This repo contains source codes, input data, and an example output for the hybrid neural network proposed in the article Wen et al. (2022), which targets improving the accuracy of a physics-based model in simulating soil reaction front using the neural network.
Wen, T., Chen, C., Zheng, G., Bandstra, J., and Brantley, S.L. (2022). Using a Neural Network – Physics-based Hybrid Model to Predict Soil Reaction Fronts. Computers & Geosciences, https://doi.org/10.1016/j.cageo.2022.105200
- Download this repo as a zip file to your local computer. Unzip the downloaded file.
- Make sure all of the dependencies are installed before running the HNN codes. The list of dependencies is listed below
| Required Package | Version |
|---|---|
| Python | 3.8.3 |
| sympy | 1.10.1 |
| tensorflow | 2.8.0 |
| pandas | 1.3.5 |
| numpy | 1.22.3 |
-
Make sure all of the above dependencies are installed before running the code
-
Specify parameters in the final.py Hyperparameters:
WHETHER_TRAIN = True ## Training the model from scratch. If using the pre-trained model, please set it to False.
ONLYBEST = True ## Only train/test the best performance models ['7a', '7b', '9a', '11a', '11c', '14a','14b']; Set to False if training all models.
HOME_DIR = "./test" ## Define the directory where the result will be saved.
-
Input data: Input data is saved in data/all_data.csv
-
Run the finaly.py script.
python final.py
This work is licensed under the MIT license.