Page 138 - LAS CIENCIAS AMBIENTALES EN EL ANTROPOCENO - DIGITAL(1)

based on the simulation of a simple the EnergyPlus API, and there were no
model implemented in the EnergyPlus errors. Regarding the evaluation metrics,
version 9.5.0 building simulator and with energy consumption and comfort
overwriting for the drive control systems hours, the proposed model failed to
through code written in Python that uses improve with respect to the control
the simulator API. For the evaluation systems with which it was compared.
of the model, the energy consumption However, not all hyperparameters of the
and the hours of comfort obtained algorithm were explored as well as other
with fixed rule systems, conventional learning alternatives. In addition, a clear
today, and with a modern one that advantage of this kind of control is that
applies artificial intelligence, but only it would allow to expand the measurable
to the active system, were compared. variables of decision-making, being able
As a result, the programmed learning to integrate an interaction of the user
algorithm worked correctly, finding an with the system, in such a way that
optimal policy. It is important to note the feeling of comfort can be increased
that it is the first application that has in a holistic way, and not just thermal
been published, according to the best comfort as was done in this work.
knowledge of the authors, of the use of

REFERENCIAS Application. Proceedings of the search. Nature 529(7587), 484–
489.
16th IBPSA Conference. Rome,
Italy, Sept. 2-4. https://doi.org/10 Singh, A.K.; Agrawal, S.; Agarwal,
.26868/25222708.2019.211427 S.; Goyal, D. (2020). Low-Cost
Brandi, S.; Piscitelli, M.S.; Martellacci, Galasiu, A.D.; Veitch, J.A. (2006). and Energy-Efficient Smart
M.; Capozzoli, A. (2020). Deep Occupant preferences and Home Security and Automation.
reinforcement learning to optimise satisfaction with the luminous Computational Netw ork
indoor temperature control and environment and control systems Application Tools for Performance
heating energy consumption in in daylit offices: a literature review. Management, M. Pant, T. K.
buildings. Energy & Buildings, 224, Energy and buildings, 38(7), 728- Sharma, S. Basterrech, and C.
110225. https://doi.org/10.1016/j. 742. Banerjee, Eds. Singapur, pp. 95–
enbuild.2020.110225 Nicoletti, F.; Carpino, C.; Cucumo, M.A.; 108.
Coraci, D.; Brandi, S.; Piscitelli, M.S.; Arcuri, N. (2020). The Control of Skansi, S. (2018). Introduction to Deep
Capozzoli, A. (2021). Online Venetian Blinds: A Solution for Learning: from logical calculus to
implementation of a Soft Actor- Reduction of Energy Consumption artificial intelligence. Springer.
Critic agent to enhance indoor Preserving Visual Comfort. Sutton, R.S.; Barto, A.G. (2018).
temperature control and energy Energies, 13(7), 1731. Reinforcement learning: An
efficiency in buildings. Energies, Silver, D.; Huang, A.; Maddison, introduction. MIT press.
14, 997. https://doi.org/10.3390/ C.J.; Guez, A.; Sifre, L.; van den Tian, Y.; Zhu, Y. (2016). Better
en14040997 Driessche, G.; Schrittwieser, J.; computer go player with neural
Dermardiros, V.; Bucking, S.; Athienitis, Antonoglou, I.; Panneershelvam, network and long-term prediction.
A.K. (2019). A Simplified Building V.; Lanctot, M. et al. (2016). arXiv preprint arXiv:1511.06410
Controls Environment with Mastering the game of go with
a Reinforcement Learning deep neural networks and tree

129