We have investigated experimentally the use of Reinforcement Learning for lighting control in our own offices. The results are awesome and now published in Building & Environment.
J. Park, T. Dougherty, H. Fritz, Z. Nagy, LightLearn: An adaptive and occupant centered controller for lighting based on reinforcement learning,
Building and Environment,2018,
In commercial buildings, lighting contributes to about 20% of the total energy consumption. Lighting controllers that integrate occupancy and luminosity sensors to improve energy efficiency have been proposed. However, they are often ineffective because they focus solely on energy consumption rather than providing comfort to the occupants. An ideal controller should adapt itself to the preferences of the occupant and the environmental conditions. In this article, we introduce LightLearn, an occupant centered controller (OCC) for lighting based on Reinforcement Learning (RL). We describe the theory and hardware implementation of LightLearn. Our experiment during eight weeks in five offices shows that LightLearn learns the individual occupant behaviors and indoor environmental conditions, and adapts its control parameters accordingly by determining personalized set-points. Participants reported that the overall lighting was slightly improved compared to prior lighting conditions. We compare LightLearn to schedule-based and occupancy-based control scenarios, and evaluate their performance with respect to total energy use, light-utilization-ratio, unmet comfort hours, as well as light-comfort-ratio, which we introduce in this paper. We show that only LightLearn balances successfully occupant comfort and energy consumption. The adaptive nature of LightLearn suggests that reinforcement learning based occupant centered control is a viable approach to mitigate the discrepancy between occupant comfort and the goals of building control.