Occupancy Estimation Based on Indoor CO2 Concentration: Comparison of Neural Network and Bayesian Methods

Abstract

The number of occupants in a space can significantly affect ventilation control. Using neural network and Bayesian Markov chain Monte Carlo (MCMC) methods, this study estimates the number of occupants based on CO2 concentration in a room. The abilities of both methods to recognize the input-parameter characteristics are compared under certain circumstances, and the parameters are optimized to improve the estimation accuracy. The neural network trains an input dataset of CO2 concentrations, ventilation rates, and occupancy patterns with tapped delay lines. Meanwhile, the Bayesian MCMC calculates the given CO2 data by a mathematical model based on a statistical approach. The present space model is a single-office room in which the CO2 concentration is determined through several simulation schemes and experiments. The estimation accuracy of the neural network depends on the complexity of the input parameters (i.e., CO2 concentration and ventilation rate), whereas the Bayesian MCMC is influenced by uncertainty in the CO2 concentration. Both methods produce acceptable estimates under certain treatments.