Operational performance and application potential of a seasonal ice storage cylinder cooling system in cold climate based on experimental investigation and deep learning

Seasonal cold storage is a method for providing energy-efficient summer cooling in buildings. This study proposes a novel seasonal ice storage cylinder (SISC), which enables charging without external energy input by directly exchanging heat with ambient cold air through convection. An experimental platform was established to evaluate the performance of the SISC under different operational stages. A transient mathematical model of the SISC-based building cooling system was developed based on thermodynamic principles and energy conservation laws. Four deep learning models were employed to predict the cooling output performance of the SISC system. The results indicate that the ice thickness in the SISC reached 600 mm at −11.7 °C during 63-day cold air convection charging stage, with a corresponding cold energy loss of 24.8 % during 64-day storage stage. The SISC system can reduce energy consumption by 30–60 %, with a payback period of about 7.62–11 years, showing considerable potential for practical application in terms of energy efficiency and operational performance. The MAE and RMSE of the CNN-LSTM model are at least 20.5 % and 25 % lower than those of other single models, and 1.84 % and 8.8 % lower than those of others hybrid models, respectively. The hybrid deep learning model can accurately capture the nonlinear variations in the SISC system's cooling output.