Development of hybrid computational model for simulation of heat transfer and temperature prediction in chemical reactors

Computational modeling based on heat transfer was developed to model temperature distribution in a liquid-phase chemical reactor. The model is hybrid by combining heat transfer and machine learning in simulation of the process. The simulated process is a tubular chemical reactor for synthesis of a target compound. CFD (Computational Fluid Dynamics) was employed for the simulations and linked to machine learning for advanced modeling. The models under investigation include Bayesian Ridge Regression, Support Vector Machine, Deep Neural Network, and Attention-based Deep Neural Network. Hyper-parameter optimization is carried out using the Jellyfish Swarm Optimizer to enhance model performance. The Bayesian Ridge Regression model exhibited a commendable performance with a score of 0.86039 using R² metric. The Deep Neural Network model, showcasing exceptional predictive accuracy, obtained an outstanding R² score of 0.99147. It was indicated that machine learning integrated CFD is a useful method for simulation and optimization of liquid-phase chemical reactors.