A data-driven model to determine the infiltration characteristics of air curtains at building entrances

The air curtain reduces heat exchange between the two sides by creating a virtual partition and works as a solution for improving building sealing and energy efficiency. Currently, the analytical numerical coupling method has achieved some success in describing the low-order theoretical descriptions of air curtain flow, but its application scope is limited. This paper introduces a data-driven model (DDM) to predict the operation state of the air curtain and the volume flow rate through the entrance. A computational fluid dynamics model is built to generate the dataset, which is validated by comparing velocity and volume flow rate with the published data in the literature. Three of the widely used algorithms are tested: support vector machine, random forest, and backpropagation neural network (BPNN). The main conclusions are as follows: (1) The combination of pressure difference and air supply velocity can quickly determine the operation state of the air curtain in the scene (f1-score = 0.9). (2) A single hidden layer BPNN can achieve high-precision prediction of volume flow rate (R2 = 0.92). (3) Compared to theoretical methods, the DDM can retain three-dimensional characteristics of the jet and capture additional details. The approach proposed in this paper can be applied to practical environments to rapidly and accurately optimize the operating parameters of air curtains.