The jupyter notebooks in this folder were created to present the training and usage of a DQ-DTC reinforcement learning drive control system. A deep Q direct torque controller can be trained via the DQ_DTC_training.ipynb notebook. The trained controller can then be tested with the DQ_DTC_validation_profile.ipynb notebook. A benchmark for the performance of conventional control methods can be received via the MP_DTC_validation_profile.ipynb notebook.
The needed python packages are listed in the requirements.txt and can e.g. be installed via
pip install -r requirements.txt
The python code files CustomKerasRL2Callbacks_torqueCtrl.py and Plot_TimeDomain_torqueCtrl.py implement extensions to the code that were outsourced from the notebooks, they do not need to be executed manually.
Please use the following BibTeX entry to cite this code:
@ARTICLE{9416143, author={Schenke, Maximilian and Wallscheid, Oliver},
journal={IEEE Open Journal of the Industrial Electronics Society},
title={A Deep Q-Learning Direct Torque Controller for Permanent Magnet Synchronous Motors},
year={2021},
volume={2},
number={},
pages={388-400},
doi={10.1109/OJIES.2021.3075521}}