Optimizing Gaussian process regression (GPR) hyperparameters with three metaheuristic algorithms for viscosity...

Area of Science:

• Materials Science
• Chemical Engineering
• Thermodynamics

Background:

• Microencapsulated phase change materials (MPCMs) are vital for thermal energy storage (TES) systems.
• Applications span building materials, textiles, and cooling technologies.
• Accurate prediction of dynamic viscosity is essential for optimizing TES performance.

Purpose of the Study:

• To accurately predict the dynamic viscosity of MPCM and MXene particle suspensions.
• To evaluate the effectiveness of Gaussian Process Regression (GPR) for this prediction task.
• To optimize GPR hyperparameters using metaheuristic algorithms.

Main Methods:

• Gaussian Process Regression (GPR) was employed for dynamic viscosity prediction.
• Twelve GPR hyperparameters were analyzed and categorized by importance.
• Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Marine Predators Algorithm (MPA) were used for hyperparameter optimization.

Main Results:

• Optimizing the four most significant GPR hyperparameters yielded R-values around 0.9983 with all tested algorithms.
• Including moderately significant hyperparameters improved models, with PSO achieving an R-value of 0.99834.
• Comprehensive optimization of all twelve hyperparameters using GA resulted in the highest accuracy (R-value of 0.999224).

Conclusions:

• Metaheuristic-optimized GPR provides a highly accurate method for predicting the dynamic viscosity of MPCM and MXene suspensions.
• The developed models offer a cost-effective and efficient alternative to extensive laboratory testing.
• This approach supports the development and optimization of various thermal energy storage and management systems.