재실자 정보 기반 실내 CO2 농도 예측모델 개발 및 성능 비교

Purpose: This study aimed to develop a CO2 prediction model to enhance indoor air quality (IAQ) management by incorporating real-time occupant-specific characteristics such as activity levels (MET), gender, and BMI. Traditional models typically rely on basic variables like occupant count, often ignoring individual factors that significantly affect CO2 emissions. By including these variables, the model seeks to improve predictive accuracy and optimize ventilation control systems for enhanced energy efficiency and IAQ management. Method: Two predictive models were developed using DNN and GRU machine learning algorithms. One model utilized basic number of occupants and environmental (outdoor CO2, indoor CO2, ventilation system flowrate) data, while the enhanced model incorporated MET, gender, and BMI. Both models were trained on mock-up data collected from a controlled environment, including CO2 concentration, ventilation flow rates, and occupant information. Their performance was evaluated using MAE, CvRMSE, and R². Result: The enhanced model, integrating occupant-specific variables, demonstrated significant improvements in prediction accuracy compared to the traditional model, with MAE and CvRMSE values improving by 3.93% and 6.92%, respectively. These results highlight the importance of detailed occupant data for real-time IAQ management and the potential for greater efficiency in ventilation control and energy savings in sustainable buildings.