Researchers Jun Liu, Qiumin Qi, Yuquan Ma, Jiwei Zhang, Qin Gu, and Shouxiao Ma have developed an intelligent prediction model designed to jointly analyze efficiency and operational costs in concentrating solar thermal storage systems. The time-attention-enhanced model, designated TAE-ECM, was designed to address long-standing modeling challenges in concentrating solar facilities. Conventional modeling methods struggle to represent the coupled evolution of thermal efficiency, operating cost, and operating state because thermal accumulation, delayed cost transmission, and system state switching occur across distinct time scales.
To capture these multi-scale temporal dynamics, the TAE-ECM framework incorporates multi-scale temporal encoding to extract feature sets from operating data. The encoding process integrates meteorological conditions, thermal variables, charging and discharging states, storage conditions, and cost parameters. A temporal-attention mechanism is then applied to identify influential historical intervals, directing the model toward the specific operational periods that govern future performance.
The architecture relies on a shared dual-task head that simultaneously predicts thermal-storage efficiency and unit effective heat-release cost. To maintain physical plausibility throughout these dual predictions, the framework enforces bounded outputs alongside an energy-balance feasibility check. Additionally, the system incorporates a rolling residual-feedback scheme that supplies incremental corrections when distribution shifts occur during ongoing plant operation.
The authors evaluated the prediction model against an operational dataset spanning 2023 and 2024. Across this dataset, TAE-ECM achieved a root-mean-square error of 0.0189 and a mean absolute percentage error of 1.71 percent for thermal-storage efficiency. For unit effective heat-release cost, the model recorded a root-mean-square error of 2.37 CNY/GJ and a mean absolute percentage error of 3.31 percent, achieving an overall composite score of 0.927.
The evaluation demonstrated the effectiveness of the rolling residual-feedback mechanism, which reduced efficiency root-mean-square error by 28.22 percent and decreased cost prediction error by 24.11 percent. Overall results indicate improved prediction accuracy and stability across diverse operational conditions, although the authors identified state-switching periods as the most challenging operating regime.
Subject of Research: Solar thermal storage efficiency and operating cost prediction
Article Title: Construction and optimization of a time-attention-enhanced intelligent prediction model for correlation analysis of solar thermal storage efficiency and cost
Article References: Liu, J., Qi, Q., Ma, Y., Zhang, J., Gu, Q., & Ma, S. (2026). Construction and optimization of a time-attention-enhanced intelligent prediction model for correlation analysis of solar thermal storage efficiency and cost. Discover Applied Sciences. https://doi.org/10.1007/s42452-026-09815-3
Image Credits: AI Generated
DOI: 10.1007/s42452-026-09815-3
Keywords: concentrating solar thermal, thermal energy storage, time-attention mechanism, dual-task learning, efficiency prediction, operating costs, residual feedback, predictive modeling, Construction, optimization, time-attention-enhanced, intelligent
News Source: Denise Maddox. (November 8, 2026). Time-Attention Model Enhances Solar Thermal Storage Efficiency and Cost Predictions. Scienmag.



