A hybrid model for packed bed thermal energy storage system

Physical models for Packed Bed Thermal Energy Storage (PBTES) system plays a crucial role in predicting its dynamic behavior and long-term performance under various operational conditions. Nevertheless, the prevailing physical models struggles to balance between computational efficiency and accuracy. This paper introduces a novel hybrid model for PBTES that combines the strengths of numerical modeling and machine learning, aiming to address this challenge. The proposed model includes a low-precision yet faster-to-solve linearized version of the state-of-the-art two-phase model, integrated with a machine learning module. In the hybrid model, the linear two phase model is solved numerically using finite volume method on a coarse mesh. Then, the ML module takes this low precision and linear solution, and maps it into a high precision non-linear solution corresponding to fine mesh. The proposed hybrid model is highly robust and generalized, and has been found to be effective in handling cases even beyond the training range of the machine learning module. During the validation process, the hybrid two-phase model delivered the non-linear solutions nearly 350 times faster than the traditional non-linear two-phase model. It demonstrated good accuracy, achieving an overall error percentage just 0.16 and a R 2 score of 0.99. The findings highlight the potential of this hybrid modeling approach for fast and accurate simulations of PBTES, suggesting its applicability in complex applications such as model control optimization and long period simulations.