科研之友
微信
新浪微博
Facebook
分享链接
摘要
Dongting Lake basin is one of the large freshwater wetland system in China, having three important nature reserves. Water level is the basis to maintain the structure, function and integrity of the lake ecosystem. In this study, two special types of recurrent neural network method, Long Short-term Memory (LSTM) and The Gated Recurrent Unit (GRU), were propused to construct the Dongting Lake's water level prediction model, in order to predict the water level change of Dongting Lake under the synthetical influence of the Yangtze river and the four sub-tributaries in the local catchment. The advantage of the LSTM and the GRU is their ability to learn the long-term dependencies between the provided input and output of the network, whichis important to simulate the cumulative change of water level affected by upstream inflow. The daily flows of five hydrological stations, which located at the stem stream of Yangtze River and the four sub-tributaries in the local catchment, were provided as the model's input conditions to predict the process of water level change in different Dongting Lake areas. The water level and flow time series from 1980 to 2002 were used to test the model, and the data set from 2003 to 2014 were applied to validate our approach. In addition, the predicted results of the two models were compared. The results show that: (1) Models based on the recurrent neural network methods of LSTM and GRU can reasonably predict the process of water level change in three Dongting Lake areas, both NSE and R2 are 0.91 - 0.95. (2) The accurate of LSTM model is slightly higher than the GRU. It's worth to say that the GRU model is a good alternative to the LSTM for that the former's calculation is more efficient than the latter. (3) The model can simulate a flood event more accurately. The maximum relative error between the predicted value of flood level and the real value is less than 5%. Moreover, the model has a good prediction ability of multi-step time series, which underlines the potential of the LSTM for hydrological modelling applications.
- 单位
