Multivariate Water Level Prediction In Dongting Lake Based On Deep Learning Model Research

Dongting Lake is located on the south bank of the Jingjiang River.It is the second largest freshwater lake in China and an important river-connecting lake in the middle reaches of the Yangtze River.In the past few decades,Dongting Lake has been one of the most frequently and severely flooded areas in the Yangtze River Basin,and the situation of floods in the lake area has become increasingly severe.The duration has grown significantly,and the lake region is facing severe seasonal drought and water shortages.How to accurately and reliably predict the water level of Dongting Lake is of great significance to the management of water resources in the lake area,flood control and drought resistance,and the improvement of flood control efficiency.Hydrological prediction using deep learning method is one of the research hotspots in recent years,which can effectively predict the continuous water level in a certain period of time in the future.Starting from the actual flood control and drought resistance needs of Dongting Lake,combined with deep learning methods,this paper carries out the following research:(1)Determine the research site and data variables according to the geographical and climatic characteristics of the study area and factors affecting the water level of Dongting Lake.Based on this,the data of water level,flow,rainfall,and temperature of multiple water level influencing variables at each site for 18 years from 2004 to 2021 were collected,and the data set was cleaned and standardized.(2)The Dongting Lake water level prediction model was constructed based on the three deep learning principles of multilayer perceptron(MLP),Elman neural network and PSOElman,and the delay response relationship between upstream flow,rainfall and water level stations in the lake area was analyzed,and the The prediction performance of the three deep learning models is comprehensively evaluated,and the PSO-Elman model with the best performance is selected for further optimization.(3)Conduct real-time water level prediction research on the PSO-Elman model with variable data from September 2021 to January 2022.The results show that the PSO-Elman model can better predict the water level of representative sites in the lake area up to 36 hours in advance.(4)In order to measure the influence of each variable on the prediction performance and results of the model,the sensitivity analysis of the model variables is carried out.The results show that the prediction results of the PSO-Elman water level prediction model are more sensitive to flow,followed by rainfall,while the model is less sensitive to air temperature.The Dongting Lake PSO-Elman water level prediction model established in this paper can better capture the complex characteristic water level variation characteristics,and has a very good multi-step long time series prediction ability.The accurate forecast of flood level can provide scientific and reasonable guidance for the relevant departments to carry out the prevention and control of flood and drought disasters.