Research Review
Prediction of two-dimensional non-isothermal indoor airflow by deep learning
2020 Volume 72 Issue 1 Pages 57-64
Details
2020 Volume 72 Issue 1 Pages 57-64
Deep Neural Network (DNN) is utilized to predict indoor velocity distribution and temperature distribution.
The non-dimensional Archimedes number (Ar) is put into the DNN as input parameter. DNN can well reproduce the indoor airflow distribution with the maximum relative error of 8%. The performance of DNN for respective prediction of velocity / temperature distribution is better than that for simultaneous prediction. It requires at least 1000 seconds for each case by CFD simulation while it takes 0.1 second for DNN prediction for each case. The high speed and high precision prediction of DNN for two dimensional non-isothermal indoor air distribution is confirmed.
