Modeling of TES Tanks by Means of CFD Simulation Using Neural Networks

Modeling of thermal energy storage (TES) tanks with computational fluid dynamics (CFD) tools exhibits limitations that hinder the time, scalability, and standardization of the procedure. In this study, an innovative technique is proposed to overcome the challenges in CFD modeling of TES tanks. This study assessed the feasibility of employing neural networks for TES tank modeling, evaluating the similarities in terms of structure and signal-to-noise ratio by comparing images generated by neural networks with those produced through CFD simulations. The results regarding the structural similarity index indicate that around 94% of the images obtained have a similarity index above 0.9. For the signal-to-noise ratio, the results indicate a mean value of 25 dB, which can be considered acceptable, although indicating room for improvement. Additional results show that our neural network model obtains the best performance when working with initial states close to the stable phase of the TES tank. The results obtained in this study are promising, laying the groundwork for a future pathway that could potentially replace the current methods used for TES tank modeling.