Hydrological Integrated Forecasting Simulation Research

As an important non-engineering measure,hydrological forecasting plays a key role in flood prevention and disaster reduction,and hydrological model is the most powerful tool for hydrological forecasting.However,the hydrological cycle is a relatively complex system,which is affected by many factors.Due to the constraints of conditions,some existing hydrological forecasting tools cannot meet the requirements of better accuracy,which limits the development and application of hydrological forecasting models to a certain extent.Therefore,it is more and more important to introduce a new integrated hydrological forecast model to improve the forecast accuracy and applicability of the hydrological model.Taking Yandu River Basin and Yujia River Basin as the research area,this paper established BP neural network model,Xin'an River model and SVR model respectively,and integrated the above models,and compared the simulation results of the integrated model and the single model.The main research contents and results include:(1)The optimization method of BP neural network predictor.A method for optimizing hydrological characteristics is proposed.Under different characteristic engineering conditions,a hydrological forecast model is established based on BP neural network.The correlation analysis method is used to select rainfall characteristics that are strongly correlated with the outlet flow of the basin,and based on the selected characteristics Formulate 4 different schemes to compare and analyze the forecast effects of models and types under different characteristic engineering conditions.And this method is used in BP neural network model and SVR model.The results show that the adopted feature selection method can make full use of rainfall data and can effectively improve the stability and accuracy of model forecast results;(2)Application of Xin'anjiang Model.Based on the existing hydrological data and DEM data,the Xin'anjiang model was applied in the two watersheds,and the model parameters were calibrated based on the data of the two watersheds from 1981 to 1985.A total of 10 floods from 1986 to 1987 were used to verify the model.The results showed that among the 21 floods in the Yandu River Basin,the average certainty coefficient was 0.85 and the pass rate was 57.1%.Among the 21 floods in the Yujia River Basin,the average certainty coefficient was 0.77 and the pass rate was19%.The simulation results of some floods were not The ideal may be caused by the inaccurate calculation of the soil water content in the previous period,the existence of base flow at the bottom of the river and the errors in the collected hydrological data itself;(3)Multi-objective optimization of predictor,kernel function and parameters of SVR model.Under different kernel functions,the grid search method is used to determine the optimal parameters of the SVR model and the optimal kernel function.According to the proposed hydrological feature optimization method,the model adopts 4 different forecasting schemes and constructs the SVR model under different forecasting influence factors.The results show that when the model uses the Gaussian(Rbf)kernel function,the simulation effect is better than other kernel functions;adding the previous cumulative rainfall to the model input data can improve the model's forecast accuracy and stability to a certain extent.which also proves the effectiveness of the proposed hydrological feature optimization method;(4)Model integration and application.Based on the above three models,the Stacking combination strategy in the machine learning integration algorithm is adopted,and the multiple linear regression algorithm is used as the secondary model to integrate the prediction results of the above three basic models.Based on considering or disregarding the previous runoff data,construct two different programs(not considered in Option 1,and considered in Option 2).The ensemble model was trained using the 5-fold cross-validation method,and the trained ensemble model was used to forecast the data of the two river basins in 1987.The results show that the reliability of the forecast results of the integrated model is improved compared with the single model;in scheme 1,the reliability of the forecast results of the integrated model is significantly improved.In scheme 2,the previous period is considered Flow data,so that a single model has a higher accuracy,therefore,the reliability of the integrated model is not significantly improved;(5)Software development.Based on the above model theory,an integrated forecasting system was developed using the Python language and corresponding development framework.The entire system realized the functions of data processing,model training,single-model forecasting and multi-model comprehensive forecasting.