This is a TensorFlow implementation of GCN-LSTM: A Spatiotemporal Prediction Model for Groundwater Level.
##Requirements:
1.tensorflow
2.scipy
3.sklearn
4.keras
5.numpy
6.matplotlib
7.pandas
8.math
9.xlrd
10.xlwt
##Directory description:
Data: Includes input data for all scripts.
GCN-LSTM-TensorFlow: py file that contains all models and methods
Models: Includes saved, trained models.
##Data Description:
well-all-Original: The groundwater level observation data of 16 wells in Zhangjiajie every five days from 2003 to 2017.
Attribute_Cij: The attribute similarity matrix is calculated based on the properties of distance, elevation, slope and aspect between 16 wells.
Spatial_adj_tin: This is a spatial weighting matrix calculated from the structure of 16-well-graph created by the Delaunay triangulation.
##Run the demo:
Our baselines included:
1.GCN_LSTM
2.LSTM_Dropout_FC
3.LSTM_FC
4.LSTM_Dropout_LSTM
5.LSTM_LSTM
The python implementations of models were in the baselines.py; The GCN Layer were in spektral_gcn.py and spektral_utilities.py.
The method.py includes the methods for calculating adjacency matrices and other general methods.
##Code call relationship
