Use of group contribution method and intelligent algorithms to predict the flash temperature of binary mixtures

Abstract 513 flash point temperatures of 42 different binary mixtures were collected from various sources. Two soft-computing models of Artificial Neural Network (ANN) and optimized Adaptive Neuro-Fuzzy Inference System with Genetic Algorithm (ANFIS-GA) were established to predict these flash temperatures. Mole fraction of component 1, flash temperature of each component and Van der Waals R and Q for each component constitute the 5 input variables of both models. It was tried to improve the predictions of these two models using the group contribution method (GCM) based on the functional groups of the UNIQUAC activity equation. The mixing parameters of 215 binary data were first calculated and then the results were extended to the entire mixture data. Two new parameters of A12 and A21 based on the GCM were also added to each model to form two new 7 input-variable models. The results of the new models were compared with the previous results and it was observed that the % AARD of the ANN and ANFIS-GA models were reduced from 0.48% and 1.8% to 0.36% and 1.7%, respectively.