Online calculation and prediction methods for hydrodynamic characteristics of boiler water-cooled wall

The hydrodynamic safety of boiler water-cooled wall directly dictates the stable operation of coal-fired power units. To achieve real-time online monitoring of the hydrodynamic characteristics of the entire water-cooled wall, two prediction models are proposed in this study. The first model (Prediction Model A) directly predicts the hydrodynamic characteristics, while the second model (Prediction Model B) first predicts the heat flux density and then calculates the hydrodynamic characteristics. The dataset originates from coupled hydrodynamic calculations and numerical simulations, which is then trained employing the Multilayer Perceptron (MLP) models. The modularized hydrodynamic calculation model, which can achieve automatic convergence without manual intervention, adopts the gradient projection method to adjust the flow distribution in parallel tubes. Model A achieves test root mean square errors (RMSEs) of 0.6 K and 30.61 kJ kg−1 for temperature and enthalpy predictions in the spiral section, respectively. The vertical section sees higher errors (7.0 K and 55.15 kJ kg−1). However, all test Pearson coefficient (r) values exceed 0.86. Model B outperforms Model A in terms of prediction accuracy. Compared with Model A, the test RMSEs of temperature, enthalpy, and quality prediction in the vertical section of Model B are reduced by 30.0%, 38.2% and 47.6%, respectively, with all test r ≥ 0.90. Nevertheless, Model A can still provide an engineering-acceptable accurate estimation of the trend of actual values. The average prediction time of Model A is 0.43 s per prediction, making it suitable for rapid prediction scenarios. Model B requires 254.61 s, ideal for steady-state accurate calculations.