Modeling the activity coefficient at infinite dilution of water in ionic liquids using artificial neural networks and support vector machines
| Autor: | Hania Benimam, Maamar Laidi, Cherif Si-Moussa, Salah Hanini |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Activity coefficient
0209 industrial biotechnology Artificial neural network Computer science Extrapolation 02 engineering and technology Least squares Dilution Support vector machine Nonlinear system chemistry.chemical_compound 020901 industrial engineering & automation chemistry Artificial Intelligence Ionic liquid Least squares support vector machine 0202 electrical engineering electronic engineering information engineering Feedforward neural network 020201 artificial intelligence & image processing Sensitivity (control systems) Biological system Software Interpolation |
| Zdroj: | Neural Computing and Applications. 32:8635-8653 |
| ISSN: | 1433-3058 0941-0643 |
| DOI: | 10.1007/s00521-019-04356-w |
| Popis: | The activity coefficient at infinite dilution of water in ionic liquids is a thermodynamic property of a paramount importance in separation processes. However, accurate modeling of this parameter remains a challenging task due to the highly nonlinear behavior of the water/ionic liquid systems. Also, available models consider a large number of inputs that are usually difficult to access and require complicated use techniques. Therefore, the main objective of this paper is to use artificial intelligence techniques to propose models (based on a reduced number of inputs that are easily accessible, and to improve the accuracy of the correlative performance for activity coefficient at infinite dilution of water in ILs). The present work features the application of artificial neural networks, support vector machine and least square support vector machine, among data-driven methods, for modeling the activity coefficient at infinite dilution of water in 53 ionic liquids. Overall, the models proposed are able to accurately correlate 318 experimental data points gathered from the literature. According to the results, the ANN is more powerful and effective computational learning machine than the two remaining ones. The correlation coefficients R2 and deviations expressed as an average absolute relative deviation for the neural network model are estimated to be 0.99997 and 0.56%, respectively. Furthermore, the neural network model’s interpolation and extrapolation capabilities are demonstrated, and its accuracy is compared to other proposed models in the literature based on multi-linear regression, least squares support vector machine and another feedforward neural network. This work also includes a graphical user interface for the proposed model, as well as an inputs’ sensitivity analysis. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full text from SpringerLink