Repositori Obert UdL
Publication
Artículo científico (JournalArticle)
Predicting state of charge in latent thermal energy storage systems using artificial intelligence modelling techniques
Accurate state of charge (SoC) estimation in latent thermal energy storage (latent TES) systems is critical for optimizing renewable energy integration and ensuring efficient energy storage operation. Traditional physics-based methods, though accurate, were computationally intensive and required complex parameter calibration, limiting their applicability for real-time applications. This study introduced an artificial intelligence (AI) framework for real-time SoC estimation in shell-and-tube latent TES systems, integrating long short-term memory recurrent neural networks (LSTM-RNN) and extreme gradient boosting (XGBoost) through an optimised weighted ensemble (0.3 LSTM-RNN, 0.7 XGBoost), with weights determined using grid search. The framework was trained and validated on 605 charging and discharging profiles from a previously experimentally validated model, significantly expanding prior studies. The dataset covered a broad range of operating conditions, including charging (140–150 °C) and discharging (100–110 °C) cycles with mass flow rates between 0.8 and 1.2 kg/s. The ensemble model achieved superior accuracy (RMSE: 1.7%, MAE: 0.9%, R2 >99.7% for charging; RMSE: 1.4%, MAE: 0.8%, R2 >99.8% for discharging) compared to individual LSTM-RNN and XGBoost models, while reducing computational costs by 80%, enabling inference within 2.6 ms per sample. This significant reduction in computational requirements transforms SoC estimation from a computational bottleneck into an enabling technology for real-time control integration, allowing direct deployment in industrial control systems where traditional physics-based methods remain impractical due to their computational demands., This study has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement COCOOL (101007976). This project was also funded by the European Union’s Horizon Europe Research and Innovation Programme under grant agreement No. 101096921 (THUMBS UP). 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. The authors also thank the Generalitat de Catalunya for the quality accreditation granted to their research group (2021 SGR 01615). GREiA is certified 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. This work was partially funded by the Ministerio de Ciencia e Innovacion ´ - Agencia Estatal de Investigacion ´ (AEI) (PID2021-123511OB-C31 - MCIN/AEI/10.13039/ 501100011033/FEDER, UE). This work was partially funded by Ministerio de Ciencia, Innovación y Universidades - Agencia Estatal de Investigación (AEI) (RED2024-153629-T). Professor A Sciacovelli also acknowledges the Novo Nordisk Foundation for the funding received (AI-LoDES, NNF25OC0100591)