In Heliyon
An artificial neural network (ANN) has been broadly developed as a design tool in various application scenarios in building sectors. One of the most important perspectives in building fields is human comfort. Various control strategies of natural ventilation schemes exist, to maintain good air quality in buildings. Nevertheless, this study presented a novel strategy by applying a simple ANN to predict the trends of indoor air temperature and determine the operation status of operable windows. Building simulations had been conducted to train, test, and validate the ANN model. The ANN model has one hidden layer and performs training using the Levenberg-Marquardt algorithm. The nodes in the hidden layer were varied to configure the best-fitting model. The best structure of the ANN model in this study is the model with one hidden layer and 20 nodes. This study compares the significance of adopting a data set between differential data with time series and raw data. The application of the differential data set exhibits better performance in predicting the indoor air temperature increase or decrease than that of the raw data. The prediction precision between the simulation and the ANN model when adopting the differential data is higher than that of raw data by 18%. This study discovered a new simple method and verified that a simple control strategy has been achieved by predicting the window operations using the increase or decrease in indoor temperatures via the ANN application.
Srisamranrungruang Thanyalak, Hiyama Kyosuke
2022-Nov
Airflow, Artificial neural network, Machine learning, Natural ventilation, Stack, Window control
