Research On Flood Forecast Of Huanren Basin Based On Machine Learning

Flood disasters cause heavy losses to people's lives and property every year,and become one of the main factors restricting the country's economic construction and sustainable development.my country has a vast territory,numerous rivers,complex topography,and remarkable monsoon climate characteristics.The annual precipitation is very concentrated.The task of flood prevention and disaster reduction is arduous.As one of the important nonengineering measures for flood control and disaster reduction,flood forecasting plays a very critical supporting role in flood control and disaster reduction in river basins.Constructing a flood forecast model with high forecast accuracy,strong applicability,simple calculation and easy promotion is of great significance for improving the effectiveness of flood warning,emergency management and reducing the loss of people's lives and property.Based on the Huanren River Basin in my country,based on long and short-term memory artificial neural networks,deep learning stacking architecture and data integration technology,this paper constructs LSTM flood forecasting models for different forecast periods under different input factors,and compares the output and output of orders under different input factors.The prediction effect of the multi-output LSTM model.The main research contents and results are as follows:(1)In order to analyze the impact of data integration on the accuracy of the LSTM model,this paper constructs a single-output LSTM flood forecast model with only rainfall input and a single-output LSTM flood forecast model based on data integration,and compares the forecasting effects of the two with the Xin'anjiang model..First,by comparing the overall Nash coefficients and correlation coefficients of the three models,it is found that the forecast effect of the single-output LSTM model when only rainfall is input is worse than that of the Xin'anjiang model.After data integration,the forecast effect of the LSTM model has been greatly improved,and It is better than the Xin'anjiang model;secondly,comparing flood forecasting peak,peak present time and NSE,it is found that the LSTM model has low flood peak forecasting,but the single-output LSTM model forecast accuracy after data integration can reach the Class A standard.(2)In order to analyze the difference between a single-output LSTM model and a multioutput LSTM,this paper constructs an LSTM model that only inputs rainfall under different forecast periods and an LSTM model based on data integration.By comparing the single-output and multi-output LSTM models under the same input conditions,it is found that the multioutput model is better than the single-output model when only rainfall is input.After data integration,the continuity of the flow data in the model input factors is destroyed.The singleoutput model is better than the multi-output model;under the same forecast period,the singleoutput or multi-output LSTM models after data integration perform better.(3)By analyzing the forecast effect of the multi-output LSTM model in each period of the forecast period under different input factors,it is found that when only rainfall is input,the model has the best forecast effect in the last period of the forecast period and is within the available range of rainfall forecast information.As the forecast period extends,the forecast accuracy becomes higher.This is because under the premise that the continuity of the input factors of the model is not destroyed,the longer the input factor sequence,the better the model learning effect;The forecast flow correction in the first time period is the most obvious.The overall Nash coefficient of the first time period under various forecast periods is higher than0.89,and the correction effect on subsequent periods decreases sequentially.