We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
多变量 LSTM 神经网络模型在地下水位预测中的应用.
- Authors
孙虹洁; 赵振华; 黄林显; 邢立亭; 郝 杰; 罗振江
- Abstract
The accurate prediction of groundwater level has great significance to the management of groundwater. A prediction model based on multivariate long short-term memory ( LSTM) neural network was constructed by taking the precipitation representing the recharge term and the temperature, vapor pressure and groundwater extraction representing the discharge term. The model used monthly precipitation, tempera-ture, vapor pressure, groundwater extraction, groundwater levels of quaternary and karst aquifer data from 2010 to 2018 of Jinan City as training set while used monthly data of 2019 as validation set. The results show that: a) using sine function signal to fit temperature data can eliminate the influence of temperature measurement errors and improve model prediction accuracy; b) the optimal prediction results can be obtained when the neuron dropout rate is 20%; the prediction Root-Mean-Square Error( RMSE) of quaternary groundwater level is 0.84 m, and the prediction Root-Mean-Square Error( RMSE) of karst groundwater level is 0.68 m; c) in general, the LSTM model can accurately simulate the dynamic characteristics of groundwater levels, and the predicting error is relatively larger only when the groundwater level is dra-matically changed.
- Publication
Yellow River, 2022, Vol 44, Issue 8, p69
- ISSN
1000-1379
- Publication type
Article
- DOI
10.3969/j.issn.1000-1379.2022.08.014