Source code for the paper "Integral-form physics constrained parallel network for identifying nonlinear dynamical systems from only noisy displacement measurements".
The code is written in Python 3.10.15, and the required packages are listed in requirements.txt. You can install them using the following command:
pip install -r requirements.txtAll the hyperparameters used in the experiments are stored in the config.py file.
The data used in the experiments are stored in the data folder, including two noise types and three different noise levels.
- 25 dB (approximately) pink Gaussian noise:
data/nonautonomous_duffing_0.11.mat - 20 dB (approximately) pink Gaussian noise:
data/nonautonomous_duffing_0.19.mat - 25 dB white Gaussian noise:
data/nonautonomous_duffing_25_white.mat - 20 dB white Gaussian noise:
data/nonautonomous_duffing_20_white.mat - 15 dB white Gaussian noise:
data/nonautonomous_duffing_15_white.mat
To train the model, you can run the following command:
python run.pyThe trained model will be saved in the models subfolder of outputs folder.
The visualization of the training process can be found in the figures subfolder of the outputs folder.
The output of the model will be saved in the results subfolder of the outputs folder.