Three-Dimensional Urban Airflow Prediction Method Based on Space-Filling Curve Autoencoder

Air flow prediction in urban is of great significance to various industries, and it is also a complex engineering problem. Using the traditional computational fluid dynamics method to predict the three-dimensional urban requires a large amount of computational resources, which cannot meet the demand of real-time prediction. This paper proposes a reduced order model for air flow prediction in urban based on deep learning. The model uses the space-filling curve method to establish the mapping from 3d unstructured mesh to 1d sequence, and then extracts features from the mapped flow field snapshots by convolutional autoencoder, and finally uses the LSTM network to predict the flow field. It is proved experimentally that this model is able to obtain better reconstruction effect than conventional reduced order model, and can predict the air flow pattern accurately.