Loris Di Natale

 

EPFL PhD Student

Physics-inspired Deep Reinforcement Learning for building control

Empa 
Swiss Federal Laboratories for Materials Science and Technology
Überlandstrasse 129
CH-8600 Dübendorf

Email:

GoogleScholar Linkedin

 

About

Loris Di Natale is a PhD student from Ecole Polytechnique Férédale de Lausanne (EPFL) working at Empa since February 2020 and funded by the NCCR automation. He previously received a B.Sc. in Mathematics and a M.Sc. in Energy Management and Sustainability, both from EPFL.
His main research interests include (Deep) Machine Learning and (Deep) Reinforcement Learning, with an application to the building control problem, to sustain the current energy transition. Advanced algorithms are indeed bound to support the energy systems of the future through automation, and he wants to bridge the gap between the control and Machine Learning communities to design the next generation of algorithms.

Ongoing research

With more and more sensors installed in new or retrofitted buildings, coupled to the higher connectivity of buildings, large amounts of data become available. Loris is looking at ways to leverage that data in control strategies, focusing on the possibilities of Deep Reinforcement Learning to learn interesting optimal policies.
One issue of particular interest, arising when deploying Deep Reinforcement algorithms, is there reliability: since a classical Reinfrocement Learning agent does not have any knowledge of the physics behind the system to controlled, it might take obviously wrong decisions - and potentially dangerous ones. In that context, Loris is looking at ways to give physics-inspired, or more generally prior knowledge to agents.
On the other hand, a challenging problem of Deep reinforcement is the large amounts of data and long training times required to find optimal control policies. This usually prompts scientists to design cumbersome models to train agents before deploying them. Loris is investigating possibilities to shorten the required engineering time to build the models and train the agents, leveraging data in full black-box pipelines to avoid lengthy procedures.

Student projects supervision

2020

  • Master Thesis of Luca Funk on Switzerland in the Energy Transition: The Role of Hydro Power and Electric Vehicles
    Co-supervisors: Dr. Martin Rüdisüli (Empa), Dr. Bratislav Svetozarevic (Empa)
    Professor: Prof. Giovanni Sansavini (ETH Zürich)

Teaching assistant activities

2020

  • Model Predictive Control at EFPL, from professor Colin Jones.

Publications

  • Bratislav Svetozarevic, Christian Baumann, Simon Muntwiler, Loris Di Natale, Philipp Heer and Melanie Zeilinger. Data-driven MIMO control of room temperature and bidirectional EV charging using deep reinforcement learning: simulation and experiments. arXiv preprint arXiv:2103.01886. 2021. (Submitted and under review in Applied Energy)
 
  • Loris Di Natale, Luca Funk, Martin Rüdisüli, Bratislav Svetozarevic, Giacomo Pareschi, Philipp Heer and Giovanni Sansavini. The Potential of Vehicle-to-Grid to Support the Energy Transition: A Case Study on Switzerland. Energies. 2021; 14(16):4812. https://doi.org/10.3390/en14164812.

 

Contact me for anything: