Performance comparison using different multilayer perceptron input–output formats to predict unsteady indoor temperature distribution

Computational fluid dynamics (CFD) is widely used to predict the indoor thermal environment; however, large time cost represents a significant disadvantage. Several deep learning approaches have been introduced reduce prediction in steady-state predictions, though their feasibility under unsteady ones has yet be investigated. Considering flexibility of multilayer perceptron (MLP) input–output format, this study compared performance two MLP formats, MLP-A (simultaneously outputting values for all cells space single calculation run) and MLP-B (outputting cell each run), when temperature distribution three scenarios: interpolation, extrapolation, varying boundary conditions. The considered formats resulted different patterns interpolation scenario: accurately predicted spatiotemporal development airflow CFD results, whereas did not. Both performed poorly extrapolation scenario but exhibited error patterns. Finally, also generally provided correct as well. This contributes an understanding by prediction.