A physics-informed subdomain proper orthogonal decomposition framework for rapid and accurate multiphysics field reconstruction in gas-fired boilers

Efficient reconstruction of three-dimensional multiphysics fields—including temperature, velocity, species concentration, and pollutant formation—is essential for monitoring the operational status of energy systems. However, direct measurements are typically sparse, while high-fidelity Computational Fluid Dynamics (CFD) simulations remain computationally prohibitive. To address these challenges, this study proposes a physics-informed subdomain Proper Orthogonal Decomposition (POD) framework for the rapid and accurate reconstruction of multiphysics fields in industrial gas-fired boilers. First, a surrogate modeling strategy is employed to map operating parameters to POD coefficients. Six surrogate models, including artificial neural interpolation, Gaussian process regression, polynomial regression, radial basis function interpolation, random forests, and support vector regression, are systematically evaluated. Among them, the random forest model is selected due to its superior predictive accuracy and robustness. Second, a reduced-order modeling approach integrates POD-based dimensionality reduction with surrogate models, enabling field reconstruction within seconds. Third, a physics-informed subdomain method partitions the combustion domain according to turbulence intensity and adaptively determines the optimal number of POD modes based on local energy thresholds. Validation on a gas-fired boiler demonstrates that the proposed framework achieves high reconstruction accuracy, outperforming both global and uniformly partitioned approaches. Overall, this framework provides a practical engineering solution that significantly enhances computational efficiency while further reducing reconstruction errors.