We have introduced a new simulation environment that is the result of merging CitySim, a building energy simulator, and TensorFlow, a powerful machine learning library for deep learning. This new simulation environment has the potential for developing building energy scenarios in which machine learning algorithms, such as deep reinforcement learning, are applied to of the major problems and opportunities modern cities face, e.g., the increased demand for heating and cooling due to increasing populations [1].
This simulation environment allows to study model-free and self-tuning control algorithms, such as deep reinforcement learning (DRL) when integrating distributed renewable energy sources and storage devices into buildings. DRL can learn on-line and off-line from historical sensor data, and it can adapt to diverse changes in the system it controls on both the demand and the supply side. Its off-line learning feature allows it to be safely implemented with a back-up controller and operational constraints.
[1] Vázquez-Canteli, J.R., Ulyanin, S., Kämpf J., and Nagy, Z., “Fusing TensorFlow with building energy simulation for intelligent energy management in smart cities”, Sustainable Cities and Society, 2018.
