Repositori Obert UdL
Publication
Contribución de congreso (ConferenceOutput)
Improved modeling of thermal energy storage tanks state of charge Using LSTM neural networks and image segmentation
This study presents a deep learning framework designed to accelerate the modeling of sensible heat thermal energy storage (TES) tanks, addressing the high computational costs associated with traditional computational fluid dynamics (CFD). By integrating long short-term memory (LSTM) neural networks with image segmentation, the authors developed a method to predict the thermal state of charge across various tank configurations characterized by different internal arrangements of solid storage spheres. A dataset comprising over 7 million thermal images from 2D CFD simulations was used to train and compare several architectures, including Convolutional Autoencoders and Inception-based models. The results demonstrate that an Autoencoder-LSTM architecture, particularly when segmentation data is introduced after the recurrent layer, yields the highest accuracy with a Structural Similarity Index (SSIM) of 86.7%. While the approach offers a rapid and viable alternative for evaluating diverse tank geometries, the study also identifies challenges regarding cumulative errors in long-term sequence predictions, suggesting future directions for optimizing AI-driven thermal analysis., This project was funded by the European Union Horizon Europe Research and Innovation Pro-gramme under grant agreement 101103552 (SUSHEAT). Views and opinions expressed are, however, those of the author(s) only and do not necessarily reflect those of the European Union or CINEA. Neither the European Union nor the granting authority can be held responsible for them. This paper is part of the RYC2023-044196-I, funded by MCIU/AEI/10.13039/501100011033 and FSE+. The authors thank the Generalitat de Catalunya for the quality accreditation granted to their research group (2021 SGR 01615). GREiA is a cer-tified agent TECNIO in the category of technology developers of the Government of Catalonia. This work is partially supported by ICREA under the ICREA Academia programme.