Simulating and Predicting of Hydrological Time Series Based on TensorFlow Deep Learning
More details
Hide details
Huazhong University of Science and Technology,Wuhan, China
Electric Power Research Institute, Jilin Electric Power Co., Changcun, China
State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin and China Institute of Water Resources and Hydropower Research, Beijing, China
Xiaohui Lei   

State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin & China Institute of Water Resources and Hydropower Research
Submission date: 2017-11-15
Final revision date: 2017-12-13
Acceptance date: 2017-12-23
Online publication date: 2018-09-19
Publication date: 2018-12-20
Pol. J. Environ. Stud. 2019;28(2):795–802
Hydrological time series refers to the observation time point and the observed time value. The simulation and prediction of hydrological time series will greatly improve the predictability of hydrological time series, which is of great significance for hydrological forecasting. TensorFlow, the second generation of artificial intelligence learning system in Google, has been favored by a large number of researchers by virtue of its high flexibility, portability, multi-language support, and performance optimization. However, the application of deep learning in hydrology is less. Based on the TensorFlow framework, the AR model and the LSTM model are constructed in Python language. The hydrological time series is used as the input object, and the model is deeply studied and trained to simulate and predict the hydrological time series. The effect of the model was tested by fitting degrees and other indexes. The fitting degree of the AR model is 0.9551, and the fitting degree of the LSTM model is 0.8012, which shows the feasibility of the model for predicting the hydrological time series, and puts forward the solution for the limitation of the existing analysis results.