Rina Hirai, Shohei Miyata, Yasunori Akashi
Languages: English | Pages: 9 pp
Bibliographic info:
41st AIVC/ASHRAE IAQ- 9th TightVent - 7th venticool Conference - Athens, Greece - 4-6 May 2022

Heating, ventilating, and air conditioning (HVAC) systems attempt to achieve a uniform indoor environment. However, this can be challenging, because the placement and control of HVAC systems and sensors are affected by many unpredictable factors. The efficacious exploitation of this nonuniformity can lead to an improvement of indoor environment around occupants. Of the many indoor environment variables, we focused on the CO2 concentration associated with ventilation. In the first part of this two-part study, computational fluid dynamics (CFD) was utilized to determine the CO2 concentration distribution in an office space. First, we investigated the effects of the recommended ventilation improvement countermeasures including  (a) increasing ventilation volume, (b) placement of air circulators (fans), and (c) restricting the number of occupants with staggered seating. Measure (a) contributed to expanding a well-ventilated area, whereas measure (b) resulted in the leveling of the CO2 concentration distribution; in the case of (c), most occupants stayed in the well-ventilated area. Based on these results, we increased the number of occupants by relocating air circulators, desks, and occupants, and 72 occupants could stay in well-ventilated areas. The target space capacity calculated from the design standard was 75. In the second part, we attempted to predict the CO2 concentration around occupants using neural network. The dataset of seating arrangement, sensor CO2 concentration, and CO2 concentration of each seat were created using CFD simulation. In this study, the CO2 concentration of each seat was calculated with a root mean square error of less than 10 ppm. In future studies, it may be possible to increase the number of occupants in a well-ventilated area by considering the nonuniformity, and the use of neural network should be tested with real world data as input.