인공신경망 모델에 따른 재실시간 중 실내 PM_2.5 예측성능 비교

다국어 초록 (Multilingual Abstract)

This study developed an artificial intelligence-based indoor PM_2.5 prediction model to proactively provide comfortable air quality in elementary school classrooms and evaluated its feasibility for real-world application. To achieve this, both forward neural network-based and recurrent neural network-based models were developed and compared to derive an optimal artificial neural network model. The forward neural network-based MLP model demonstrated the best performance and high accuracy prediction within a living lab environment. This suggests that the use of a forward neural network-based model is effective in predicting PM_2.5 levels during occupancy times in elementary school classrooms. The final prediction model can be integrated into the control algorithm of classroom ventilation systems for proactive PM_2.5 management. However, the study identified periods where predictive performance declined due to insufficient training data, highlighting the need for continuous retraining. This ensures adaptation to changing environmental conditions and guarantees optimized indoor conditions. The findings of this study enable the continuous maintenance of a comfortable indoor environment during specific times in elementary school classrooms.