Dynamic Structure Learning through Graph Neural Network for Forecasting Soil Moisture in Precision Agriculture

This is a Pytorch implementation of “Dynamic Structure Learning through Graph Neural Network for Forecasting Soil Moisture in Precision Agriculture”.

Requirements

• matplotlib=3.3.4=pypi_0
• networkx=2.4=py_1
• numpy=1.18.5=py38h1da2735_0
• scikit-learn=0.23.1=py38h603561c_0
• scipy=1.5.0=py38hbab996c_0
• sklearn=0.0=pypi_0
• statsmodels=0.11.1=py38haf1e3a3_0
• torchvision=0.7.0=py38_cpu

Dependency can be installed using the following command:

pip install -r requirements.txt

Model Training & Testing

Hyperparamters can be sent through command line or can be changed in the train files itself. Exact values to be used is given in the appendix of the paper.

• DGLR (Shared) ``` #USA python train.py –dataset USA –nofstations 68 –model DGLR_SHARE

#Spain python train.py –dataset Spain –nofstations 20 –model DGLR_SHARE

#Alabama python train.py –dataset Alabama –nofstations 8 –model DGLR_SHARE

#Mississippi python train.py –dataset Mississippi –nofstations 5 –model DGLR_SHARE

- DGLR (w/o SL)

#USA python train.py –dataset USA –nofstations 68 –model DGLR

#Spain python train.py –dataset Spain –nofstations 20 –model DGLR

#Alabama python train.py –dataset Alabama –nofstations 8 –model DGLR

#Mississippi python train.py –dataset Mississippi –nofstations 5 –model DGLR

- DGLR (w/o Sm)

#USA python train_graphrefine.py –dataset USA –nofstations 68

#Spain python train_graphrefine.py –dataset Spain –nofstations 20

#Alabama python train_graphrefine.py –dataset Alabama –nofstations 8

#Mississippi python train_graphrefine.py –dataset Mississippi –nofstations 5

- DGLR (our model)

#USA python train_graphrefine_smoothness.py –dataset USA –nofstations 68

#Spain python train_graphrefine_smoothness.py –dataset Spain –nofstations 20

#Alabama python train_graphrefine_smoothness.py –dataset Alabama –nofstations 8

#Mississippi python train_graphrefine_smoothness.py –dataset Mississippi –nofstations 5 ```

Training Details

We have run all the experiments on a CPU with 2.6GHz 6-core Intel Core i7. We conduct extensive hyperparameter tuning for all the baseline algorithms and report the best results obtained. Detailed hyperparameter set up is given in the appendix of the paper. The presentation of the paper is available in slides and the poster is available here.