Estimation of activity coefficient of aqueous ionic liquids using a machine learning method: The artificial neural network coupled with group contribution approach

Abstract

In this study a Machine Learning (ML) approach based on the artificial neural network (ANN) coupled with Group Contribution (GC) has been developed to estimate the mean ionic activity coefficient (MIAC) of aqueous quaternary ammonium salts (QASs). ML is a subset of Artificial intelligence (AI) that focuses on the development of algorithms that allow computers to learn from and make decisions based on data. Prediction of the activity coefficient of QAS using a thermodynamic model such as the equation of state (EoS) or empirical models at high concentrations is difficult. Therefore, the ML methods are a good alternative to estimate the MIAC of the new-designed QASs. In this work, 31 aqueous QASs and 880 data points have been collected to develop the ANN network. The collected data have been divided into three subsets; 620 training, 130 testing, and 130 validation data points. The critical temperature (T_c), critical volume (V_c), acentric factor (ω), molality, and molecular weight of QASs have been considered as input layer. The T_c, V_c, and ω have been estimated using the group contribution (GC) methods. Therefore, the input variables can be estimated using the molecular structure of ionic liquids. The training correlating coefficient (R²), and the training performance (MSE) have been obtained 0.9994 and 0.0027, respectively. The results of the ANN + GC method have been compared to the Pitzer, e-NRTL, and ePC-SAFT models. The average ARD value of the ANN + GC model is lower than the aforementioned models. This work shows that the ANN + GC approach can be utilized as a robust model for the estimation of the activity coefficient of aqueous ionic liquids up to high concentrations. It must be noted that, the ANN + GC model can be used to predict the MIAC of new-designed QASs in the absence of experimental data.