CO₂-based occupancy forecasting with an Agent-Based Model

In the realm of building performance optimization, understanding occupancy dynamics is pivotal for enhancing both energy efficiency and occupant comfort. Occupancy forecasts, serving as critical inputs for data-driven predictive control technologies, play a significant role in this domain. To address this need, we propose a novel model that directly estimates building occupancy levels. This model is particularly applicable to buildings equipped with mechanical ventilation systems and CO 2 concentration sensors. The number of persons is estimated by utilizing the CO 2 production rate of people and applying the principle of mass conservation. The CO 2 -based approach has been validated with manually recorded ground-truth measurements. A forecast is generated using the first order Markov chain model in combination with an Agent-Based Modell (ABM). The probability transition matrix of the Markov chain defines the behaviour of the occupant-agents, which is used in the ABM to generate behaviour profiles. The model has been tested on four office rooms, with a one-year measurement dataset. The Markov chain with ABM provides a forecast, which encompasses the stochasticity of people’s behaviour. The presence True Positive Rate (TPR) reaches 50 % and the False Positive Rate (FPR) is 15 %, in average. The occupancy TPR is only 30 % and the FPR 15 %. The proposed approach offers a framework to easily implement further variables, like occupancy-related power consumption, lighting operation, window opening etc.