Critical heat flux prediction through machine learning model for narrow rectangular channels

Accurately predicting critical heat flux (CHF) is crucial for ensuring efficient energy transfer and safe operation in thermal system, particularly in narrow rectangular channels, which are favored for their superior thermo-hydraulic performance and compactness. This study compiles 660 experimental CHF data points from narrow rectangular channels under wide variety of operating conditions and evaluates existing CHF correlations. Four machine learning (ML) models—back-propagation (BP) neural networks, random forest (RF), support vector regression (SVR), and Long Short-Term Memory (LSTM) are employed to predict CHF using input parameters encompassing thermal-hydraulic and geometrical characteristics. The results indicate that the BP approach has a high level of performance, with a root-mean-square error (RMSE) of 38.37 and a mean absolute error (MAE) of 24.33. To further optimize the BP neural network, four metaheuristic algorithms—particle swarm optimization (PSO), genetic algorithms (GA), cuckoo search (CS), and ant colony algorithm (ACA) are applied. The GA optimization yields the most accurate predictions, achieving a RMSE of 33.53 and a MAE of 19.13. The trained GA-BP model exhibits robust performance in predicting CHF in narrow rectangular channels under varying pressure conditions.