A computational intelligence scheme for estimating electrical conductivity of ternary mixtures containing ionic liquids

A computational intelligence scheme for estimating electrical conductivity of ternary mixtures containing ionic liquids

    A computational intelligence scheme for estimating electrical conductivity of ternary mixtures containing ionic liquids Mohsen Hossei...

591KB Sizes 40 Downloads 78 Views

    A computational intelligence scheme for estimating electrical conductivity of ternary mixtures containing ionic liquids Mohsen Hosseinzadeh, Abdolhossein Hemmati-Sarapardeh, Ameli, Fereshteh Naderi, Mohammadmahdi Dastgahi PII: DOI: Reference:

S0167-7322(16)31175-8 doi: 10.1016/j.molliq.2016.05.059 MOLLIQ 5873

To appear in:

Journal of Molecular Liquids

Received date: Accepted date:

11 May 2016 20 May 2016

Forough

Please cite this article as: Mohsen Hosseinzadeh, Abdolhossein Hemmati-Sarapardeh, Forough Ameli, Fereshteh Naderi, Mohammadmahdi Dastgahi, A computational intelligence scheme for estimating electrical conductivity of ternary mixtures containing ionic liquids, Journal of Molecular Liquids (2016), doi: 10.1016/j.molliq.2016.05.059

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT A Computational Intelligence Scheme for Estimating Electrical

PT

Conductivity of Ternary Mixtures Containing Ionic Liquids Mohsen Hosseinzadeh a, Abdolhossein Hemmati-Sarapardeh b,*, Forough Ameli c,

RI

Fereshteh Naderi d, Mohammadmahdi Dastgahi e,

Department of Chemical Engineering, Amirkabir University of Technology, Tehran, Iran

b

Department of Petroleum Engineering, Amirkabir University of Technology, Tehran, Iran

c

Chemical Engineering Department, Islamic Azad University, North Tehran Branch, Tehran, Iran

a

Department of Chemistry, Shahr-e Qods Branch, Islamic Azad University, Shahr-e Qods, Tehran, Iran

e

Department of Electrical Engineering, Dezful Branch, Islamic Azad University, Dezful. Iran

NU

SC

a

MA

Abstract: Due to unique physical and chemical properties of ionic liquids (ILs), they received

AC CE P

TE

D

lots of attention in many industrial fields and are widely under research. Ionic liquids, also are emerging as important components for applications in electrochemical devices. To develop their applications and achieving desire properties, they are usually mixed with organic solvents. Applying ionic liquids in many applications needs the accurate and reliable data of electrical conductivity of ILs and their mixtures. To this end, a total of 224 experimental data were collected from literature and divided randomly into two datasets: 179 data was selected as training set and the remained 45 data was used as a testing set. Afterwards, a reliable modeling technique is developed for modeling the electrical conductivity of the ILs ternary mixtures. This approach is called least square support vector machine (LSSVM). The model parameters were optimized using the method of couple simulated annealing (CSA). The input model parameters were, temperature of the system, melting point, molecular weight and mole percent of each component. A comprehensive error investigation was carried out, yielding the well accordance between the predictions of the model and experimental data. The presented model can predict the dependency of electrical conductivity variations with input variables. Moreover, the sensitivity analyses demonstrated that, among the selected input parameters, the average melting point of mixture has the largest effect on the electrical conductivity. Furthermore, suspected data were detected using the Leverage approach, residual, Williams plot and statistical hat matrix. Except seven data points, the all data appear to be reliable. Keywords: Ionic liquids; ternary mixture; electrical conductivity; least square support vector machine. *Corresponding

author

email

addresses:

A.

Hemmati-Sarapardeh

[email protected] )

1

([email protected]

&

ACCEPTED MANUSCRIPT 1. Introduction

PT

Ionic liquids (ILs) are known since 1914 [1], however the interest for them have been increased since 1992, when Wilkes and Zaworotko introduced new compounds with hydrolytically stable

RI

ions [2]. Due to distinctive physical and chemical properties of ionic liquids, they received lots

SC

of attention in the past decade. ILs are liquid salts with melting points normally below 373.15 K

NU

[3-5]. They are usually composed of nonsymmetrical organic cations and numerous different inorganic or organic anions. ILs have a series of outstanding properties, including large

MA

temperature range of liquid state, negligible vapor pressure, wide electrochemical windows, good solubility of organic and inorganic compounds, flame resistance, tunable nature, remarked

D

catalytic property [6-12], strong solubilization power, recyclability, nonvolatility, high thermal

TE

stability (up to 673 K), low toxicity, and conduction properties [13-18]. ILs are used as

AC CE P

nonflammable, nonvolatile ILs-based electrolytes in high temperature fuel cells and batteries [19-22]. Chemical and physical properties of ILs such as their polarity, hydrophobicity and viscosity can be selected by choosing the cationic or the anionic constituents [23]. Normally, ILs are composed of bulky, nonsymmetrical organic cations such as pyrrolidinium, pyridinium, quaternary ammonium or phosphonium, imidazolium, and numerous different inorganic or organic anions such as tetrafluoroborate, hexafluorophosphate, trifluoromethanesulfonate, bis(trifluoromethylsulfonyl)imide and bromide anions [4, 24-30]. ILs can be used as recyclable ―green solvents‖ [31, 32] as well as conductors in electrochromic devices [33]. These liquids have been used for liquid electrolytes based on aliphatic tetraalkylammonium or aromatic pyrrolidinium and isoquinolinium cations [34, 35]. Using ILs in the aforementioned applications requires knowledge of some physical and thermodynamic properties. The electrical conductivity of ILs is of paramount importance. Knowing the exact value of electrical conductivity of ILs in 2

ACCEPTED MANUSCRIPT some applications such as electrolytes is crucial. Therefore, up to now, the properties of pure ILs have been studied by a number of scientists [33, 36, 37]. For example, in 2011, Kanakubo et al.

PT

[38] have measured the densities, viscosities and electrical conductivities of N-methoxymethyl-

RI

N-methylpyrrolidinium bis(trifluoromethanesulfonyl) amide over a wide temperature at atmospheric pressure. In 2013, Tomida et al. [39] have studied the densities and thermal

SC

conductivities of N-butylpyridinium tetrafluoroborate, N-hexylpyridinium tetrafluoroboratte and

NU

N-octylpyridinium tetrafluoroborate. Then, in 2014, Makino et al. [40] have reported the densities, viscosities, and electrical conductivities of ethylimidazolium and 1-ethyl-3-

numerous advantages of ILs, there are a strong electrostatic force and

D

In spite of

MA

methylimidazolium ILs.

TE

hydrogen-bonding interaction among ions, so there is a microscopic aggregation in ILs and because of that the viscosities of ILs are always very high [41-45]. Adding less viscous

AC CE P

molecular solvents into ILs have a significant effect on the physical properties of ILs, and thus improves the large-scale applications of ILs [7, 46, 47]. Therefore, a number of scientists have studied the physical properties of binary solutions of ILs [48-52]. Study of physicochemical properties of mixtures of ILs and molecular solvent is too necessary. It provides valuable data for the application of ILs as well as excellent information about the variation of microscopic structures of those mixtures [53, 54]. Until now, most studies related to ILs are limited to pure ILs or binary mixtures. In spite of extensive studies on pure ILs properties such as densities, viscosities, conductivities, etc., little information on thermodynamic properties of ternary mixtures have been published [55]. Zhang et al. determined the conductivities and viscosities of the room-temperature IL, 1-butyl-3-methylimidazolium hexafluorophosphate ([bmim][PF6]) + water + ethanol and [bmim][PF6] + water + acetone ternary mixtures at different temperatures 3

ACCEPTED MANUSCRIPT [55]. Chen et al. measured the densities, viscosities, and conductivities of the ternary solutions [N-EMP]Br

(N-ethyl,methylpiperidinium

bromide)

+

[N-PMP]Br

(N-propyl,methyl-

PT

piperidinium bromide) +H2O, [N-EMP]Br + [NBMP] Br (N-butyl,methyl-piperidinium bromide)

RI

+H2O, and [N-PMP]Br + [N-BMP]Br +H2O and their binary subsystems at different temperatures and atmospheric pressure. Recently, Hosseinzadeh and Hemmati-Sarapardeh

SC

proposed a model for prediction of viscosity of IL ternary mixtures. The variables of this

NU

equation were selected as molecular weight, composition, and normal boiling point [56]. The viscosity of ternary mixtures including ionic liquids was calculated based on a new, prompt, and

MA

precise technique namely least square support vector machine (LSSVM).

D

In this study, we follow up our previously published work [56] and develop a computation

TE

method, namely LSSVM to predict electrical conductivities of ternary mixtures containing ILs at various temperatures and atmospheric pressure. Coupled simulated annealing (CSA) approach

AC CE P

was applied for estimation of LSSVM model parameters. Afterward, statistical and graphical error analyses are utilized simultaneously to check the performance of the developed model. In addition, trend analysis is used to examine if the model can capture the physically expected trends. Besides, sensitivity analysis is employed through the relevancy factor to deepen our understanding about the relative effect of input parameters on electrical conductivity. Finally, the Leverage approach is used to find the applicability domain of the developed model as well as to assess the quality of experimental data points.

2. Data Gathering 4

ACCEPTED MANUSCRIPT The model accuracy and robustness depend highly on precaution of the dataset used for forecasting the determined properties. To this end, our attempt is to assemble all available

PT

experimental data on the electrical conductivities of ternary mixtures containing IL data sets,

RI

over temperatures ranging from 288.15 to 308.15 K which are available in open literature sources [55, 57]. A total of 224 experimental published data points were gathered from the

SC

reliable literatures. The component of each ternary mixtures are tabulated in Table1 and the

NU

properties of the selected compounds are gathered in Table 2, which were provided form these references [58-63].

MA

The input variables of this model are selected temperature, melting point (Tf), molecular weight (Mw) of the compounds, and the composition of the two first substances. It should be noticed

TE

D

that the third component‘s mole fraction is a dependent variable, since the summation of all the mole fractions is equal to one .

AC CE P

  f (T , Mwi , T f i , X 1 , X 2 )

(1)

The developed model based on these data could be reliable and efficient for predicting electrical conductivity of other ternary mixtures containing IL.

3. Model development The support vector machine (SVM) is a proper mathematical tool to develop nonlinear relationships among the available experimental data considered as inputs of the model and the desired output. This theory which has been applied in many fields, mainly has been discussed in computational science for a set of related supervised learning methods that analyze data and recognize patterns and are used for regression analysis [64-66]. Among the large number of machine learning methods which have already been used to solve a wide range of difficulties in 5

ACCEPTED MANUSCRIPT science and engineering, artificial neural networks (ANNs) have yielded highly accurate results [67]. However, ANN has lost its favor after appearance of SVM [68, 69]. Moreover, ANN

PT

usually encounters over-fitting or under-fitting issues due to their empirical risk minimization

RI

characteristics; however, such possibilities have been minimized in SVM paradigm by incorporating a structural risk minimization strategy. As a result, SVM can overcome several

SC

shortage difficulties in ANN models mentioned earlier. These reasons attract the attention of

NU

several researchers to SVM models [70-73].

Recently, a modified version of SVM has been presented [69] called least squares support

MA

vector machine (LSSVM) trying to minimize its complexity and improve its convergence speed. LSSVM employs equality constraints rather than inequality ones. This reformulation introduces

TE

D

a system of linear equations which can be iteratively solved in several consecutive steps [67, 69]. Moreover, only a portion of support vectors are applied to construct an approximation model

AC CE P

because of spars feature of SVM, whereas LSSVM utilizes all data points in order to find a satisfactory approximation [69]. The function f(x) based on SVM principle is defined as below: f ( x)  wT  (x)  b

(2)

where  (x) and wT are the kernel function and the transposed output layer, respectively, and b is the bias, the intercept of the linear regression in the modified SVM method (LSSVM). x shows the input vector of the model parameters . It has a dimension of N×n, where N and n represent the number of data points and input parameters, respectively. Vapnik proposed minimization of the below cost function in order to calculate w and b: Cost function 

N 1 T w  c ( k   k* ) 2 k 1

(3)

Subjected to the following constraints:

6

ACCEPTED MANUSCRIPT  y k  wT  ( x k )  b     k , k  1,2,...., N  T *  w  ( x k )  b  y k     k , k  1,2,...., N   ,  *  0, k  1,2,...., N  k k

PT

(4)

where xk and yk stand for kth data input and kth data output, respectively. Moreover, ε expresses

RI

the fixed precision of the function approximation, and  k and  k* are slack variables. Using a

SC

small value for ε would lead to creating an exact model that would lead to placing some data

NU

point out of the ε precision. This would lead to impractical solution. As a result, the stack variables would apply for designation of the permissible limit for the error. The deviation

MA

quantity from the preferable ε is evaluated using the tuning parameter of SVM (the c  0 in Eq. (3)). The lagrangian as follows is applied for minimization of the cost function which is



N

k 1

k











N N 1 N * * *   a  a a  a K x , x   a  a  yk ak  ak*    k k l l k l k k 2 k ,l 1 k 1 k 1

AC CE P

 a



TE



L a, a *  

D

presented in Eq. (3) with constraints in Eq. (4):



 ak*  0 , ak , ak*  0, c



(5)

(5a)

K xk , xl    xk   xl , k  1,2,..., N T

(5b)

where  k and  k show Lagrangian multipliers. Eventually, the final version of SVM can be expressed as follows:

f ( x) 

N

 (a

k , l 1

k

 ak* )K ( x, xk )  b

(6)

The above stated parameters namely, ak , ak* , and b are determined by solving the quadratic programming problem. This is a difficult task. To overcome this problem, Suykens and Vandewalle [69, 74] introduced the modification of SVM technique least square to assist the

7

ACCEPTED MANUSCRIPT main SVM technique. The SVM method is reformulated in LSSVM method which is presented as below [75]: 1 T 1 N w w    ek2 2 2 k 1

PT

Cost function 

RI

Subjected to the following constraint:

SC

yk  wT xk   b  ek

(7)

(8)

NU

where γ is tuning parameter in LSSVM method, and ek is the error variable. The Lagrangian for this problem is:



N 1 T 1 N w w    ek2   ak wT  xk   b  ek  yk 2 2 k 1 k 1

MA

Lw, b, e, a  



(9)

D

where ak represents the multipliers of Lagrangian. To solve the problem the derivatives of Eq.

TE

(9) are equated to zero. This leads to the following equations:

AC CE P

N  L  0  w  ak ( xk )   w k 1  N  L  0  a  0  k  b k 1  L   0  ak  ek , k  1,2,..., N  ek  L  0  wT   xk   b  ek  yk  0 k  1,2,..., N   a  k

(10)

As it is obvious, the number of equations and the unknowns are 2N+2 (ak , ek , w, and b) . Solving the system of equations (Eq. (10)), leads to obtaining the LSSVM parameters. As mentioned before, there is a tuning parameter  in LSSVM. The parameters of the kernel functions are introduced as the second tuning parameter, as both LSSVM and SVM are kernel-based techniques. For the present study, the RBF kernel function is introduced as below:

8

ACCEPTED MANUSCRIPT K ( x, xk )  exp(  || xk  x ||2 /  2 )

(11)

Another tuning parameter is σ2. As a result, there are two tuning parameters in LSSVM method using the

PT

RBF kernel functions. These are obtained by minimization of the deviations of LSSVM technique using

RI

the experimental values [75]. The root mean square error (RMSE) from the outputs of LSSVM

n

i 1

rep. / predi

 Oexpi ) 2

n

1/ 2

(12)

NU

RMSE  

 (O

SC

algorithm is determined as below:

MA

where O is the output, subscripts rep./pred. and exp. stand for the represented/predicted, and experimental values, respectively, and n shows the number of samples from the initial

D

population. In this study, the LSSVM algorithm developed by Suykens and Vandewalle [69] has

AC CE P

TE

been used. Coupled simulated annealing is applied for optimizing LSSVM parameters.

4. Model Evaluation 4.1

Statistical and Graphical Error Analyses

To evaluate the accuracy and robustness of the developed model, several statistical parameters have been used consisting of root mean square error (RMSE), average percent relative error (APRE), average absolute percent relative error (AAPRE), and coefficient of determination (R2). Definitions and equations of these parameters are given below:

1. Root Mean Square Error:

9

ACCEPTED MANUSCRIPT  O n

i exp

i 1

 Oi rep./pred



2

(13)

PT

1 N

RMSE 

1 N

n

E i 1

i

(14)

SC

Er 

RI

2. Average Percent Relative Error:

   100  i  1,2,3,..., n 

(15)

MA

 O i exp  O irep./pred Ei   Oi exp 

NU

where Ei (Percent Relative Error) is determined as follows:

1 N

n

| E i 1

i

|

(16)

TE

Ea 

D

3. Average Absolute Percent Relative Error:

 O n

R2  1

AC CE P

4. Coefficient of Determination (R2):

i 1

 Oi rep./pred 

2

i exp

 O n

i 1

i rep./pred

O

(17)



2

where O is the average of the experimental data points. For model evaluation and visualization of the results, two graphical analyses namely crossplot and error distribution curve were used. The definition of the graphical analyses can be found elsewhere [76, 77]

4.2. Identifying Outliers in Experimental Data 10

ACCEPTED MANUSCRIPT Outliers detection are of paramount importance for model development [78]. For detection of outliers, numerical and graphical methods are introduced [78-82]. Leverage approach for

PT

detection of outliers, determines the residual values and Hat matrix (H) [78, 79, 81, 82]. In this

RI

study, an algorithm is represented to determine the data of interest. The Hat or leverage indices

SC

are determined based on Hat matrix (H), using the following equation [78-81]:

H  X ( X t X ) 1 X t

(18)

NU

where X is an (n  k) matrix consisting of n data and k parameters of the model, and t denotes the

MA

transpose matrix. The Hat values of the data are the diagonal elements of the H value. Graphical identification of the outliers is usually carried out through sketching the William plot

D

based on the H values calculated from Eq. (18). Correlation of Hat indices and standardized

TE

cross-validated residuals (R) is represented by this plot. A warning Leverage (H*) has a constant

AC CE P

value of 3(k+1)/n, where n denotes the number of data points and k stands for the number of model parameters. The cut-off value of three is regarded to R, which would select the data points in the range of ±3 of standard deviation to cover 99% of the distributed data. To develop a reliable model, the values of data points should be located in the criteria of 0 ≤ H ≤ H* and -3 ≤ R ≤ 3. This approach would lead to a valid model from statistical point of view. The ‗‗Good High Leverage‘‘ points are located in H*≤ H and -3 ≤ R ≤ 3 domain. The data points which are located out of the applicability domain are called "Good High Leverage". If the data points are placed in the range of R ˂ -3 or 3 ˂ R, they are called ‗‗Bad High Leverage‘‘ points which are outliers.

5. Results and discussion

11

ACCEPTED MANUSCRIPT In the present work, CSA-LSSVM algorithm was implemented to construct an accurate, reliable and robust model for predicting electrical conductivity of IL ternary mixtures. As mentioned

PT

earlier, the input model parameters were temperature, molecular weight (Mw), melting point

RI

(Tf), and composition of the components. In fact, there are nine inputs consist of temperature, molecular weight and melting point of three components and mole fraction of the two first

SC

components. It is worth noting that as the summation of the mole fractions of components is

NU

equal to one, the third compound's mole fraction would not be an independent variable. Using various sources, a large data bank of the ternary mixtures was selected. The collected data was

MA

initially divided into two subsets including training and testing sets. The former set is employed to perform and generate the model structure. The latter set is used to investigate the final

TE

D

performance and validity of the proposed model for unseen data. To increase the model applicability and robustness, the whole database was divided

AC CE P

randomly into two subsets: 80% (179 data points) and 20% (45 data points) as training and testing sets, respectively. To optimize the LSSVM parameters, coupled simulated annealing technique was applied. The optimization process has been performed repeatedly to obtain the most adequate probable optimum of the objective function, resulting in 0.0446 for σ2 and 269724.1282 for γ, where σ2 and γ are the two main parameters of this model. The proposed model was evaluated using the statistical parameters, including RMSE, APRE, R2, and AAPRE, which are reported in Table 2. This table shows the statistical parameters for both of the ―Training‖ and ―Test‖ sets. These parameters show the accuracy of the model in all data sets including the ―Training‖ and ―Test‖ sets. The results reveal that the developed model, reports the electrical conductivity of IL mixtures with APRE and AAPRE

12

ACCEPTED MANUSCRIPT equal to -2.54% and 5.42%, respectively. Moreover, the values of other statistical parameters are reported as R2 = 0.9998 and RMSE = 0.51.

PT

In addition to statistical error analysis, graphical error analysis was performed to

RI

visualize the accuracy and performance of the proposed CSA-LSSVM model. Fig. 1 shows the crossplot of experimentally measured electrical conductivity values versus predicted values by

SC

the proposed model. As it is obvious, a light cloud around the unit slope line reveals the

NU

precision and validity of the developed model without underestimation or overestimation in the training and testing data sets, demonstrating the superior proficiency of the proposed model. In

MA

general, the 45° straight line between the experimental values and represented/predicted data shows the perfect model line. The closer the plotted data to the 45° perfect model line, the higher

TE

D

is the consistency of the model. Deviations of the electrical conductivity values predicted by the model from experimentally measured data are also represented in Figs. 2 and 3, for training and

AC CE P

test sets, respectively. As indicated in Fig. 2, a well accordance exists between electrical conductivity values predicted by CSA-LSSVM model and electrical conductivity from experimental data set in training set. As it is obvious, the developed model, predicts the values of electrical conductivity data accurately in the training set. Besides, Fig 3 illustrates the excellent results of the test set using the developed model. Using the graphical and statistical error results, it is clear that this model can accurately determine the value of electrical conductivity of the ternary mixtures including ILs; therefore, the developed model is an excellent candidate for determining the electrical conductivity of IL mixtures, instead of performing the expensive and time consuming experimental techniques. The percent relative error from experimental values of the training and test datasets are shown in Fig. 4. As can be seen in Fig. 4, at very low electrical conductivities, or high concentration of ionic liquids the deviations from experimental values are

13

ACCEPTED MANUSCRIPT generally larger than for higher values of electrical conductivities or lower concentration of ionic liquids. The estimated electrical conductivities by the proposed LSSVM model with the

PT

experimental ones in different ranges of temperature and composition for the [bmim][PF6] +

RI

H2O (2) + ethanol (3) and [bmim][PF6] + H2O (2) + acetone (3) systems were compared in Figs. 5 and 6, respectively. As can be seen in both of the systems, by variation of the composition and

SC

temperature, the developed model follows the trends that are physically expected and the

NU

predicted data match well with the corresponding experimental ones. In order to detect and identify the suspected data, the Leverage statistical approach was

MA

accomplished in this study, in which the residuals of the model, Williams Plot, and statistical Hat

matrix lead to recognition of possible outliers. The h values were calculated through Eq. (18).

D

Moreover, the Williams plot has been sketched in Fig. 7 for the output obtained from CSA-

TE

LSSVM. Being the majority of data points within the ranges 0 < H < 0.134 and −3 < R < 3

AC CE P

reveals that the proposed model is statistically valid and accurate. As is shown in Fig. 7, the entire data point except seven points are located within the applicable domain for the presented CSA-LSSVM model. Accordingly, these seven points can be stated as probable doubtful datum. These doubtful data are presented in Table. 4. It is interesting to demonstrate that the qualities of the electrical conductivity data are not the same. In other words, lower value of R, and H would lead to more accurate experimental data. To evaluate the influence of input parameters (i.e., melting point, molecular weight and mole percent of each compound) on the output, a sensitivity analyses was performed. For this purpose, to evaluate the impact of each parameter on the electrical conductivity of ternary mixture, the relevancy factor (r) [83-85] was introduced. Higher values for r between input and output parameters demonstrate the higher impact of that input variable on the output result. The negative or positive impact of input parameters on ternary

14

ACCEPTED MANUSCRIPT mixture electrical conductivity, nevertheless, is not demonstrated by the absolute value of r. The relevancy factor values with directionality causes a clearer and sensational comprehension about

PT

the entire impact, which was calculated in the present study. The values of r are determined as

n



( Inp k ,i  Inp k )(i   )

( Inp k ,i  Inpk ) 2 i 1 (i   ) i 1 n

n

2

(19)

NU

r ( Inp k ,  g ) 

i 1

SC



RI

follows:

where Inpk,i and Inpk are the ith value and the average value of the of the kth input variable, respectively

MA

(k=average melting point and average molecular weight), λi represents ith value of the reported electrical conductivity, and  denotes the average value of the presented electrical conductivity. First of all, it was

D

aimed to figure out the relevancy factor between each of the nine input parameters and the electrical

TE

conductivity; nevertheless, not meaningful outcome was resulted, because the combination of these

AC CE P

parameters affects electrical conductivity of ternary mixtures. To overcome this problem, the relevancy factor between temperature, average melting point (Tfa) of the ternary mixtures and average molecular weight (MWa) of the ternary mixtures with electrical conductivity was calculated. These average parameters were determined using the following formula: i 3

(20)

Tf a   Tf i  xi i 1

i 3

(21)

MWa   MWi  xi i 1

The result of the sensitivity analysis conducted in this study is shown in Fig. 8. As it is demonstrated obviously in the figure, either of temperature and average molecular weight has positive impacts on electrical conductivity of ternary mixture, while average melting point has a negative impact on ternary mixtures electrical conductivity. As can be seen in this figure, the 15

ACCEPTED MANUSCRIPT relevancy factors are 0.56, -0.79 and 0.05 for the average molecular weight, average melting

PT

point and temperature, respectively.

RI

6. Conclusion

SC

In this article, a large data set of ternary mixtures containing ILs was prepared from literature sources and a supervised learning algorithm which is least square support vector machine has

NU

been presented to estimate the electrical conductivity of IL ternary mixtures as a function of

MA

temperature, composition, molecular weight and melting points of components. About 80% of the whole dataset (179 data points) were used for training and 20% (45 data points) of the whole

D

dataset were applied to test the model performance. The developed CSA-LSSVM model predicts

TE

electrical conductivity of ternary mixtures containing ILs with a good accuracy. Graphical and statistical error analyses shows that the developed model is efficient and reliable. Predictions of

AC CE P

the model were compared with experimental measurements and well accordance was observed, and the overall R2 of 0.9999 and AAPRE 5.42% were obtained. The relevancy factor revealed that the average melting point has the greatest impact on the ternary mixture electrical conductivities. At the end, the Leverage approach revealed that the developed model is valid and reliable form statistical point of view.

16

ACCEPTED MANUSCRIPT References

AC CE P

TE

D

MA

NU

SC

RI

PT

. Walden, Ueber die Molekulargr sse und elektrische Leitf higkeit einiger geschmolzenen Salze, Известия Российской академии наук. Серия математическая, 8 ( 9 4) 405-422. [2] J.S. Wilkes, M.J. Zaworotko, Air and water stable 1-ethyl-3-methylimidazolium based ionic liquids, J. Chem. Soc., Chem. Commun., (1992) 965-967. [3] J.F. Brennecke, E.J. Maginn, Ionic liquids: innovative fluids for chemical processing, AIChE Journal, 47 (2001) 2384-2389. [4] K. Marsh, J. Boxall, R. Lichtenthaler, Room temperature ionic liquids and their mixtures—a review, Fluid Phase Equilibria, 219 (2004) 93-98. [5] M.P. Singh, R.K. Singh, S. Chandra, Ionic liquids confined in porous matrices: Physicochemical properties and applications, Progress in Materials Science, 64 (2014) 73-120. [6] M.J. Earle, J.M. Esperança, M.A. Gilea, J.N.C. Lopes, L.P. Rebelo, J.W. Magee, K.R. Seddon, J.A. Widegren, The distillation and volatility of ionic liquids, Nature, 439 (2006) 831834. [7] Q. Yang, H. Xing, Y. Cao, B. Su, Y. Yang, Q. Ren, Selective separation of tocopherol homologues by liquid− liquid extraction using ionic liquids, Industrial & Engineering Chemistry Research, 48 (2009) 6417-6422. [8] K. Ui, K. Yamamoto, K. Ishikawa, T. Minami, K. Takeuchi, M. Itagaki, K. Watanabe, N. Koura, Development of non-flammable lithium secondary battery with room-temperature ionic liquid electrolyte: Performance of electroplated Al film negative electrode, Journal of Power Sources, 183 (2008) 347-350. [9] D.R. MacFarlane, M. Forsyth, P.C. Howlett, J.M. Pringle, J. Sun, G. Annat, W. Neil, E.I. Izgorodina, Ionic liquids in electrochemical devices and processes: managing interfacial electrochemistry, Accounts of chemical research, 40 (2007) 1165-1173. [10] K. Kubota, T. Nohira, T. Goto, R. Hagiwara, Novel inorganic ionic liquids possessing low melting temperatures and wide electrochemical windows: Binary mixtures of alkali bis (fluorosulfonyl) amides, Electrochemistry Communications, 10 (2008) 1886-1888. [11] S. Ahrens, A. Peritz, T. Strassner, Tunable aryl alkyl ionic liquids (TAAILs): The next generation of ionic liquids, Angewandte Chemie International Edition, 48 (2009) 7908-7910. [12] M.J. Earle, K.R. Seddon, Ionic liquids. Green solvents for the future, Pure and applied chemistry, 72 (2000) 1391-1398. [13] M. Królikowska, T. Hofman, Densities, isobaric expansivities and isothermal compressibilities of the thiocyanate-based ionic liquids at temperatures (298.15–338.15 K) and pressures up to 10MPa, Thermochimica Acta, 530 (2012) 1-6. [14] H. Jiang, J. Wang, F. Zhao, G. Qi, Y. Hu, Volumetric and surface properties of pure ionic liquid n-octyl-pyridinium nitrate and its binary mixture with alcohol, The Journal of Chemical Thermodynamics, 47 (2012) 203-208. [15] O. Ciocirlan, O. Croitoru, O. Iulian, Densities and viscosities for binary mixtures of 1-butyl3-methylimidazolium tetrafluoroborate ionic liquid with molecular solvents, Journal of Chemical & Engineering Data, 56 (2011) 1526-1534. [16] W. Qian, Y. Xu, H. Zhu, C. Yu, Properties of pure 1-methylimidazolium acetate ionic liquid and its binary mixtures with alcohols, The Journal of Chemical Thermodynamics, 49 (2012) 8794. [17] J. Canongia Lopes, M.F. Costa Gomes, P. Husson, A.A. Pádua, L.P.N. Rebelo, S. Sarraute, M. Tariq, Polarity, viscosity, and ionic conductivity of liquid mixtures containing 17

ACCEPTED MANUSCRIPT

AC CE P

TE

D

MA

NU

SC

RI

PT

[C4C1im][Ntf2] and a molecular component, The Journal of Physical Chemistry B, 115 (2011) 6088-6099. [18] Y. Xu, J. Yao, C. Wang, H. Li, Density, viscosity, and refractive index properties for the binary mixtures of n-butylammonium acetate ionic liquid+ alkanols at several temperatures, Journal of Chemical & Engineering Data, 57 (2012) 298-308. [19] J. Dupont, R.F. de Souza, P.A. Suarez, Ionic liquid (molten salt) phase organometallic catalysis, Chemical reviews, 102 (2002) 3667-3692. [20] J.-P. Belieres, D. Gervasio, C.A. Angell, Binary inorganic salt mixtures as high conductivity liquid electrolytes for> 100 C fuel cells, Chemical communications, (2006) 4799-4801. [21] H. Nakamoto, M. Watanabe, Brønsted acid–base ionic liquids for fuel cell electrolytes, Chemical Communications, (2007) 2539-2541. [22] H. Ye, J. Huang, J. Xu, N. Kodiweera, J. Jayakody, S. Greenbaum, New membranes based on ionic liquids for PEM fuel cells at elevated temperatures, Journal of Power Sources, 178 (2008) 651-660. [23] D. Han, K.H. Row, Recent applications of ionic liquids in separation technology, Molecules, 15 (2010) 2405-2426. [24] M.H. Ghatee, M. Bahrami, N. Khanjari, Measurement and study of density, surface tension, and viscosity of quaternary ammonium-based ionic liquids ([N 222 (n)] Tf 2 N), The Journal Of Chemical Thermodynamics, 65 (2013) 42-52. [25] R.L. Gardas, J.A. Coutinho, A group contribution method for viscosity estimation of ionic liquids, Fluid Phase Equilibria, 266 (2008) 195-201. [26] F. Endres, D. MacFarlane, A. Abbott, Electrodeposition from ionic liquids, John Wiley & Sons, 2008. [27] S. Hwang, Y. Park, K. Park, Measurement and prediction of phase behaviour for 1-alkyl-3methylimidazolium tetrafluoroborate and carbon dioxide: Effect of alkyl chain length in imidazolium cation, The Journal of Chemical Thermodynamics, 43 (2011) 339-343. [28] J. Safarov, I. Kul, W.A. El-Awady, A. Shahverdiyev, E. Hassel, Thermodynamic properties of 1-butyl-3-methylpyridinium tetrafluoroborate, The Journal of Chemical Thermodynamics, 43 (2011) 1315-1322. [29] G. Vakili-Nezhaad, M. Vatani, M. Asghari, I. Ashour, Effect of temperature on the physical properties of 1-butyl-3-methylimidazolium based ionic liquids with thiocyanate and tetrafluoroborate anions, and 1-hexyl-3-methylimidazolium with tetrafluoroborate and hexafluorophosphate anions, The Journal of Chemical Thermodynamics, 54 (2012) 148-154. [30] E. Gómez, N. Calvar, Á. Domínguez, E.A. Macedo, Synthesis and temperature dependence of physical properties of four pyridinium-based ionic liquids: Influence of the size of the cation, The Journal of Chemical Thermodynamics, 42 (2010) 1324-1329. [31] M. Earle, A. Forestier, H. Olivier-Bourbigou, P. Wasserscheid, P. Wasserscheid, T. Welton, Ionic liquids in synthesis, in, Wiley-VCH: Weinheim, 2003. [32] P. Wasserscheid, T. Welton, Ionic liquids in synthesis, Wiley Online Library, 2008. [33] W. Lu, A.G. Fadeev, B. Qi, B.R. Mattes, Fabricating conducting polymer electrochromic devices using ionic liquids, Journal of The Electrochemical Society, 151 (2004) H33-H39. [34] M. Galin, A. Chapoton, J.-C. Galin, Dielectric increments, intercharge distances and conformation of quaternary ammonioalkylsulfonates and alkoxydicyanoethenolates in aqueous and tirfluoroethanol solutions, J. Chem. Soc., Perkin Trans. 2, (1993) 545-553. [35] M. Yoshizawa, A. Narita, H. Ohno, Design of ionic liquids for electrochemical applications, Australian journal of chemistry, 57 (2004) 139-144. 18

ACCEPTED MANUSCRIPT

AC CE P

TE

D

MA

NU

SC

RI

PT

[36] J. Chen, H. Song, C. Xia, Z. Tang, A method of synthesizing trioxane through formaldehyde cyclizing by using bi-functional ILs as the catalyst, Chinese Patent: CN, 102020630 (2011). [37] A. Biswas, R. Shogren, D. Stevenson, J. Willett, P.K. Bhowmik, Ionic liquids as solvents for biopolymers: Acylation of starch and zein protein, Carbohydrate Polymers, 66 (2006) 546550. [38] M. Kanakubo, H. Nanjo, T. Nishida, J. Takano, Density, viscosity, and electrical conductivity of N-methoxymethyl-N-methylpyrrolidinium bis (trifluoromethanesulfonyl) amide, Fluid Phase Equilibria, 302 (2011) 10-13. [39] D. Tomida, S. Kenmochi, K. Qiao, T. Tsukada, C. Yokoyama, Densities and thermal conductivities of N-alkylpyridinium tetrafluoroborates at high pressure, Fluid Phase Equilibria, 340 (2013) 31-36. [40] T. Makino, M. Kanakubo, Y. Masuda, T. Umecky, A. Suzuki, CO 2 absorption properties, densities, viscosities, and electrical conductivities of ethylimidazolium and 1-ethyl-3methylimidazolium ionic liquids, Fluid Phase Equilibria, 362 (2014) 300-306. [41] Y. Wang, G.A. Voth, Unique spatial heterogeneity in ionic liquids, Journal of the American Chemical Society, 127 (2005) 12192-12193. [42] C. Rey-Castro, L.F. Vega, Transport properties of the ionic liquid 1-ethyl-3methylimidazolium chloride from equilibrium molecular dynamics simulation. The effect of temperature, The Journal of Physical Chemistry B, 110 (2006) 14426-14435. [43] R. Brookes, A. Davies, G. Ketwaroo, P.A. Madden, Diffusion coefficients in ionic liquids: Relationship to the viscosity, The Journal of Physical Chemistry B, 109 (2005) 6485-6490. [44] W. Jiang, Y. Wang, G.A. Voth, Molecular dynamics simulation of nanostructural organization in ionic liquid/water mixtures, The Journal of Physical Chemistry B, 111 (2007) 4812-4818. [45] Q. Yang, K. Yu, H. Xing, B. Su, Z. Bao, Y. Yang, Q. Ren, The effect of molecular solvents on the viscosity, conductivity and ionicity of mixtures containing chloride anion-based ionic liquid, Journal of Industrial and Engineering Chemistry, 19 (2013) 1708-1714. [46] P.D. de María, A. Martinsson, Ionic-liquid-based method to determine the degree of esterification in cellulose fibers, Analyst, 134 (2009) 493-496. [47] Q. Yang, H. Xing, B. Su, K. Yu, Z. Bao, Y. Yang, Q. Ren, Improved separation efficiency using ionic liquid–cosolvent mixtures as the extractant in liquid–liquid extraction: A multiple adjustment and synergistic effect, Chemical Engineering Journal, 181 (2012) 334-342. [48] S. Zhang, X. Li, H. Chen, J. Wang, J. Zhang, M. Zhang, Determination of physical properties for the binary system of 1-ethyl-3-methylimidazolium tetrafluoroborate+ H2O, Journal of Chemical & Engineering Data, 49 (2004) 760-764. [49] S. Tian, S. Ren, Y. Hou, W. Wu, W. Peng, Densities, Viscosities and Excess Properties of Binary Mixtures of 1, 1, 3, 3-Tetramethylguanidinium Lactate+ Water at T=(303.15 to 328.15) K, Journal of Chemical & Engineering Data, 58 (2013) 1885-1892. [50] D. Le Botlan, B. Mechin, G. Martin, Proton and carbon-13 nuclear magnetic resonance spectrometry of formaldehyde in water, Analytical Chemistry, 55 (1983) 587-591. [51] A. Arce, O. Rodríguez, A. Soto, A comparative study on solvents for separation of tert-amyl ethyl ether and ethanol mixtures. New experimental data for 1-ethyl-3-methyl imidazolium ethyl sulfate ionic liquid, Chemical engineering science, 61 (2006) 6929-6935. [52] C.J. Rao, K. Venkatesan, K. Nagarajan, T. Srinivasan, P.V. Rao, Treatment of tissue paper containing radioactive waste and electrochemical recovery of valuables using ionic liquids, Electrochimica acta, 53 (2007) 1911-1919. 19

ACCEPTED MANUSCRIPT

AC CE P

TE

D

MA

NU

SC

RI

PT

[53] A. Heintz, Recent developments in thermodynamics and thermophysics of non-aqueous mixtures containing ionic liquids. A review, The Journal of Chemical Thermodynamics, 37 (2005) 525-535. [54] Z. Lei, B. Chen, C. Li, H. Liu, Predictive molecular thermodynamic models for liquid solvents, solid salts, polymers, and ionic liquids, Chemical reviews, 108 (2008) 1419-1455. [55] J. Zhang, W. Wu, T. Jiang, H. Gao, Z. Liu, J. He, B. Han, Conductivities and viscosities of the ionic liquid [bmim][PF6]+ water+ ethanol and [bmim][PF6]+ water+ acetone ternary mixtures, Journal of Chemical & Engineering Data, 48 (2003) 1315-1317. [56] M. Hosseinzadeh, A. Hemmati-Sarapardeh, Toward a predictive model for estimating viscosity of ternary mixtures containing ionic liquids, Journal of Molecular Liquids, 200 (2014) 340-348. [57] Q. Liang, Y. Hu, W. Yue, Electrical conductivities for four ternary electrolyte aqueous solutions with one or two ionic liquid components at ambient temperatures and pressure, Chinese Journal of Chemical Engineering, (2015). [58] J. Vogel, L. Yarborough, The effect of nitrogen on the phase behavior and physical properties of reservoir fluids, in: SPE/DOE Enhanced Oil Recovery Symposium, 1980. [59] J. Anwar, D. Frenkel, M.G. Noro, Calculation of the melting point of NaCl by molecular simulation, The Journal of chemical physics, 118 (2003) 728-735. [60] D.W. Mayo, R.M. Pike, D.C. Forbes, Microscale organic laboratory: with multistep and multiscale syntheses, John Wiley & Sons, 2010. [61] P. Ballirano, Laboratory parallel-beam transmission X-ray powder diffraction investigation of the thermal behavior of nitratine NaNO3: spontaneous strain and structure evolution, Physics and Chemistry of Minerals, 38 (2011) 531-541. [62] T. Suzuki, M. Itoh, Y. Takegami, Y. Watanabe, Chemical structure of tar-sand bitumens by< sup> 13 C and< sup> 1 H nmr spectroscopic methods, Fuel, 61 (1982) 402410. [63] S. Zhang, X. Lu, Q. Zhou, X. Li, X. Zhang, S. Li, Ionic Liquids:: Physicochemical Properties, Elsevier, 2009. [64] S. Ayatollahi, A. Hemmati-Sarapardeh, M. Roham, S. Hajirezaie, A rigorous approach for determining interfacial tension and minimum miscibility pressure in paraffin-CO2 systems: Application to gas injection processes, Journal of the Taiwan Institute of Chemical Engineers. [65] S.-M. Tohidi-Hosseini, S. Hajirezaie, M. Hashemi-Doulatabadi, A. Hemmati-Sarapardeh, A.H. Mohammadi, Toward prediction of petroleum reservoir fluids properties: A rigorous model for estimation of solution gas-oil ratio, Journal of Natural Gas Science and Engineering, 29 (2016) 506-516. [66] A. Hemmati-Sarapardeh, F. Ameli, B. Dabir, M. Ahmadi, A.H. Mohammadi, On the evaluation of asphaltene precipitation titration data: Modeling and data assessment, Fluid Phase Equilibria, 415 (2016) 88-100. [67] F. Gharagheizi, A. Eslamimanesh, F. Farjood, A.H. Mohammadi, D. Richon, Solubility parameters of nonelectrolyte organic compounds: determination using quantitative structure– property relationship strategy, Industrial & Engineering Chemistry Research, 50 (2011) 1138211395. [68] N. Cristianini, J. Shawe-Taylor, An introduction to support vector machines and other kernel-based learning methods, Cambridge university press, 2000. [69] J.A. Suykens, J. Vandewalle, Least squares support vector machine classifiers, Neural processing letters, 9 (1999) 293-300. 20

ACCEPTED MANUSCRIPT

AC CE P

TE

D

MA

NU

SC

RI

PT

[70] V. Vapnik, The nature of statistical learning theory, Springer Science & Business Media, 2000. [71] A. Baylar, D. Hanbay, M. Batan, Application of least square support vector machines in the prediction of aeration performance of plunging overfall jets from weirs, Expert Systems with Applications, 36 (2009) 8368-8374. [72] E. Byvatov, U. Fechner, J. Sadowski, G. Schneider, Comparison of support vector machine and artificial neural network systems for drug/nondrug classification, Journal of Chemical Information and Computer Sciences, 43 (2003) 1882-1889. [73] B. Schölkopf, A.J. Smola, Learning with kernels: Support vector machines, regularization, optimization, and beyond, MIT press, 2002. [74] K. Pelckmans, J. Suykens, T. Van Gestel, J. De Brabanter, L. Lukas, B. Hamers, B. De Moor, J. Vandewalle, LS-SVMlab toolbox, Dept. Elect. Eng., ESAT-SCD-SISTA, Leuven, Belgium, (2002). [75] J.A. Suykens, T. Van Gestel, J. De Brabanter, B. De Moor, J. Vandewalle, J. Suykens, T. Van Gestel, Least squares support vector machines, World Scientific, 2002. [76] M. Fathinasab, S. Ayatollahi, A. Hemmati-Sarapardeh, A Rigorous Approach to Predict Nitrogen-Crude Oil Minimum Miscibility Pressure of Pure and Nitrogen Mixtures, Fluid Phase Equilibria, 399 (2015) 30-39. [77] E. Mohagheghian, H. Zafarian-Rigaki, Y. Motamedi-Ghahfarrokhi, A. HemmatiSarapardeh, Using an artificial neural network to predict carbon dioxide compressibility factor at high pressure and temperature, Korean Journal of Chemical Engineering, 32 (2015) 2087-2096. [78] A.M. Leroy, P.J. Rousseeuw, Robust regression and outlier detection, Wiley Series in Probability and Mathematical Statistics, New York: Wiley, 1987, 1 (1987). [79] C.R. Goodall, Computation using the QR decomposition, Elsevier, 1993. [80] F. Gharagheizi, A. Eslamimanesh, M. Sattari, B. Tirandazi, A.H. Mohammadi, D. Richon, Evaluation of thermal conductivity of gases at atmospheric pressure through a corresponding states method, Industrial & Engineering Chemistry Research, 51 (2012) 3844-3849. [81] P. Gramatica, Principles of QSAR models validation: internal and external, QSAR & Combinatorial Science, 26 (2007) 694-701. [82] A.H. Mohammadi, A. Eslamimanesh, F. Gharagheizi, D. Richon, A novel method for evaluation of asphaltene precipitation titration data, Chemical Engineering Science, (2012). [83] G. Chen, K. Fu, Z. Liang, T. Sema, C. Li, P. Tontiwachwuthikul, R. Idem, The genetic algorithm based back propagation neural network for MMP prediction in CO 2-EOR process, Fuel, 126 (2014) 202-212. [84] A. Hemmati-Sarapardeh, B. Aminshahidy, A. Pajouhandeh, S.H. Yousefi, S.A. HosseiniKaldozakh, A soft computing approach for the determination of crude oil viscosity: Light and intermediate crude oil systems, Journal of the Taiwan Institute of Chemical Engineers, (2015). [85] S. Hajirezaie, A. Hemmati-Sarapardeh, A.H. Mohammadi, M. Pournik, A. Kamari, A Smooth Model for the Estimation of Gas/Vapor Viscosity of Hydrocarbon Fluids, Journal of Natural Gas Science and Engineering, 26 (2015) 1452–1459.

21

ACCEPTED MANUSCRIPT

Table 1 Components' Properties of each ternary mixtures Molar weight (g/gmol)

melting point (K)

Ref.

Water

18.01

273.15

]85[

NaCl

58.44

1074

[59]

Ethanol

46.07

159.15

[60]

NaNO3

84.9947

581

Acetone

58.08

[c4mim][PF6]

284.18

[C4mim][BF4]

226.03

[C6mim][Cl]

202.73

[C6mim][BF4]

254.08

[61]

178.15

]85[

283.1

[62]

192.15

[63]

198.15

[63]

191.15

[63]

AC CE P

TE

D

MA

NU

SC

RI

PT

Component

22

ACCEPTED MANUSCRIPT

Table 2: Statistical parameters of the ternary mixtures at different conditions Component 3

Temperature Range, K

[bmim][PF6] [bmim][PF6] H2O H2O H2O

Ethanol Acetone [C6mim][BF4] NaCl [C6mim][BF4]

288.15-308.15 288.15-308.16 273.15-274.32 273.16-275.27 272.85-273.15

RI

SC

AC CE P

TE

D

MA

NU

H2O H2O NaNO3 [C6mim][Cl] [C6mim][Cl]

Number of Data points

PT

Component 1 Component 2

23

Ref

56 49

[55] [55]

33 30 56

[57] [57] [57]

ACCEPTED MANUSCRIPT Table 3: The suspected experimental data based on the Leverage approach Tf 1

[55]

[bmim][PF6]

H2O

Ethanol

[55]

[bmim][PF6]

H2O

[55]

[bmim][PF6]

[55]

[bmim][PF6]

[57]

H2O

[57] [57]

Tf 2

MW 1 MW MW 2 3

Tf 3

T (K)

X1

X2

Hat

R

283.1

273.15 159.15 284.18 18.01 46.07 308.15

0.8

0.10

0.047

3.679

Ethanol

283.1

273.15 178.15 284.18 18.01 58.08 288.15

0.42

0.17

0.061 -3.524

H2O

Ethanol

283.1

273.15 178.15 284.18 18.01 58.08 308.15

0.52

0.34

0.070

4.682

H2O

Ethanol

283.1

273.15 178.15 284.18 18.01 58.08 308.15

0.52

0.14

0.048

3.197

[C6mim][Cl] [C6mim][BF4] 273.15 192.15 191.15

18.01 226.0 254.0 298.15 0.99729 0.002032 0.018

3.122

H2O

[C6mim][Cl] [C6mim][BF4] 273.15 192.15 191.15

18.01 226.0 254.0 298.15 0.99640 0.001803 0.018

3.036

H2O

[C6mim][Cl] [C6mim][BF4] 273.15 192.15 191.15

PT

Component 3

RI

Component 1 Component 2

AC CE P

TE

D

MA

NU

SC

Ref

24

18.01 226.0 254.0 298.15 0.99639 0.00271 0.018

3.438

ACCEPTED MANUSCRIPT Table 4: statistical error analysis of the proposed model in both training and testing sets. APRE (%)

AAPRE (%)

RMSE

R2

This Study, Training Set

-3.06

5.85

0.3999

0.9999

This Study, Test Set

-0.46

3.75

0.8276

0.9999

This Study, Total

-2.54

5.42

0.5152

0.9999

AC CE P

TE

D

MA

NU

SC

RI

PT

Data Sets

25

ACCEPTED MANUSCRIPT

200

PT

160

RI

140

SC

120

NU

100 80

MA

60 40 20

Training Set Testing Set

D

Predicted Electrical Conductivity (S/cm)×10

4

180

50 100 150 4 Experimental Electrical Conductivity (S/cm)×10

AC CE P

0

TE

0

200

Fig. 1: Crossplot for electrical conductivity of ternary mixtures in this study

26

ACCEPTED MANUSCRIPT

200 Predicted Data

180

PT

160

RI

140

SC

120 100 80

NU

Electrical Conductivity (S/cm)×10

4

Experimental Data

60

MA

40

0 20

40

60

TE

0

D

20

80 100 Data Index

120

140

160

180

AC CE P

Fig. 2: Comparison between CSA-LSSVM model predictions with the experimental data in training phase

27

ACCEPTED MANUSCRIPT

200

Predicted Data Experimental Data

PT

160

RI

140

SC

120 100 80

NU

Electrical Conductivity (S/cm)×10

4

180

60

MA

40

0 5

10

15

TE

0

D

20

20 25 Data Index

30

35

40

45

AC CE P

Fig. 3: Comparison between CSA-LSSVM model predictions with the experimental data in testing phase

28

ACCEPTED MANUSCRIPT

20

Training Set

PT

10

RI

5

Testing Set

0

SC

Relative Error (%)

15

-5

NU

-10 -15

0

MA

-20 40

80

120

Experimental Electrical Conductivity (S/cm)×10

160

200

4

AC CE P

TE

D

Fig. 4: Relative error versus experimental electrical conductivity in both traring and testing phases

29

PT

ACCEPTED MANUSCRIPT

180.00

X1=0.6, X2=0.2 (Pred) X1=0.8, X2=0.06 (Exp) X1=0.8, X2=0.06 (Pred) X1=0.52, X2=0.14 (Exp) X1=0.52, X2=0.14 (Pred) X1=0.38, X2=0.31 (Exp)

RI

120.00 100.00 80.00 60.00 40.00 20.00

D

X1=0.38, X2=0.31 (Pred)

SC

X1=0.6, X2=0.2 (Exp)

140.00

NU

X1=0.3, X2=0.35 (Pred)

MA

X1=0.3, X2=0.35 (Exp)

Electrical Conductivity (S/cm)×104

160.00

TE

0.00 285

290

AC CE P

[bmim][PF6] (1) H2O (2) Ethanol (3)

295 300 Tempreture (K)

305

310

Fig. 5: The estimated electrical conductivities by the proposed LSSVM model with the experimental ones in different ranges of temperature and composition for the [bmim][PF6] + H2O (2) + ethanol (3)

30

ACCEPTED MANUSCRIPT

X1=0.8, X2=.06 (Pred) X1=0.52, X2=0.14 (Exp) X1=0.52, X2=0.14 (Pred) X1=0.52, X2=0.34 (Exp) X1=0.52, X2=0.34 (Pred)

PT

X1=0.8, X2=0.06 (Exp)

140.00 120.00

RI

X1=.42, X2=.17 (Pred)

160.00

100.00

SC

X1=.42, X2=.17 (Exp)

180.00

80.00 60.00 40.00

MA

20.00

NU

Electrical Conductivity (S/cm)×104

200.00

0.00

285

D

[bmim][PF6] (1) H2O (2) Acetone (3)

290

295

300

305

310

Tempreture (K)

AC CE P

TE

Fig. 6: The estimated electrical conductivities by the proposed LSSVM model with the experimental ones in different ranges of temperature and composition for [bmim][PF6] + H2O (2) + acetone (3) systems

31

ACCEPTED MANUSCRIPT

6 Valid Data

Lower Suspected Limit

PT

3

Upper Suspected Limit

RI

Leverage

0

SC

Standarized Residuals

Suspected Data

NU

-3

0

0.02

MA

-6 0.04

0.06

0.08

0.1

0.12

0.14

Hat

AC CE P

TE

D

Fig. 7: Detecting the probable outlier data and applicability domain of the LSSVM model

32

ACCEPTED MANUSCRIPT 0.6 0.56

Average Tf

NU

-0.79

0 -0.2

Temperature

SC

Average MW

RI

0.05

PT

0.2

-0.4

Relevancy Factor

0.4

-0.6 -0.8

AC CE P

TE

D

MA

Fig. 8: Relevancy factor of temperature, average molecular weight and average melting point of ternary mixtures with electrical conductivity

33

ACCEPTED MANUSCRIPT

PT

Highlights

Electrical conductivity of ternary mixtures containing ionic liquids is modeled.



The developed model has enough accuracy.



The model is capable of simulating the actual physical trends.



The Leverage approach is performed to identify the probable outliers.

AC CE P

TE

D

MA

NU

SC

RI



34