Hydrodynamic characteristics prediction models for water-cooled wall under different loads based on Artificial neural network

Real-time monitoring of hydrodynamic characteristics in water-cooled wall is crucial for boiler safety. Comprehensive numerical simulations are employed in this study to determine the heat flux distribution on the water-cooled wall across 38 different operating conditions. The heat flux results are applied to calculate hydrodynamic characteristics. Subsequently, the hydrodynamic characteristics results, along with the corresponding operating parameters, form the dataset for the Artificial Neural Network (ANN) models. The proposed methodology generates high-quality datasets, with a maximum root mean square error (RMSE) of only 6.8 K when comparing the results of working fluid temperature to the measured values. Two dataset partitioning methods are compared. Compared with random partitioning (control group), Considering each working condition as a whole during dataset partitioning (experimental group) rises the average correlation coefficient (r) and coefficient of determination (R2) of predicted results in the test set by 17.88 % and 6.48 %, respectively, along with a 31.8 % decrease in average number of neurons. The developed models exhibit excellent agreement with four working conditions in the test set, both in trends and absolute values, with acceptable error ranges. On the validation and test sets, the average values of r for flow rate, pressure, temperature, and enthalpy are 0.9024, 0.9904, 0.9528, and 0.8609, separately. The average values of R 2 for these variables are 0.8956, 0.9932, 0.9525, and 0.8548, respectively, underscoring the reliability and practicality of the predictive models.