An Interpretable Predictive Model for High-Gravity NO _<i>x</i> Oxidation-Absorption Driven by Physics-Informed Neural Networks

High-gravity technology offers a frontier solution for efficient NOx removal, yet its intricate internal mass transfer-reaction coupling poses significant modeling challenges, particularly for “black-box” machine learning methods that lack physical consistency. To this end, we introduce a Physics-Informed Neural Network (PINN) with a hard parameter-sharing architecture to simulate the oxidation-absorption of NOx by an alkaline H2O2 solution in a high-gravity field. By learning common physical laws through a shared base network and handling multispecies dynamics via dedicated branches, this model markedly improves convergence speed and generalization. The results demonstrate that the model achieves high-precision predictions of NOx absorption efficiency (R2 = 0.931, RMSE = 4.871%) and exhibits excellent robustness, even with sparse or noisy data. Furthermore, concentration field predictions combined with SHAP analysis reveal the underlying mechanisms: rotational speed is the dominant factor governing process efficiency by intensifying fluid shear and interfacial renewal to promote the reaction between NO and the hydroperoxide anion (−OOH), with its importance followed by H2O2 concentration, pH, and liquid flow rate. This study not only delivers an interpretable tool for predicting and optimizing the high-gravity De-NOx process but also establishes a new intelligent modeling paradigm for elucidating other complex multiphase environmental systems such as advanced oxidation processes or gas purification processes.