Heating, ventilating, and air conditioning (HVAC) systems account for 51% of the total energy usage in buildings. Adaptive lighting, variable air volume hoods, indirect evaporative pre-cooling, and demand response systems are important for achieving energy efficient sustainable buildings. However, the current state of these technologies will only take us so far towards achieving the net-zero energy buildings. The majority of these systems assume maximum room occupancy; all rooms are conditioned without regard to actual usage. For example, a conference room could be conditioned assuming an occupancy of 30 people when only 20 people actually use the room. Also, it is possible to avoid conditioning the room when empty. Thus, effective energy management requires real-time occupancy measurement. A system for occupancy monitoring is essential to solving this fundamental energy management problem. Perhaps just as important as an occupancy estimation system is occupancy prediction models. This needs to be addressed since conditioning a room is not instantaneous and requires time for adjustments. For example, if it is known that a large number of people are in a lobby area, we want the HVAC system to know an adjacent conference room will be used with high probability and begin conditioning the room beforehand.
SCOPES, a distributed Smart Cameras Object Position Estimation System that balances the trade-offs associated with camera sensor networks. Each node in the system is comprised of a Cyclops camera that performs local detection and processing of the visual information and a Tmote sensor node, which provides power and multi-hop communication capability. SCOPES uses local adaptive techniques to maximize the the active sensing time of the camera. The system switches between fast and simple background subtraction algorithms for object detection and the more computationally intensive pixel grouping algorithms for estimating the number and direction of travel of multiple persons in the local field of view. By aggregating meta-information generated by each node, SCOPES is able to minimize the total data transmitted in the network and still be able to generate an accurate density estimation map of the coverage area.
Models of occupancy can be created by gathering data over a long period of time using a system such as SCOPES. We have developed a Multivariate Gaussian model and an agent based model using several days of ground truth occupancy data. The agent based model simulates occupancy by modeling the behavior of the individual. Agents are given paths, walking speed, and iterates based on the occupancy changes seen in the training data. Occupancy is simulated by creating multiple agents that follow probabilistically generated instructions. Based on their simulated movement, room occupancies over the course of the day can be estimated. The multivariate Gaussian model creates a Gaussian fit of room occupancies for each hour. These fits allows us to calculate probabilities for occupancy changes over time. Currently we are developing models based on Gaussian mixture models and Markov chains. These models are useful design tools for simulating how occupants utilize space. In particular, these models can be used for creating control strategies for HVAC systems.
Using occupancy models to examine user mobility patterns in buildings, we can predict room usage thereby enabling us to control the HVAC systems in an adaptive manner. By controlling HVAC using a multivariate Gaussian model and an agent based model occupancy predictions, simulations indicate a 14% reduction in HVAC energy usage by having an optimal control strategy based on occupancy estimates and usage patterns.