Nhảy đến nội dung

Rheological properties of SWCNT/EG mixture by a new developed optimization approach of LS-Support Vector Regression according to empirical data

Authors: 

Jalal Alsarraf, Seyed Amin Bagherzadeh, Amin Shahsavar, Mahfouz Rostamzadeh, Phan Van Trinh, Minh Duc Tran*

Source title: 
Physica A: Statistical Mechanics and its Applications, 525: 912-920, 2019 (ISI)
Academic year of acceptance: 
2019-2020
Abstract: 

Present work aims to introduce a new novel method of Support Vector Regression as a substitute for Artificial Neural Network to predict nanofluid properties, for the first time. Then its performance is evaluated according to the empirical results of SWCNT/EG versus temperature and concentration. Hence two LS-SVM and ANN models are trained to estimate the dynamic viscosity of nanofluid made of single-wall carbon nanotubes in ethylene glycol in terms of the temperature (T = 30 to 60 °C) and solid concentration (a to 0.1%). The results indicate that the precision of the LS-SVM and ANN models are comparable; nevertheless, the LS-SVM generalization is much better than the ANN. This is due to the fact that the LS-LSM models have a less number of parameters in comparison with the ANN. Therefore, the LS-LSM is more resistant to overfitting than the ANN, especially in handling small-size datasets. Hence, the LS-SVM may be a more reliable method for function estimation problems with small-size datasets.