nash efficiency coefficient,the prediction accuracy of the model is Class B,which meets the requirements of the code.Furthermore,a total of 51 flood peaks with discharge greater than 1200 m/s were selected for flood analysis,and Pearson-ⅲ curve conforming to flood frequency distribution was used to obtain the runoff of all levels of floods in the basin.On the whole,Xin ’anjiang model can play a very good role in flood peak prediction,and it is very suitable for the Hutou Lake Basin.(4)Based on the current research status of drought indicators,this paper analyzes the applicability and limitations of each indicator,and puts forward a method of drought prediction using machine learning,which directly corresponds hydrological and meteorological data to drought grades.Through correlation analysis,four parameters with strong correlation(namely precipitation,runoff,evaporation and soil water content)are selected as classification features,supervised learning is carried out according to drought grades classified by comprehensive meteorological drought index(MCI),sample data of all grades are balanced,and various common classification methods are used to classify drought grades.Based on four evaluation indexes(namely accuracy,misclassification cost,prediction speed and training time),a fine KNN classification method is selected.On this basis,the time scale of the characteristic parameters was studied,and it was finally determined that the accuracy of classification results was the highest when the time scale was 60 days of rainfall,60 days of evaporation,90 days of runoff and 90 days of soil water content.Finally,the classification results of machine learning are displayed intuitively by using the two-dimensional zoning map,which makes the drought prediction and analysis more intuitive and convenient. |
