A Novel Dynamic Hidden Semi-Markov Model (D-HSMM) for Occupancy Pattern Detection from Sensor Data Stream

Occupant presence and behaviour have a large impact on building energy performance. With the availability of low cost and affordable sensors, accurate occupancy detection by combining sensor stream data with machine learning approaches becomes possible. In this paper, we propose a novel dynamical hidden semi-Markov model (D-HSMM) which can accurately detect occupancy pattern from sensor data stream in real time. Our approach extends traditional hidden Markov (HMM) and hidden semi-Markov models (HSMM). The novelty of the proposed approach consists in 1) a new dynamic duration modelling way in which the duration is dynamically changing, instead of using fixed duration in traditional HSMM based models; 2) a new approach to state prediction (i.e. occupant presence or absence in this case) based on a weighted function with partially available observations instead of using the whole set of observations. In order to evaluate the performance of our model, we have compared our results with traditional HMM and HSMM approaches. The experimental evaluation shows that our D-HSMM model outperforms the conventional HMM and HSMM-based approaches with very high accuracy.