Machine learning predictions and optimization for thermal energy storage in cylindrical encapsulated phase change material

Abstract

Accurate prediction of melting time is crucial in designing Thermal Energy Storage (TES) systems based on cylindrically encapsulated Phase Change Materials (PCMs). The melting time of a cylindrical encapsulated PCM directly correlates with the energy stored in the system. This study introduces a precise prediction model for the total melting time of cylindrically encapsulated PCM, utilizing a machine learning algorithm. The model, developed with the Multilayer Perceptron (MLP) method, demonstrated superior performance compared to the correlation equation proposed in the literature. The Mean Absolute Percentage Error (MAPE) value for the correlation equation was 16.68%, while the MLP model achieved a significantly lower MAPE of 4.07%, indicating its success in capturing the intricate relationship between input parameters and melting time. Furthermore, optimization results using the Non-dominated Sorting Genetic Algorithm II (NSGA-II) underscore the importance of striking a balance between stored energy and power during the design process. Maximizing stored energy (81.78 kJ) minimizes power (12.69 W), and vice versa, maximizing power (73.38 W) minimizes stored energy (37.10 kJ). In the case of equal weighting for stored energy and power in the design (56.05 kJ and 38.89 W, respectively), a 31.5% decrease in energy and a 206.5% increase in power were observed compared to the scenario where energy is maximized. Additionally, a 44% decrease in power and a 51.1% increase in energy were noted compared to the case where power is maximized. These findings collectively highlight the robustness and effectiveness of the developed MLP model in accurately predicting melting time and providing optimal solutions for energy storage parameters.