Prediction of Eutectic Composition and Melting Point of Reciprocal Eutectic using Principal Components based Feed Forward Neural Network (PCA-FFNN)

The current study presents a novel technique incorporating Artificial Neural Network and dimensionality reduction (Principal Component Analysis) for predicting the two thermophysical properties of the binary fluoride/chloride reciprocal eutectic salt systems namely the composition and melting point (MP) temperature of the eutectic system. This model considers 12 molecular and atomic parameters to compute the composition and melting point of binary fluoride/chloride reciprocal eutectic salt systems. The Principal Component Analysis-Feed Forward Neural Network methodology demonstrated enhanced prediction accuracy, with a mean root mean squared error of 5.533 for melting point and 1.329 for eutectic composition, as assessed by the leave-one-out cross-validation method (LOOCV). Further, the R2 values of the melting point and eutectic composition, for the models with the least RMSE on test data, were 97.69 % and 97.26 % respectively. This modelling technique offers the potential to forecast the composition and melting points of multi-component reciprocal eutectic salt systems, and also to ascertain the other properties of reciprocal eutectic salt systems, such as their densities, enthalpies, conductivities, and so on.