Bayesian filtering for building occupancy estimation from carbon dioxide concentration

This paper proposes a new framework based on Bayesian filtering for building occupancy estimation from the observation of carbon dioxide concentration. The proposed framework can fuse a statistical model and an observation model for better occupancy estimation. The statistical model can capture the temporal dependency of the building occupancy, and the first-order inhomogeneous Markov model is utilized for the estimation of occupancy transition probability. The observation model can estimate the occupancy level from carbon dioxide concentration. The likelihood is obtained from the solution of the
observation model. To identify the observation model, we present a novel ensemble extreme learning machine technique. Applying the Bayes filter technique, we can fuse the transition probability and the likelihood for better occupancy estimation. The proposed framework can be applied for general cases of occupancy estimation, and the solution outperforms the results of the observation model. The results of a real experiment show the effectiveness of the proposed method.