Codes for A Cutting-Edge Conceptual Reservoir Operation-Based Deep Learning Framework
Description
Training and analysing code for A Cutting-Edge Conceptual Reservoir Operation-Based Deep Learning Framework (CRO-LSTM). Descriptions for model training and analyzing steps can be found at “Steps to reproduce” in the link of source code. Data description: (1) runoff, rain and reservoir data in the testing stage collected from the management authority of the Minjiang basin. hydro_data.csv : Include basin runoff data and reservoir data. rain_data.csv: Include rainfall observation data. (2) The processed data for model teseting. Resampling and averaging methods were used to handle missing values and mitigate fluctuations. data.csv: processed data. data.pth: processed data packed for model development. shap_values.pth: computed shapley values from the test datasets.
Files
Steps to reproduce
Trainning and analysing steps. 1. Hyperparam_search.py. Using the Optuna tool for hyperparameter searching.(Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. 2019.Optuna: A Next-generation Hyperparameter Optimization Framework. In KDD.).Detailed information for Optuna can be found in https://optuna.org. 2.MultiTrain.py. Taking a multi-round training and testing process for the CRO-LSTM, LSTMr, and LSTMc models. 3.Shapley.py. Using the Shap tool for Shapley analyse.Detailed information for Shap can be found in https://shap.readthedocs.io
Institutions
- Wuhan University