This project provides a U-Net based model for solving a Laplace equation-related temperature distribution problem. The model takes boundary values as input and predicts internal temperature distributions.
Run the matlab code heat_eq.m to genrate the training data. Training data are temperature mapping in with different boundary condition settings.
All data would be saved in a .csv file, each row represents a data point in training.
Python Version: Python 3.8 or higher is recommended.
Packages:
- TensorFlow (2.x)
- NumPy (1.19+)
- Matplotlib
- SciPy
To install the required packages, run:
pip3 install tensorflow numpy matplotlib scipy
The code expects a CSV file named laplace_results_v3.csv containing the data. Each line of the CSV file (except the header) should represent a flattened 100x100 temperature field.
Ensure that laplace_results_v3.csv is present in the same directory as the code.
Set train_mode in the code:
If train_mode = True, the code will train a U-Net model using the specified dataset. It will save the best model as best_model_v3.h5.
If train_mode = False, the code will load the previously saved best_model_v3.h5 and evaluate it on the test set. It will also visualize a prediction sample.
Execute the Code:
python3 main.py
During training (train_mode = True):
The code splits the dataset into training, validation, and testing sets (70%/20%/10%).
It trains the model for up to 50 epochs, using early stopping to prevent overfitting.
After training, it evaluates the model on the test set and plots training/validation curves for Loss and MAE.
It saves the best performing model as best_model_v3.h5.
During evaluation/prediction (train_mode = False):
The code loads the pre-trained best_model_v3.h5.
It evaluates performance on the test set. It selects one sample from the test set, makes a prediction, and visualizes both the true and predicted temperature fields.
Examples:
