Thermal analysis of PCM magnesium chloride hexahydrate using various machine learning and deep learning models

Abstract

The heat storage through Phase change material (PCM) is important for sustainable development. Previously a set of experiments were carried out on 14.5 kg of PCM, Magnesium chloride hexa hydrate in a shell and tube heat exchanger. The results are used to build machine learning and deep learning models such as Support vector machine, Random forest, Extreme gradient boost, Catboost and Artificial neural network to replicate the experiment virtually. The input variables are heat transfer fluid’s inlet and outlet temperature and time elapsed while PCM’s temperature is predicted in both charging and discharging cycles. The hyperparameter tuning is done using hyperband. Catboost model outperformed all other models with highest variance in both the cycles (0.9945 and 0.9299) and highest score: 0.993 and 0.9466. It has got the lowest MAPE (0.2% and 0.1 %). Hence this model can be recommended over these other models for doing system simulation of problems involving both charging and discharging cycles of the PCM.