Flame speed prediction and combustion characteristics of hydrogen-enriched methane jets: A POD-mixed-BPNN framework for CFD flame dynamics

With the rising demand for hydrogen energy, blending hydrogen into methane pipelines presents a promising strategy. This study develops a Reduced-Order Model (ROM) framework based on Proper Orthogonal Decomposition (POD) via Singular Value Decomposition (SVD) to predict jet flames under high Hydrogen Doping Ratios (HDR). Key parameters such as HDR, equivalence ratio, and nozzle radius are considered. Three interpolation methods—POD-linear, POD-spline, and POD-RBF—are compared with a Back Propagation neural network (BPNN) for temperature prediction. While POD-linear achieves good performance (runtime <5.4 ms, error <6.51 %) and even outperforms BPNN under smooth parametric transitions, the BPNN improves overall accuracy (<4.75 %) and reduces computation time to 1.74 ms. To further enhance generalization and robustness under limited samples, a Mixed-BPNN is proposed, reducing the prediction error from 4.75 % to 1.34 %, while slightly increasing the prediction time to 2.35 ms.