PREDICTION OF PCME'S THERMAL BEHAVIOR USING A DEEP NEURAL NETWORK

Phase Change Material Emulsions stand out as potential latent thermal fluid in different thermal applications. This paper presents results concerning the prediction of the heat transfer behaviour of a 30 wt.% paraffin in water emulsion for a temperature range of 0-20 degrees C. Based on data obtained empirically, we developed a deep neural network to predict the PCME's heat transfer coefficient. The artificial neural model was developed using the most complex and new scientific and statistical tools - TensorFlow, Keras, Pandas and Python programming language. A complex statistical study was performed a priori to model's development. For comparisons, the model was first trained with 24 features (comprehensive model) and then with only 5 (lumped model), as they were the most statistically relevant. The two models have similar prediction mean squared errors (around 5% for the comprehensive model and around 6% for the lumped model), but the full model tends to converge faster (used only 30 epochs compared with 130). Both models showed very good prediction capabilities on new and unseen data: the comprehensive model predicted with only 5.0% error, while the lumped model had a mean squared error equal to 6%.