State Space Neural Networks - A PyTorch-powered interface to perform NN-based system identification

For more information on this code, please refer to my PhD dissertation (in particular, to Chapter 3 and Chapter 4)

Fabio Bonassi, “Reconciling deep learning and control theory: recurrent neural networks for model-based control design,” 2023, Politecnico di Milano. PhD Dissertation. Supervisor: Prof. Riccardo Scattolini, Prof. Marcello Farina [link]

If you use the code, please consider citing the PhD dissertation.

@phdthesis{bonassi2023reconciling,
  title = {Reconciling deep learning and control theory: recurrent neural networks for model-based control design},
  author = {Bonassi, Fabio},
  year = {2023},
  month = feb,
  address = {Milan, Italy},
  school = {Politecnico di Milano},
  type = {PhD thesis},
}

Or, alternatively, the corresponding Springer Brief

@incollection{bonassi2024reconciling,
  title = {Reconciling Deep Learning and Control Theory: Recurrent Neural Networks for Indirect Data-Driven Control},
  author = {Bonassi, Fabio},
  booktitle = {Special Topics in Information Technology},
  pages = {77--87},
  year = {2024},
  publisher = {Springer},
  doi = {10.1007/978-3-031-51500-2_7},
}

To illustrate how to use this python library, the script example_ph.py is now included. The script fits a stable deep GRU to the data collected from the pH neutralization process.

You can install the requirements by running pip install -r requirements.txt. You need Python 3.10 or later.