Occupant Behavior Prediction and Real-Time Correction-based Smart Building Energy Optimization

Buildings are one of the biggest energy consumers and greenhouse gas producers. Technology could help reduce their environmental impact by deploying sensors to collect relevant data about the way energy is consumed, and with the aim to optimize it. For this sake, understanding building occupant's behavior and occupancy patterns can help minimize energy consumption while satisfying occupant's comfort. Indeed, occupants directly influence building appliances that consume energy, such as HVAC, ovens, hot water tanks, etc. In this paper, we aim at predicting occupants' movements among rooms and use the predicted movements to deduce room and space occupancy in the building. The latter is then used to preheat/pre-cool rooms. However, since prediction models are not always that accurate, it is possible to face situations where HVAC of some rooms are activated while these are empty or vice-versa, leading to either a waste of energy or a lack of occupant's comfort. To deal with this issue, we make use of sensors to detect real-time occupancy of building rooms and then correct the prediction when necessary. To achieve this, we developed a graph mining-based optimization approach that combines occupant behavior prediction and a real-time correction. We experimented our approach on simulated data and results showed that our model optimizes up to 39.09% of HVAC energy consumption, and provides up to 99.39% of occupants' comfort.