POD-Based Reduced-Order Modeling of Indoor Temperature Fields in Impinging Jet Ventilation: Parameter-Based Prediction and Sparse Sensor Reconstruction

Proper orthogonal decomposition (POD) has been widely used for reduced-order modeling of fluid flows; however, its application to complex indoor airflow remains limited, particularly in systems with strong sensitivity to boundary conditions and regime-dependent behavior. This study considers impinging jet ventilation (IJV), where the flow pattern is governed by the balance between momentum and buoyancy forces. Two POD-based approaches are investigated for predicting indoor temperature fields. The first approach employs radial basis function (RBF) to predict POD coefficients from operating parameters (POD-RBF), enabling rapid prediction under varying inlet velocities and heat fluxes. The second approach reconstructs the temperature field from sparse sensor measurements using linear stochastic estimation (LSE). The effects of sensor number and placement strategies, including random, QR-pivoting, and K-means methods, are systematically examined. Computational fluid dynamics (CFD) simulations are used to generate training and validation datasets. The results demonstrate that POD-RBF can efficiently predict the flow field under known boundary conditions. Furthermore, POD-LSE enables reliable flow-field reconstruction from a limited number of optimally placed sensors. Nevertheless, both approaches show reduced accuracy in the transitional flow regime. These findings highlight the complementary roles of parameter-based prediction and sensor-based reconstruction.