Our PhD student June is presenting his research on intelligent environments at the IBPC conference in Syracuse. His talk is entitled LightLearn: Occupant centered lighting controller using reinforcement learning to adapt systems to humans.
Abstract:
Humans spend up to 90% of their time indoors, thus maintaining and improving indoor environmental quality can increase their comfort and productivity. In this regard, building systems are employed to control the indoor environment within the comfort range. With the rapid development, decreasing size, and reduced costs of information and communication technology (ICT) systems, existing buildings can be retrofitted with smart technologies (i.e., sensors, actuators, communication, Internet-of-Things (IoT) platforms). In this paper, we propose a retrofit solution for an occupant centered controller (OCC) for lighting in existing buildings. In the OCC framework, the lighting control agent interacts with the occupant non-intrusively, learning from the behavior of the occupant, and determining adaptive set-points for achieving both human comfort and energy saving. We present both hardware and algorithm. The central control node is a Raspberry Pi microcontroller, which collects illuminance (lux), light switch position (on/off), and occupancy data (occupied/unoccupied). The light switch position is tracked by an off-the-shelf product via Bluetooth. On the other hand, occupancy data is monitored via the Bluetooth signal on mobile devices (smartphones, wearables, etc). The OCC algorithm in this paper is based on model-based reinforcement learning to control the lighting system. During the training phase, occupant behavior and system are modeled as a Markov Decision Process (MDP). Then, the agent calculates the optimal policy (i.e., the policy that leads to the maximum cumulative reward) using value iteration. Subsequently, in the control phase, the agent determines the switch action (on/off) based on this optimal policy. We present the experimental results of the OCC approach for an office space in a university building.
