Prediction Of Runoff Series In Jiulong River Basin Based On LSTM Model

Water resource is an indispensable element of human beings.Accurate runoff prediction plays a decisive role in water resources scheduling and environmental quality inspection management.As a time series,runoff elements are limited and influenced by natural factors and human factors in the formation process,showing strong nonlinear relations and chaotic characteristics.Therefore,how to effectively predict runoff sequences is the focus and difficulty of research in the field of hydrology.The lack of runoff-related data in the Jiulong River Basin site has resulted in a lack of continuity of data,which cannot meet the needs of runoff time series research,and has had a tremendous impact on the subsequent water quality environment research.Therefore,the study of runoff sequence prediction by the model is also indispensable.In order to effectively predict the runoff of the stations in the basin and improve the accuracy of runoff prediction,this paper focuses on the study of the Jiulong River Basin Water Environment Simulation Research by the Institute of Urban Environment of the Chinese Academy of Sciences.In 2018,a total of 10 years of runoff is the research object,and LSTM,GAN and other deep learning related technologies are used to establish a runoff sequence prediction model and a data simulation model.The main content of this article is divided into the following:(1)This paper analyzes the characteristics of the inner diameter of the basin,the meteorological characteristics,and the model expression between meteorology and runoff,and conducts basic research for subsequent feature selection and modeling;(2)Based on the analysis of LSTM model and BP model,based on the autocorrelation of runoff sequence and partial meteorological data,the generalization ability of BP network and the time series processing ability of LSTM network are fully utilized,and the runoff prediction model is improved.The LSTM-BP coupled runoff prediction model is designed,and the runoff sequence prediction model is designed to determine the structure and related parameters of the model.(3)Based on the serious situation of the missing data and the characteristics of time series,based on the GAN network,the LSTM network is added,the data simulation model of GAN-LSTM coupling is proposed,the model structure is designed,and the relevant parameters are determined.(4)Simulate and analyze the LSTM-BP runoff sequence prediction model and GANLSTM data simulation model on the Tensorflow simulation platform.The results show that the overall fitting degree of the LSTM-BP coupled runoff prediction model is good,and the simulation results are basically consistent with the actual runoff variation trend.Compared with LSTM and BP two runoff prediction models,the prediction accuracy is higher,more reasonable and accurate,and the simulation effect is optimal.The data generated by the GAN-LSTM simulation model learns the underlying laws in the original dataset,improves the data quality,and indirectly improves the predictive performance of the LSTM-BP model.