Research And Application Of Runoff Time Series Prediction Based On Ensemble Learning

The runoff time-series is a representative time-series.There have been many precedents for correlation analysis and research,and good results have been obtained.However,the time-series of runoff is affected by a variety of complex factors,and traditional methods are not enough to analyze the rich information contained in it,and it is also impossible to make efficient and accurate predictions.However,runoff has a significant impact on production and life.Therefore,in order to improve the accuracy of runoff prediction,it is urgent to introduce new prediction methods,improve the prediction process,and improve the prediction capability of the model.In this paper,two representative stations A and B located in the upstream and downstream of a certain river reach are selected as representatives.It uses traditional machine learning and ensemble learning as the research theme,and collects related annual and monthly runoffs.After the data obtained,the research questions on runoff prediction are progressively step by step,and the exploration of model application,the improvement of model forecasting accuracy,and the improvement of forecasting models were realized.The main research contents and innovations of this article are as follows:Use traditional machine learning methods for runoff prediction.First,use traditional machine learning basic methods,and use ARMA,KNN,ridge regression,SVR,BP,and decision tree models to predict annual runoff,monthly runoff,and flood season runoff respectively;then combined with relevant evaluation indicators,evaluate the prediction results;and finally Analyze and compare the advantages and disadvantages of the results of each model,and analyze the limitations of traditional machine learning models in predicting runoff.The ensemble learning method is used for runoff prediction.First of all,a systematic description of various ensemble learning methods,based on the respective characteristics of annual runoff,monthly runoff,and flood season runoff data,combined with theory to explore data processing and division;then various ensemble learning methods,such as random forest,Ada Boost,GBDT and XGBoost are used to predict the value of annual runoff,monthly runoff,and flood season runoff,and evaluate the results with indicators;finally,analyze the prediction results of related models,discuss the advantages and disadvantages of each model prediction in the ensemble learning,and compare the traditional learning method points out that there are improvements,and the experimental results prove that the ensemble learning method has a large degree of improvement in runoff prediction compared with the traditional learning.Use ensemble learning for runoff interval prediction.For the runoff time-series,it has a wide range of fluctuations,is affected by a variety of complex factors,but has good periodic characteristics.Combining actual production needs,interval forecasts can provide a certain range of fluctuations.For runoff series with changes in flood and dry seasons,it has high research value.Firstly,the various interval forecasting methods used in this article are introduced.There are three areas: interval fluctuation before prediction,interval fluctuation after prediction,and fine-grained indirect interval prediction;then monthly runoff is taken as an example to predict the interval,and finally the forecast is combined with evaluation indicators.The results are analyzed.Experiments show that the interval fluctuation method after prediction can achieve better results,and the finegrained indirect prediction model can obtain interval prediction results that are more suitable for real production.The three methods provide a new direction for runoff research through interval prediction examples.