Assessment of different airflow modeling approaches on a naturally ventilated Mediterranean building

• Full-scale experiment in a residential building with natural cross ventilation.
• Proposition of a compromise between measurements and simulation.
• Use of statistical tools to model airflow rate.
• Artificial Neural Networks provide better results than calibrated empirical models.
• Proposition of a simplified model adapted to building ventilation control.

This paper focuses on a full-scale experiment to assess and model the airflow rate in a naturally ventilated room using different approaches. The building studied is located in a coastal area of Corsica and mostly affected by thermal breezes phenomena which lead to high airflow rate during day (between 8 and 30 ACH) and lower during night (between 2 and 8 ACH). The first aim of this work is to set up a method in order to measure continuously the airflow rate in cross ventilation configuration using a minimal number of sensors. Our methodology involves direct measurements of velocity on a mesh and use of statistical methods. The second objective is to develop and evaluate different airflow modeling approach in cross natural ventilation configuration. Various levels of complexity are tested and compared: empirical modeling, model calibration and behavioral modeling based on artificial neural networks. In terms of error, the artificial neural network appears to be the best compromise to model the airflow rate and allow to reach a MAE of 1.75 ACH with a one minute time step.Suggested model in this paper can be coupled with a thermal model and is suitable for model based natural ventilation control.