Bottom-cut optimization with multilayer perceptron prediction for melting performance in horizontal latent heat storage units

To address the issues of a slow melting rate and the accumulation of low temperature phase change material (PCM) at the bottom of horizontal thermal energy storage (TES) units, this study proposes a novel bottom-cut optimization strategy. A numerical model is developed and validated experimentally for the horizontal TES unit with various bottom-cut ratios (H/R from 0.28 to 1.0) is developed and validated experimentally. The effects of the cut ratio on melting time, melting front evolution, temperature field, velocity field, and Nusselt and Grashof numbers are analyzed. The results indicate that reducing the cut ratio optimizes the thermal path in the upper region, enhancing natural convection and thermal uniformity. Compared with the circular unit, the unit with a cut ratio of 0.28 reduces the complete melting time from 28,310 s to 7380 s, representing a decrease of 73.93%. Meanwhile, the average Nusselt number and average Grashof number increase by 155.45% and 75.95%, respectively, indicating enhanced convective heat transfer intensity. Furthermore, a multilayer perceptron (MLP) model is trained on 1135 numerical datasets to rapidly predict the melting performance. The model achieves relative errors within ±10% on both the training and testing sets. This work offers an efficient approach for designing high-performance TES devices through bottom-cut optimization combined with machine learning for rapid prediction.