Comparison between Cuckoo Search, Genetic Algorithm and Particle Swarm Optimization Algorithms for Multi-Component Systems

In order to confirm our capability with CS, a comparison with GA and PSO algorithms was made for quaternary and quinary systems. Three quaternary systems, Ethanol + Water + Pentane + Hexane; Ethanol + Water + Pentane + Cyclohexane and Ethanol + Water + Hexane + Cyclohexane and one quinary system, Ethanol + Water + Pentane + Hexane + Cyclohexane, as studied by Khansary and Sani (2014) were selected for the same. The experimental data for the above-mentioned systems has been taken from Huang, Chung, Tseng and Lee (2010).

For each system, 20 trials have been carried out and the lowest RMSD along with the corresponding interaction parameters was selected as the final result and reported in Tables 7.19 and 7.20. For the quaternary system, CS gave RMSD (%) values in the range 0.14%-0.44% against ~1.0% for the GA and the PSO algorithm (Table 7.19), whereas for the quinary system, CS gave RMSD (%) values ~0.85% against ~2.0% for GA and PSO (Table 7.20). This clearly shows that the CS algorithm is a reliable tool to correlate the LLE data and in some cases, it is even superior to the GA and the PSO algorithm.

TABLE 7.14

Effect of the Cation Type on the Success Performance (%SR) of CS with the UNIQUAC Model for Selected Ternary Systems

Tolerance (e)

Iter max

SYS-3

SYS-4

1.00E-03

200

13.3

16.7

500

66.7

90.0

1000

100.0

96.7

1500

100.0

96.7

2000

100.0

100.0

1.00E-04

200

0.0

0.0

500

0.0

6.7

1000

6.7

46.7

1500

53.3

86.7

2000

73.3

100.0

1.00E-05

200

0.0

0.0

500

0.0

0.0

1000

0.0

10.0

1500

16.7

53.3

2000

30.0

90.0

Notes: SYS-3: [BMIM][TFO] + Ethanol + ETBE; SYS-4: [EMIM] [TFO] + Ethanol + ETBE.

In summary, based on %SR and RMSD analyses, it can be concluded that the CS algorithm gave a good result in terms of both SR and the number of iterations, especially for the UNIQUAC model. LLE systems containing imidazolium ILs appear to be challenging for PE for higher precision. Further studies should be focused on the performance improvement of the CS algorithm to get result with good precision using low numerical effort. Thus, it can be concluded that the UNIQUAC and the NRTL models, with interaction parameters estimated by the CS algorithm, were able to correlate the LLE data successfully.

 
Source
< Prev   CONTENTS   Source   Next >