Prediction of liquidus temperature for complex electrolyte systems Na3AlF6-AlF3-CaF2-MgF2-Al2O3-KF-LiF based on the machine learning methods

Abstract

Liquidus temperature is an important physicochemical property parameter of electrolyte system. It is significant to construct a practical model to predict the liquidus temperature of electrolyte with universal adaptability, since the current formulas is unsatisfactory with many limitations. In this work, different kinds of machine learning algorithms were used to explore the correlations of the liquidus temperature with the components of electrolyte. It was found that the performance of SVM(Support vector machine) model and GBR(Gradient boosting regression) model were better than those of PLS (Partial least squares regression), MLR(Multiple linear regression), ANN(Artificial neural network) and RFR(Random forest regression) for predicting liquidus temperatures with the component of electrolyte. Furthermore, the optimal models were applied to 20 independent electrolyte samples to predict the liquidus temperatures. The predicted results based on SVM model is also much better than that of the linear formula reported in the literature and GBR model. The SVM model outlined here can provide the accurate prediction method of the liquidus temperature for industrial complex electrolyte, it is crucial for balance and stability of the electrolytic cell in the process of electrolytic aluminum production.