A Modeling Method of Pre-training and Fine-tuning for Non-uniform Indoor Environment

The detailed information of indoor environment is crucial for the proper operation control of non-uniform thermal comfort and energy efficiency. With the concept of transfer learning, this study proposed a pre-training and fine-tuning modeling method which can provides detailed information of an indoor environment. The proposed modeling method was comprised of two stages. In the first stage, the full-scale experiment was conducted using computational fluid dynamics to obtain the dataset to train a surrogate model of numerical simulation. In the second stage, quad-views color sequence particle streak velocimetry was used to obtain the dataset to fine-tune the constructed surrogate model. In the above two stages, physics informed neural network was used to build the model of the non-uniform environment. The performance was experimentally validated to apply to a displacement ventilation in a test chamber.