Deep Learning for Energy-Efficient Building Control
Buildings account for more than one-third of the worldwide energy consumption, and they are an equally important source of CO2 emissions. A critical task to improve energy efficiency is to adapt HVAC (heat, ventilation, and air-conditioning) equipment operation to energy demand, minimizing consumption while keeping occupants comfortable. Model-predictive control (MPC) approaches have proved successful to generate HVAC operational control plans, but they still present several drawbacks: they require constructing a detailed physical model of each building, which implies a considerable effort, and have important performance limitations, which strongly limit the opportunities for implementing new intelligent optimization procedures.
PROFICIENT aims at solving these issues by developing novel deep reinforcement learning techniques capable of: (1) learning a more efficient predictive model of the building from sensor data; and (2) optimizing the computation of operational plans without using heuristic knowledge.
Stay tuned for updates!
Principal Investigator: Dr. Juan Gómez-Romero
PROFICIENT is a 2-year project funded by the EXPLORA programme of the Spanish Ministry of Science, Innovation and Universities in 2018-2020 (TIN2017-91223-EXP).