| 수록사항 |
한국생활환경학회지 , Vol.30 No.6(2023-12) |
| 주제어 |
; Machine Learning; Residential house; CO2 concentration prediction |
| 요약2 |
In this study, the performance of an indoor CO2 concentration machine learning model learned using minimal data was verified, such as day, time, temperature, and CO2 concentration. These variables are easy to acquire in residential spaces. The prediction accuracy of three models?ANN, KNN, and LSTM?was confirmed using data from residential units, which are considered standard households in Korea. ANN achieved an R2 of 0.96, CvRMSE of 1.1%, and MBE of 2.14%, showing the best performance. Both KNN and LSTM demonstrated appropriate prediction stability (over 0.8), CvRMSE (within 2%), and MBE (within 4%). When applied to a unit scheduled for future demonstration of the ventilation algorithm, a similar level of prediction accuracy was confirmed, despite differences in shape and residential furniture compared to the standard unit. This study is a basic exploration of applying the ventilation algorithm to the demonstration unit. In future studies, we plan to apply this model to the demonstration unit and operate the ventilator according to the predicted values, aligning with the current legal operating standard of 0.5 ACH for regular operation. We intend to conduct a multifaceted performance evaluation of this ML model by comparing and analyzing the results with fan energy consumption and indoor CO2 concentration values. |
