Real-time building occupancy sensing using neural-network based sensor network

Current occupancy sensing technologies may limit the effectiveness of buildings controls, due to a number of issues ranging from unreliable data, sensor drift, privacy concerns, and insufficient commissioning. More effective control of Heating, Ventilation and Air-conditioning (HVAC) systems may be possible using a smart and adaptive sensing network for occupancy detection, capable of turning off services out of hours, and not over-ventilating, thus enabling energy savings, and not under-ventilating during occupied periods, giving comfort and health benefits. A low-cost and non-intrusive sensor network was deployed in an open-plan office, combining information such as sound level, case temperature, carbon-dioxide (Co₂) and motion, to estimate occupancy numbers, while an infrared camera was implemented to establish ground truth occupancy levels. Symmetrical uncertainty analysis was used for feature selection, and a genetic based search to evaluate an optimal sensor combination. Selected multi-sensory features were fused using a neural network. From initial results, estimation accuracy reaching up to 75% for occupied periods was achieved. The proposed system offers promising opportunities for improved comfort control and energy efficiency in buildings.