Research On Medium And Long Term Runoff Probability Prediction Method Coupled With CFS Ensemble Forecast And Deep Learning

High-precision runoff forecast is the key to realize optimal operation management of reservoir group.However,due to the restriction of weather forecast technology,it is difficult for the runoff forecast method based on the mechanism of runoff production and confluence to realize the medium and long term runoff forecast.In recent years,with the development of artificial intelligence technology and deep learning theory,data-driven runoff prediction methods have attracted wide attention.The data-driven forecasting method pays attention to the internal relationship of the long-term runoff series and makes predictions by mining the hidden laws of the data in depth with the method of mathematical statistics.This kind of method is simple in modeling and has good adaptability and prediction performance.In this paper,the research on medium and long term runoff forecast model based on deep learning is carried out.By analyzing the correlation between meteorological factors and runoff,the maximum correlation and minimum redundancy criteria are used to screen key runoff and meteorological factors,and the medium and long term runoff forecast model coupled with CFS ensemble forecast information correction is constructed.Taking the inflow runoff of the Three Gorges Reservoir as the research object,the simulation results verify the high prediction accuracy and reliability of the proposed model.The main research work and achievements of this paper are as follows:(1)Study on the runoff forecast method based on data driven process of runoff formation of explanatory is not strong,get CFS global weather forecast from the NCEP data and parse,by using inverse distance weighted interpolation for study area meteorological forecast results,the maximum coefficient of introduction of a maximum minimum redundancy feature selection method,combined with the first-order incremental selection,candidate factors are introduced one by one,and the optimal feature subset is selected according to the model performance evaluation,which provides data support for the medium and long term runoff prediction model.(2)Under the background of changing climate environment and underlying surface conditions,extreme gradient boosting deep learning model is introduced into the medium and long term runoff forecast.Built the three gorges reservoir on the runoff forecast model and compared with the commonly used machine learning and deep learning model analysis,case study results show that the built model has higher prediction accuracy.At the same time,by analyzing the characteristics of meteorological forecast information,a ridge regression method for the correction of runoff forecast results are proposed.The results show that the corrected forecast with coupled meteorological forecast information has higher prediction performance and better applicability in the medium and long term runoff forecast under climate change.(3)Based on the idea of combination prediction,the mixed probability prediction model of limit gradient lifting tree and Gaussian Process Regression combination is constructed.Compared with the single prediction model,the hybrid probability prediction model has better point prediction performance and can provide the probability interval under certain confidence level.Compared with the common probability prediction model,the results show that the prediction accuracy of this model is higher and the probability prediction results are more reliable.The research results can provide an effective method for medium and long term runoff probability prediction of reservoirs.