Predicting solid–solid phase transition of quaternary ammonium salts by machine learning

Abstract

Solid–solid phase change is the key to energy storage technology. As important solid–solid phase change materials (SS-PCMs), quaternary ammonium salts, provide a variety of options for the development of SS-PCMs with different properties due to their diverse molecular structures. However, the relationship between the molecular structure of quaternary ammonium salts and their solid–solid phase change behavior is unclear. This study investigates the effect of three structural factors: type of anion, length and number of n-alkyl chains on the solid–solid phase transition behavior of quaternary ammonium salts. It is found that the ability of quaternary ammonium salts to undergo solid–solid phase transition is not determined by a single structural factor, but is influenced by a synergistic effect of multiple factors, which makes the prediction of their phase-transition behavior extremely difficult. In order to accurately predict the solid–solid phase transition behavior of quaternary ammonium salts, a prediction model based on a machine learning algorithm was constructed. Three different machine learning models: support vector machine (SVM), random forest (RF) and deep neural network (DNN) were used to analyze the dataset. By comparing the performances of the models, SVM was finally identified as the optimal solution with an accuracy of 0.9524 in predicting whether solid–solid phase transition can occur in quaternary ammonium salts. This study provides an efficient and accurate method to predict whether unknown quaternary ammonium salts possess solid–solid phase change capability. This is valuable in guiding the design and development of new high-performance SS-PCMs.