Implementing Bayesian inference on a stochastic CO2-based grey-box model

Abstract

The COVID-19 pandemic brought global attention to indoor air quality (IAQ), which increases public’s awareness on monitoring indoor ventilation conditions significantly. Indoor CO₂ monitoring has been widely accepted as an effective way for indicating IAQ conditions, attributed to its close relationships with indoor air change rates. However, real-time estimation of air change rates or CO₂ emission rates from CO₂ measurement data remains challenging due to uncertainties in factors like random air movements, dynamic conditions (e.g., weather and occupancy), and the limitations of deterministic equations. This study addresses these challenges by applying Bayesian inference to a stochastic CO₂-based grey-box model, enabling the accurate estimation of ventilation and CO₂ emission rates while accounting for uncertainty. The model’s accuracy and robustness were validated through CO₂ tracer gas experiments, employing constant injection and decay methods in a large-scale aerosol chamber. Both prior and posterior predictive checks (PPC) were performed to verify this approach. The approach proposed by this study improves the interpretation of CO₂ monitoring data, thereby facilitating the future real-time IAQ management.