Open AccessShort-term forecast of Yangtze River water level based on Long Short-Term Memory neural network
Author(s) -
Shinan Chen,
Yunkai Qiao
Publication year - 2021
Publication title -
iop conference series. earth and environmental science
Language(s) - English
Resource type - Journals
eISSN - 1755-1307
pISSN - 1755-1315
DOI - 10.1088/1755-1315/831/1/012051
Subject(s) - yangtze river , artificial neural network , term (time) , epoch (astronomy) , mean squared error , mean squared prediction error , selection (genetic algorithm) , computer science , water level , meteorology , series (stratigraphy) , environmental science , statistics , artificial intelligence , mathematics , machine learning , geography , geology , cartography , stars , physics , archaeology , quantum mechanics , china , computer vision , paleontology
The water level fluctuation forecast of Yangtze River plays an important role in the navigation planning and other areas. This study used LSTM neural network to make short term forecast of the water level of Nanjing navigable river, focusing on the 2 days, 3 days and 5 days-forecast from the past 14 days. The error of mean square reached 0.064, 0.121 and 0.195, showing a rather accurate prediction. The model was also suitable for longer time series prediction. This study optimized the model by adjusting the hyper-parameters using qualitative and quantitative analysis and further decided that the batch size equals to 90 and the epoch equals to 250 in the case of 5 days-forecast from the past 14 days. The prediction accuracy increased by 21% with a good prediction performance, and the error of mean square was reduced to 0.153. It provided a reference for the selection of hyper-parameters for the prediction of water level in Nanjing station by LSTM neural network.