Application Analysis Of Machine Learning In Huai River Water Level

The Huaihe River Basin spans the three provinces of Henan,Anhui,and Jiangsu,and is mostly a plain area.Therefore,the area of cultivated land in this region is extensive,and the annual grain output plays an important role in our country;and as part of the north-south dividing line in my country,its importance is self-evident.However,based on historical experience,the Huaihe River has frequent flood disasters and poor environmental carrying capacity.As one of the important preventive measures for flood disasters,water level prediction can effectively promote the construction of water conservancy safety in the Huaihe River Basin.Water level prediction is one of the long-term concerns of the academic community,and the development of its theory and methods has been iteratively upgraded.Traditional forecasting methods such as grey system theory and multi-layer hierarchical model have been gradually updated and replaced under the background of data mining and artificial intelligence.According to the research of many scholars,data-driven big data analysis methods can effectively improve the accuracy of forecasting..This paper sorts out the methods used by scholars to predict the water level in the past,and reviews the factors that affect the water level,and organizes and forms a better data set and a better model for water level prediction.First,in order to explore how to improve the accuracy of water level prediction results,two data sets are used for prediction,one is to use a single variable(water level)to predict a single variable(water level),and the other is to use multiple variables(water level,rainfall,flow)to predict a single variable(water level).Secondly,in terms of data processing,interpolation and generative adversarial neural network(GAN)are used for filling.When the amount of data is larger,the filling effect of GAN is better.Thirdly,in the selection of the single-step prediction method,singular spectrum analysis(SSA)is used to decompose the data,and the residual term after data decomposition is decomposed twice using the fully integrated empirical mode decomposition(CEEMDAN),and then decomposed.The results are put into Long Short Term Memory(LSTM),Gradient Boosting Regression(GBR),Extreme Gradient Boosting(XGBoost),Lightweight Gradient Booster(Light GBM)prediction models,and Long Short Term Memory(LSTM)and Temporal Convolutional Networks(TCN)Finally,the grid search method is used to adjust the parameters of the machine learning model,and the genetic algorithm(GA)or the gradient descent method(GD)is used to solve the optimal solution of the parameters of the neural network model.Finally,try multi-step water level forecasting,plan ahead,predict risks,and help plan rationally.The research results show that the single-step prediction effect of the SSA-CEEMDAN-LSTM-TCN model is the best under the multivariate input and univariate output datasets(water level,flow,rainfall-water level),indicating that more effective influencing factors are added to the prediction.Make the model prediction effect better.When Autoformer predicts water level in multi-step prediction,the longer the prediction time,the better the effect.