3.6 Comparison results with existing methods
In Table 2, the ACGC models are compared with the classical group contribution methods (Joback: first-order group technique, C-G: second-order group technique, M-G: third-order group technique, M-G+: third-order group technique) and S-S (Sun and Sahinidis: third-order group technique). The groups in Joback method describe part of atomic adjacent relationship, and the AAG method is more refined than the Joback method, which is theoretically more accurate than the Joback method. And the groups we have divided are guaranteed to appear in more than five compounds, thus the ARE for the critical temperature looks higher than the Joback method. The ARE ofV c in ACGC model was 1.34%, which was lower than that of Joback (2.30 %), C-G (1.79 %), M-G (1.80 %), M-G+ (2.05 %) and S-S (3.10 %). These results show that the ACGC models are accurate. The ARE of critical properties calculated by ACGC method is lower than M-G+ method and S-S method. Although ARE of the T c ACGC model is higher than that of the C-G and M-G methods, we screened the data so that the number of occurrences of groups in the data set is not less than 5 times, so the model is more reliable. In addition, this work aims to develop reliable models with good predictive ability, using 20 % of the data set as test sets and the rest as training sets to fit the property model, which is different from the traditional group contribution method.
Table 2 Comparison of ACGC with GC methods.