processes

Article
Machine Learning Prediction of Critical Temperature of
Organic Refrigerants by Molecular Topology
Yi Que 1,2, *, Song Ren 1,2 , Zhiming Hu 1 and Jiahui Ren 2, *

1 China Petroleum Engineering and Construction Corporation Southwest Company, Chengdu 610041, China;
[email protected] (S.R.); [email protected] (Z.H.)
2 Key Laboratory of Low-Grade Energy Utilization Technologies and Systems, Ministry of Education,
School of Energy and Power Engineering, Chongqing University, Chongqing 400030, China
* Correspondence: [email protected] (Y.Q.); [email protected] (J.R.)

Abstract: In this work, molecular structures, combined with machine learning algorithms, were
applied to predict the critical temperatures (Tc ) of a group of organic refrigerants. Aiming at solving
the problem that previous models cannot distinguish isomers, a topological index was introduced.
The results indicate that the novel molecular descriptor ‘molecular fingerprint + topological index’
can effectively differentiate isomers. The average absolute average deviation between the predicted
and experimental values is 3.99%, which proves a reasonable prediction ability of the present method.
In addition, the performance of the proposed model was compared with that of other previously
reported methods. The results show that the present model is superior to other approaches with
respect to accuracy.

Keywords: refrigerants; critical temperature; molecular structure; machine learning





 1. Introduction
Citation: Que, Y.; Ren, S.; Hu, Z.; Ren,
The proposal of carbon neutrality will accelerate the utilization of renewable energy,
J. Machine Learning Prediction of
such as solar energy and geothermal energy [1,2]. Additionally, thermodynamic cycles,
Critical Temperature of Organic
including novel power systems represented by the organic Rankine cycle (ORC), and re-
Refrigerants by Molecular Topology.
frigeration/heat pump cycles, represented by a vapor compression cycle, are the effective
Processes 2022, 10, 577. https://
approaches to use the medium and low energy. A working fluid is the energy carrier
doi.org/10.3390/pr10030577
of thermodynamic cycles, which plays a key role in designing and enhancing the ther-
Academic Editor: Sergio Bobbo modynamic cycles [3,4]. Organic refrigerants, as a kind of compounds with low boiling
Received: 16 February 2022
point, apart from their application in the refrigeration industry, also have unique advantage
Accepted: 9 March 2022
to be used as working fluids in ORC to recover low-grade energy and improve energy
Published: 16 March 2022
utilization efficiency.
In recent years, with increasing attention paid to environmental problems, such as
Publisher’s Note: MDPI stays neutral
ozone layer depletion and greenhouse effect, it is urgent to develop new environmentally
with regard to jurisdictional claims in
friendly and efficient working fluids with zero ozone depletion potential (ODP) and low
published maps and institutional affil-
global warming potential (GWP) [5,6]. Critical parameters are basic thermophysical prop-
iations.
erties of a working fluid consisting of critical temperature (Tc ), critical pressure (pc ), and
critical volume (vc ). Among them, critical temperature is not only the demarcation point
of a subcritical and supercritical cycle, but also the basis for estimating other physical
Copyright: © 2022 by the authors.
properties. Besides, the efficiency of a subcritical cycle is a function of Tc . For example,
Licensee MDPI, Basel, Switzerland. when the evaporation and condensation temperatures are given, working fluids with high
This article is an open access article Tc usually have better cycle efficiency [7]. Therefore, the accuracy of critical tempera-
distributed under the terms and ture largely determines the reliability of thermophysical property estimation and relevant
conditions of the Creative Commons computational design.
Attribution (CC BY) license (https:// Currently, there are two main approaches to obtain critical temperature: experimental
creativecommons.org/licenses/by/ measurement and theoretical estimation. The measurement methods can be categorized
4.0/). into direct and indirect way. For the direct observation method, the reappearance of the

Processes 2022, 10, 577. https://doi.org/10.3390/pr10030577 https://www.mdpi.com/journal/processes

Processes 2022, 10, 577 2 of 12

phase interface and the critical opalescence phenomenon are usually used for judging the
emergence of critical points [8]. On the other hand, the indirect calculation method usually
takes the specific heat peak on the Cv -T diagram and the inflection point of the isotherm
to find critical points [9]. However, an experiment alone cannot meet the requirements of
industrial application for Tc . Until now, only hundreds of experimental values of critical
temperature are reported, which is far from the demands of industrial production and
design [10]. Hence, it is of necessity to predict Tc by using theoretical methods. There are
several routes to calculate the critical temperature of working fluids.
The empirical correlation method is mainly based on the physical properties that are
common and easy to measure, such as boiling point and density. Guldberg et al. [11]
proposed a formula correlating the critical temperature with the boiling point. Veja-
hati et al. [12] proposed a simple exponential model to estimate the critical temperature of
alkanes. Klincewicz et al. [13] correlated the molecular weight and boiling point with Tc .
The calculation of empirical correlation is simple and fast, but lacks theoretical foundation
and has poor universality.
A group contribution method (GCM) is a generally applied method to estimate phys-
ical properties. As regards Tc , Riedel et al. [14] proposed a model using the group con-
tribution concept to estimate the critical temperature of organic compounds. Based on
their work, Lydersen et al. [15] proposed the first group contribution method with a de-
tailed group division and better estimation results. Joback et al. [16] further improved this
method. Considering that the previous methods did not consider the interaction between
adjacent groups and distinguish isomers, several new models have been developed later,
such as the secondary group contribution method proposed by Constantinou et al. [17],
group-interaction contribution method by Marrero et al. [18], and position contribution
method by Wang et al. [19]. It is convenient to apply a group contribution method with
complete functional group parameters. However, the group division is complex, and the
calculation is complicated, which slows down the calculation process of physical property.
Besides, molecular simulation (MS) is a numerical prediction method based on atomic
interaction [20]. Raabe [21] investigated the vapor-liquid phase equilibria of several bi-
nary mixtures of R-1234yf and R-1234ze(E) via Gibbs ensemble Monte Carlo simulation.
Yang et al. [22] reported the vapor-liquid equilibrium properties of R152a and its mixture by
Gibbs ensemble Monte Carlo simulation and molecular dynamics simulation. Cai et al. [23]
studied the evaporation process of R32/R152a by molecular dynamics method. More
reports [24–26] have shown that MS is a powerful method for predicting the properties of
materials. However, the prediction accuracy of MS heavily relies on the atomic interaction
potential model.
Recently, machine learning (ML) has become a popular method for physical property
estimation due to high accuracy [27]. Theoretically, a three-layer neural network can
approximate any rational function with any precision [28]. It is easy to set and implement
an ML model in computer programs without a given expression. Compared with previous
models, when a machine learning method is applied to predict the physical properties
of a working fluid, such as melting point [29], boiling point [30], density [31], and heat
capacity [32], it usually has better prediction accuracy. The application of ML on the
estimation of critical temperature has also been explored. Gharagheizi et al. [33] attempted
to apply group contribution as the input of an artificial neural network to estimate the
critical temperature of pure compounds. However, most of the existing group contribution
methods cannot effectively distinguish isomers, which inevitably affects the prediction
results. Therefore, the appropriate expression of a molecular structure is a prerequisite for
the construction of a machine learning model.
The development of cheminformatics provides a new idea for the programming
language expression of a molecule structure. In the 1980s, molecular fingerprints (MF)
appeared along with the study of similarity searching in medicinal chemistry [34]. MF uses
the Boolean value ‘1’ or ‘0’ to describe whether there is a specific substructure in a molecule.
Processes 2022, 10, 577 3 of 12

When it comes to the prediction of toxicity [35] and viscosity [36], taking MF as the input
feature of a machine learning algorithm can achieve satisfactory results.
Consequently, in this paper, four kinds of molecular fingerprints were used to represent
the structures of working fluid molecules, acting as the input of four ML algorithms. A
total of 16 different prediction models of Tc (16 = 4 MFs multiplied by 4 ML algorithms)
were established. Then a topological index was introduced to further optimize and obtain
the optimal model to predict the critical temperature of working fluids. The performance
of the proposed model was then compared with those of other previous methods.

2. Methods
2.1. ML Algorithms
Four types of supervised ML algorithms were used in this research, including sup-
port vector regression (SVR), decision tree (DT), random forest (RF), and multilayer
perceptron (MLP).

2.1.1. Support Vector Regression

Support vector regression is an algorithm that uses an appropriate kernel function
to map nonlinear data to a high dimensional feature space and transform the nonlinear
relationship into a linear form [37]. The accuracy of SVR depends on the optimization
of the model parameters, including the choice of kernel function, kernel parameter, tube
radius, and regularization parameter, which balance the model complexity and training
error. In this work, 10-fold cross validation paralleled with grid research was applied to
find the optimal combination of these parameters.

2.1.2. Decision Tree

A decision tree is composed of nodes and directed edges. There are two types of
nodes, the internal node, which represents a feature or attribute, and the leaf node, which
represents a category or certain value [38]. When DT is used for a regression task, it tests
a certain feature of the sample from the root node and assigns the sample to the child
nodes according to the test results. At this time, each child node corresponds to one of the
characteristic values. The samples are tested and distributed recursively until they reach
the leaf node. The prepruning of DT in this study was realized by 10-fold cross validation
and grid research to get the optimal parameters.

2.1.3. Random Forest

Random forest is a bagging algorithm based on multiple decision trees, which are
grown from different bootstrap samples of the training data. Bootstrap samples are gener-
ated from a random selection with replacement of the training samples during tree growth.
The data that are not chosen in the construction of forests is called ‘out-of-bag’ samples.
Each tree predicts its out-of-bag sample as the tree is added to the forest, and the average
of these results gives an overall evaluation [39]. Random forest usually has a better per-
formance than an individual tree, which helps to decrease the variance of a model. The
number of trees is obtained from drawing the learning curve, and the process of prepruning
is similar to that of DT.

2.1.4. Multilayer Perceptron

An artificial neural network is designed to simulate the structure and function of
a neural system for data processing; it can constantly adjust the weight of the chain
between the simulated neurons so that the entire network can better fit the relationship of
the training data. Multilayer perceptron is a feedforward neural network that simulates
nonlinear relationships through interconnected artificial neurons and complex topological
structures [40]. Its basic structure includes input layer, hidden layer, and output layer. Each
input node is connected to the output node through a weighted chain, which is used to
simulate the connection strength between neurons. Here, a multilayer perceptron with two
 the simulated neurons so that the entire network can better fit the relationship of the train-
ing data. Multilayer perceptron is a feedforward neural network that simulates nonlinear
relationships through interconnected artificial neurons and complex topological struc-
tures [40]. Its basic structure includes input layer, hidden layer, and output layer. Each
Processes 2022, 10, 577 4 of 12
input node is connected to the output node through a weighted chain, which is used to
simulate the connection strength between neurons. Here, a multilayer perceptron with
two hidden layers was built based on Keras. The optimal parameters, including the acti-
hidden layers waslearning
vation function, built based
rate,on
andKeras. Thelayer
hidden optimal parameters,
sizes, including
were obtained the activation
by random search,
function, learning rate,
and the remaining and hidden
parameters tooklayer sizes, were obtained by random search, and the
defaults.
remaining parameters took defaults.
2.2. Datasets
2.2. Datasets
The experimentally measured critical temperature data points of working fluids were
The
taken fromexperimentally measured
an open database critical
of the Design temperature data
Institute for pointsProperties
Physical of working(DIPPR)
fluids were
and
taken from
relevant an open[41].
literature database of the
The data Design
consist Institute
of the forpure
Tc of 155 Physical Properties
substances (DIPPR)
of working and
fluids.
relevant literature [41]. The data consist of the Tc of 155 pure substances of working fluids.
Based on this, the Tc databases of pure working fluids was built.
Based on this, the Tc databases of pure working fluids was built.
The pure chemicals basically cover the working fluids, which are commonly used in
The pure chemicals basically cover the working fluids, which are commonly used in
engineering practice. In order to improve the performance of the prediction model for
engineering practice. In order to improve the performance of the prediction model for pure
pure substances, the chemicals in the database are divided into three categories: (halogen-
substances, the chemicals in the database are divided into three categories: (halogenated)
ated) alkanes, (halogenated) alkenes, and ethers. Seventy percent data of each category
alkanes, (halogenated) alkenes, and ethers. Seventy percent data of each category were
were randomly selected to construct the training set, which was used to train the models,
randomly selected to construct the training set, which was used to train the models,
establishing a relationship between the molecular structure and the critical temperature.
establishing a relationship between the molecular structure and the critical temperature.
The remaining 30% data formed the testing set, which was used to evaluate the prediction
The remaining 30% data formed the testing set, which was used to evaluate the prediction
accuracy of the established model.
accuracy of the established model.
2.3. Feature
2.3. FeatureExtraction
Extractionand
andData
DataPreprocessing
Preprocessing
2.3.1. Molecular
2.3.1. Molecular Fingerprints
Fingerprints
The working
The working fluids
fluids were
were represented
represented by
by fingerprints
fingerprints derived
derived from
from their
their molecular
molecular
structures. Molecular
structures. Molecular fingerprints
fingerprints encode
encode aa structure
structure into
into an
anarray
arrayofofbit
bitstrings,
strings,the
the11s
s
and
and 00ss describe the presence
presence or
or absence
absence ofof particular
particular substructures
substructures inin the
themolecule.
molecule. AA
schematic
schematicofofmolecular
molecularfingerprints
fingerprintsisisshown
shownininFigure
Figure1.1.

Figure1.1.Schematic
Figure Schematicdiagram
diagramof
offingerprint
fingerprintof
ofaamolecule.
molecule.

In this
In this work,
work, four
four different
different lengths
lengths of
of fingerprints
fingerprints were
were chosen:
chosen: MACCS
MACCS (166(166 bits),
bits),
Pubchem
Pubchem (881 (881 bits),
bits), Extended
Extended (1024
(1024 bits),
bits), Morgan
Morgan (2048
(2048bits).
bits). All
All the
the fingerprints
fingerprints were
were
calculated
calculatedthrough
throughan anonline
onlinetransformer,
transformer,ChemDes
ChemDes[42].
[42].Since
Sincethe
thestructures
structuresof
ofworking
working
fluids
fluids are
are simple,
simple, fingerprint
fingerprint bits
bits with
with zero
zerovariance
variance were
werefiltered.
filtered. The
The MFs
MFs were
were then
then
applied to build regression models. Relevant information is listed in
applied to build regression models. Relevant information is listed in Table 1. Table 1.

Table 1. Length of fingerprints after variance threshold.

Fingerprints MACCS Pubchem Extended Morgan

Length 166 881 1024 2048
After removal 42 80 191 376
Processes 2022, 10, 577 5 of 12

2.3.2. Topological Index

Compared with a group contribution method, molecular fingerprints can uniquely
express the structural characteristics of most working fluid molecules. However, there are
still a few isomers that cannot be effectively distinguished. Therefore, the topological index
MTI0 was introduced to tell the difference between structural isomers, and the geometric
correction number GM was added to MTI0 to further distinguish the cis-trans isomers [43].
A topological index is an approach to quantify a molecular structure, which is a constant
of a molecular graph. It is obtained by performing certain numerical operations on the
matrices that characterize the graph [44]. The topological index S is calculated as follows:

DvVw = Dv DV Dw (1)
N
MTI0 = ∑ (vDvVw )i (2)
i =1
N h  i
GM = ∑ T
MGF DvVw + DvVw (3)
i =1 i

S = MTI0 + GM (4)
In Equations (1)–(4), Dv , DV , and Dw represent the valence matrix, vertex weight
matrix, and adjacency matrix of a working fluid molecule, respectively. N is the number of
atoms, v represents the valence vector, and MGF is a diagonal matrix that distinguishes cis
and trans isomers. The detailed process to obtain topological indices can be found in the
Supplementary Materials.

2.4. Model Validation

The performance of a model is evaluated by comparing the predicted and experimental
values using the following statistical parameters: correlation coefficient (R2 ), root mean
square error (RMSE), and average absolute deviation (AAD).
m  
∑ (yi − ŷi )2
 i =1
R2 = 1 − 

 m
 (5)
2
∑ ( yi − yi )
i =1
v
u m
u ∑ (yi − ŷi )2
u
t i =1
RMSE = (6)
m
1 m
 
 y − ŷi 

AAD = ∑ 100 ×  i  (7)
m i =1 yi 
In Equations (5)–(7), m is the number of samples; yi and ŷi are the measured and
predicted value of chemical i, respectively; and y is the mean value of data points.

3. Results and Discussion

3.1. Preliminary Screening of Models
Sixteen prediction models were obtained by employing the molecular fingerprints
of pure working fluids as the input of machining learning algorithms. The prediction
performance of each regression model in the testing subset is shown in Figure 2 (taking the
coefficient of determination R2 as a statistical indicator).
 reason accounting for it is that an Extended fingerprint works by counting all molecular
fragments along the path from a certain atom to a specified number of bonds and judging
whether these fragments are presented in a preassigned list of substructures. While in
ChemDes the maximum path length is set to 5 by default, this leads to those molecules
Processes 2022, 10, 577 with chain length ≥5 maybe having the same fingerprints (14 couples of fluids have6the
of 12
same Extended fingerprints in this work), therefore affecting the model performance.

Predictiveperformance
Figure2.2.Predictive
Figure performanceofofeach
eachmodel
modelon
onthe
thetesting
testingset.
set.

AsThe
theMACCS
longest fingerprint
fingerprints, is the shortest
Morgan canamong the four
distinguish the fingerprints; it cannot of
molecular structures distin-
all
the pure working fluids in the dataset. Nevertheless, an optimal R of 0.6661 obtained byas
guish the four couples of cis and trans isomers in the dataset. When 2 MACCS were input
2
anthe
SVRfeature
modelof ML algorithms,
is still the model
unsatisfactory. established
Small datasetsby andSVR exhibited
excessive an unsatisfactory
features may be theR
value
main of 0.5956
reason theyincause
a testing set. It isand
overfitting because
reducethis
thekind of short-length
robustness of models. fingerprint
Therefore,covers less
a Mor-
molecular
gan structure
fingerprint is not information, which ismodels
suitable to establish insufficient
for a to train with
dataset the prediction models, thus
limited samples.
limiting the performance of models.
Compared with other fingerprints, Pubchem achieved the best prediction perfor-
manceExtended
among allistheanfour
extension of the The
ML models. Chemistry
optimal Development
combination Kit fingerprint
is MLP [45].which
+ Pubchem,
2 value achieved by SVR is only 0.6807. The
Taking
Extended fingerprints as input, the highest R
attained an R2 of 0.8712. This proves that Pubchem can effectively characterize molecular
reason accounting for it is that an Extended fingerprint works by counting all molecular
information and construct the relationship between the molecular structure and the criti-
fragments along the path from a certain atom to a specified number of bonds and judging
cal temperature with a limited training dataset. A comparison between the predicted and
whether these fragments are presented in a preassigned list of substructures. While in
experimental values and the deviation of each point are shown in Figure 3. It can be seen
ChemDes the maximum path length is set to 5 by default, this leads to those molecules
that most of the data points have deviations less than 7.5%, only four samples’ deviations
with chain length ≥5 maybe having the same fingerprints (14 couples of fluids have the
same Extended fingerprints in this work), therefore affecting the model performance.
As the longest fingerprints, Morgan can distinguish the molecular structures of all the
pure working fluids in the dataset. Nevertheless, an optimal R2 of 0.6661 obtained by an
SVR model is still unsatisfactory. Small datasets and excessive features may be the main
reason they cause overfitting and reduce the robustness of models. Therefore, a Morgan
fingerprint is not suitable to establish models for a dataset with limited samples.
Compared with other fingerprints, Pubchem achieved the best prediction performance
among all the four ML models. The optimal combination is MLP + Pubchem, which
attained an R2 of 0.8712. This proves that Pubchem can effectively characterize molecular
information and construct the relationship between the molecular structure and the critical
temperature with a limited training dataset. A comparison between the predicted and
experimental values and the deviation of each point are shown in Figure 3. It can be seen
that most of the data points have deviations less than 7.5%, only four samples’ deviations
are more than 10%, which proves a relatively good prediction ability of the model. More
details about Pubchem fingerprints are provided in the Supplementary Materials.
From the point of ML algorithms, as a strong learner, SVR has a stable and satisfactory
prediction performance, and the R2 of SVR + Pubchem model in the testing set reaches
0.8184. Apart from MLP + Pubchem, the feedforward neural network MLP has a general
effect when other fingerprints are used as input. The ensemble algorithm RF, which is based
on the weak learner DT, has higher prediction accuracy compared with DT. However, the
performance of RF varies with different fingerprints. Thus, the comprehensive prediction
efficiency of the four models can be sorted as follows: SVR > MLP > RF > DT.
 RMSE = 24.19
550 10
R2 =
Processes 2022, 10, x FOR PEER
0.8712REVIEW 7 of 12

Predicted values (K)

500

Deviations (%)
5

450
0
Processes 2022, 10, 577 are more than 10%, which proves a relatively good prediction ability of the model. More
7 of 12
400
details about Pubchem fingerprints are provided in the Supplementary Materials.
-5
350

600 15
-10
300
RMSE = 24.19
550 10
250 -15
2
250 R = 0.8712
300 350 400 450 500 550 600 250 300 350 400 450 500 550 600
Predicted values (K)

500
Experimental values (K) Experimental values (K)

Deviations (%)
5

450
Figure 3. Predictive performance of MLP + Pubchem
0 model.
400
From the point of ML algorithms, as -5 a strong learner, SVR has a stable and satisfac-
350
tory prediction performance, and the R2 of SVR + Pubchem model in the testing set reaches
300 0.8184. Apart from MLP + Pubchem, the -10feedforward neural network MLP has a general

effect when other fingerprints are used as input. The ensemble algorithm RF, which is
250 -15
250 300 350
based
400 450
on 500
the weak
550
learner
600
DT, has higher250prediction
300 350accuracy
400 compared
450 500 with
550 DT.
600 How-
ever, the performance
Experimental values (K) of RF varies with different fingerprints. Thus,
Experimental values (K)the comprehensive
prediction efficiency of the four models can be sorted as follows: SVR > MLP > RF > DT.
Figure 3. Predictive performance of MLP + Pubchem model.
Figure 3.
3.2. Modification of Models
3.2. Modification of Models
From
Whenthe point ofthe
checking ML algorithms,
data as a strong
of a Pubchem learner,itSVR
fingerprint, has a that
is found stable and satisfac-
certain cis and
tory When checking
prediction the data of
performance, anda Pubchem
the R fingerprint,
2 of SVR + Pubchem it is model
found thatin certain
the testingcissetand trans
reaches
trans isomers of working fluids cannot be distinguished by Pubchem. Therefore, the pre-
isomers
0.8184. of working fluids cannot be distinguished by Pubchem. Therefore, the prediction
diction Apart
modelfromcan be MLP + Pubchem,
further the Based
optimized. feedforward
on the neural
analysisnetwork
above, aMLP has a general
topological index
model
effect can beother
when further optimized.
fingerprints Based
are used onas the analysis
input. The above, a topological
ensemble algorithm index
RF, S was
which is
S was considered to be added as a new feature of Pubchem. Then the modified fingerprint
considered
based on thetoweak
be added learner as aDT,
new feature
has higher ofprediction
Pubchem.accuracy
Then thecompared
modified fingerprint
with DT. was
How-
was used
used as theas input
the input of the
of the two
MLML algorithms with topperformance:
performance:SVR SVR andand MLP.
MLP. By By
ever, the
comparing performance
statistical of two
RF varies
parameters,
algorithms
with
the optimal
with
different top
fingerprints.
critical temperature Thus, the comprehensive
prediction model
comparing
prediction statisticalof
efficiency parameters,
the four the optimal
models can be critical
sorted temperature
as follows: SVR prediction
> MLP >model
RF > of aa
of
DT.
pure working
pure working fluid
fluid waswas finally
finally selected.
selected.
The prediction
The prediction results of SVR
results of SVR and andMLP
MLPafter aftermodification
modificationare areshown
shown inin Figures
Figures 4 and
4 and 5,
3.2.
5, Modification
respectively. ItofisModels
clear that the prediction accuracy of the models significantly improved
respectively. It is clear that the prediction accuracy of the models significantly improved
with When
with the checking the
the introduction
introduction of adata
of of a Pubchem
a topological
topological index.fingerprint,
index. The R
The R22 of it is +
SVR
of SVR +found
Pubchem
Pubchemthat certain
in cis and
in aa testing
testing set
set
trans isomers
increased from
increased of working
from 0.8184
0.8184 to fluids
to 0.8426, cannot
0.8426, while be
while that
that ofdistinguished
of MLP
MLP ++ Pubchem by Pubchem.
Pubchem reached Therefore,
reached 0.9143.
0.9143. the pre-
diction model can be further optimized. Based on the analysis above, a topological index
600 S was considered to be added as a new feature 20
of Pubchem. Then the modified fingerprint
was used as the input of the two ML algorithms with top performance: SVR and MLP. By
550 15 RMSE = 27.73
comparing statistical parameters, the optimal critical temperature prediction model of a
R2 = 0.8426
pure working fluid was finally selected. 10
Predicted values (K)

500
The prediction results of SVR and MLP after modification are shown in Figures 4 and
Deviations (%)

5
450
5, respectively. It is clear that the prediction accuracy of the models significantly improved
0
400
with the introduction of a topological index. The R2 of SVR + Pubchem in a testing set
increased from 0.8184 to 0.8426, while that -5 of MLP + Pubchem reached 0.9143.
350
-10
600 20
300 -15

550 15 RMSE = 27.73

250 -20
250 300 350 2400
R = 0.8426450 500 550 600 250 300 350 400 450 500 550 600 650
10
Predicted values (K)

500
Experimental values (K) Experimental values (K)
Deviations (%)

5
450
Figure 4. Predictive performance of the SVR + modified Pubchem model.
0
400
Comparing Figure 3 with Figure 5,-5it is obvious that the data points of a modified
350 model are more concentrated around the-10line y = x, the working fluids with deviations of
more than 7.5% reduced from 7 to 4, proving its better prediction ability. This shows that
300 -15
the selected topological index can well solve the problem that Pubchem cannot differentiate
250 cis and trans isomers, thereby improving -20 the overall prediction performance of models.
250 300 350 400 450 500 550 600
250 300 350 400 450
Thus, the 500 550 600
final critical temperature prediction model of pure working fluids was650
obtained.
Experimental values (K) Experimental values (K)
3.3. Comparisons with Existing Methods
Three existing group contribution methods (GCMs) and an empirical correlation for
the estimation of critical temperature were used to compare with the proposed model. The
 Processes 2022, 10, 577 8 of 12

Processes 2022, 10, x FOR PEER REVIEW 8 of 12

GCMs include the Lydersen, Joback, and Constantinou-Gani methods, and the differences
between them are the division of groups and whether the boiling point (Tb ) is needed in the
estimation. All the methods were applied to 120 pure working fluid samples with available
TFigure 4. Predictive performance of the SVR + modified Pubchem model.
b collected in this article.

600 15

RMSE = 21.74
550 10
2
R = 0.9143
Predicted values (K)

500 5

Deviations (%)
450 0

400
-5

350
-10

300
-15

250
250 300 350 400 450 500 550 600 300 350 400 450 500 550 600

Experimental values (K) Experimental values (K)

Figure 5. Predictive performance of

Figure 5. of the
the MLP
MLP ++ modified Pubchem model.

The comparison
Comparing Figureresults areFigure
3 with shown5,init Figure
is obvious6. It that
is noticed thatpoints
the data the Joback method,
of a modified
based on the experimental boiling points (T exp ), exhibits relatively good performance in
model are more concentrated around the line b y = x, the working fluids with deviations of
the
moreestimation
than 7.5% Tc . However,
ofreduced from the
7 toexperimentally
4, proving determined
its better values
prediction of Tb This
ability. may shows
not always
that
Processes 2022, 10, x FOR PEER REVIEW 9 of 12
est ) were considered in the Joback method,
be available. When the estimated values (T b
the selected topological index can well solve the problem that Pubchem cannot differen- its
accuracy showed a marked decline.
tiate cis and trans isomers, thereby improving the overall prediction performance of mod-
els. Thus, the final critical temperature prediction model of pure working fluids was ob-
60
tained.
8
AAD
3.3.7 Comparisons with Existing Methods RMSE 50

Three existing group contribution methods (GCMs) and an empirical correlation for
6
the estimation of critical temperature were used to compare
40 with the proposed model.
The5 GCMs include the Lydersen, Joback, and Constantinou-Gani methods, and the differ-
AAD/%

RMSE

ences between them are the division of groups and 30 whether the boiling point (Tb) is
4
needed in the estimation. All the methods were applied to 120 pure working fluid samples
with
3
available Tb collected in this article.
20
The comparison results are shown in Figure 6. It is noticed that the Joback method,
based
2 on the experimental boiling points (Tbexp), exhibits relatively good performance in
the estimation of Tc. However, the experimentally determined
10 values of Tb may not always
1
be available. When the estimated values (Tbest) were considered in the Joback method, its
accuracy
0 showed a marked decline. 0

sen exp ) est ) st) ond

)
Re
id ork
der (T b k(
Tb (fir sec cz- is w
Ly ac k a c C- G ( wi T h
Job Job C-G nce
Kli

Figure
Figure 6. 6.
TheThe comparisons
comparisons between
between thethe proposed
proposed model
model and
and previous
previous models.
models.

TheThe Constantinou-Gani
Constantinou-Gani (C-G)
(C-G) method
method does
does notnot need
need toto use
use a normal
a normal boiling
boiling point;
point; it it
involves two orders of groups, and the second groups are used to overcome
involves two orders of groups, and the second groups are used to overcome the limitationthe limitation
that
that thethe first
first groups
groups cannot
cannot distinguish
distinguish isomers
isomers in in molecular
molecular structures.
structures. The
The C-G
C-G method,
method,
taking second groups into account, has higher accuracy than single first
taking second groups into account, has higher accuracy than single first groups, and groups, and the
the
estimated value is more reliable. However, there are still some problems in the C-G method.
estimated value is more reliable. However, there are still some problems in the C-G
For example, the estimation performance of substances composed of very small molecules
method. For example, the estimation performance of substances composed of very small
is poor. Besides, many substances cannot be correctly separated because the division of
molecules is poor. Besides, many substances cannot be correctly separated because the
second groups is not comprehensive; thus only a small part of isomers can be discriminated
division of second groups is not comprehensive; thus only a small part of isomers can be
discriminated by the C-G method. The Klincewicz–Reid method correlates critical tem-
perature with molecular weight (Mw) and boiling point and gives a simple linear regres-
sion function. This function provides relatively reasonable estimations. However, as men-
tioned earlier, correlating molecular weights with Tc still lacks theoretical basis.
Processes 2022, 10, 577 9 of 12

by the C-G method. The Klincewicz-Reid method correlates critical temperature with
molecular weight (Mw ) and boiling point and gives a simple linear regression function.
This function provides relatively reasonable estimations. However, as mentioned earlier,
correlating molecular weights with Tc still lacks theoretical basis.
By means of comparison, the MLP + Pubchem model proposed in this paper effectively
solves the problem of distinguishing isomers and obtains the best accuracy on the premise
of not relying on experimental values of boiling points.

3.4. Distinction of Isomers

The C-G method can partially discriminate isomers with the aid of a second group.
The MLP + Pubchem model also has this ability depending on the molecular fingerprints
and topological index. Table 2 shows the estimation results of our proposed model as
opposed to the C-G (second) method.

Table 2. Prediction samples of isomers in pure working fluids.

Compounds S Texp/K Tcal1/K Deviation/% Tcal2/K Deviation/%

(Z)-1,2-Dichloroethylene 3846 507.25 518.97 2.3105 558.45 10.094
(E)-1,2-Dichloroethylene 2838 535.8 533.2 0.4853 558.45 4.228
(Z)-1,2,3,3,3-Pentafluoropropene 7758 379.25 376.13 0.822 435.30 2.003
(E)-1,2,3,3,3-Pentafluoropropene 6636 386.75 376.21 2.727 435.30 13.789
(Z)-2-Butylene 180 435.5 437.40 0.436 430.03 1.257
(E)-2-Butylene 68 428.6 426.33 0.530 430.03 0.333
1,1,1,2,2,3-Hexafluoropropane 8276 403.35 411.48 2.017 404.06 0.175
1,1,1,2,3,3-Hexafluoropropane 8741 412.45 411.01 0.349 494.52 19.897
1,1,1,3,3,3-Hexafluoropropane 8984 398.1 410.77 3.183 386.51 2.912
2,2,3-Trimethylpentane 424 563.5 573.40 1.757 566.24 2.736
2,2,4-Trimethylpentane 460 543.8 545.11 0.241 545.16 0.250
2,3,3-Trimethylpentane 412 573.5 573.06 0.077 594.42 3.648
2,3,4-Trimethylpentane 426 566.4 567.14 0.130 588.60 3.920

Tcal 1 and Tcal 2 denote the estimated critical temperature values of the MLP + Pubchem
model and the C-G (second) method, respectively. Apparently, while the C-G (second)
method cannot distinguish cis and trans isomers and achieve worse estimation perfor-
mance in structural isomers, the proposed model can recognize isomers with decent
prediction accuracy. The detailed calculation process and results can be found in the
Supplementary Materials.

4. Conclusions
In this work, molecular fingerprints, which are derived from molecular structures,
were used as the input of machine learning algorithms to establish the critical temperature
prediction models of working fluids. By analyzing the prediction performances, it is found
that Pubchem fingerprints can effectively characterize the molecular structures of work-
ing fluids when acting as the input of MLP to establish the ‘molecular structure-critical
temperature’ relationship. In order to address the problem that Pubchem fingerprints
cannot distinguish small parts of cis and trans isomers, the topological index S was intro-
duced as a new feature of fingerprints. The modified fingerprints have a better structure
recognition ability, leading to a significant improvement in the prediction performance
of the Pubchem + MLP model. The R2 of the testing set reaches 0.9143, with an average
deviation of 3.99%. Finally, the performance of the proposed model was compared with
those of other previous methods, and the results indicate that the present model is superior
to other approaches with respect to accuracy. This research provides a new approach to
build the ‘molecular structure-critical temperature’ relationship for working fluids. By
this ‘molecular structure-property’ method, the proposed model can also be applied in the
physical property prediction of other common working fluid systems.
Processes 2022, 10, 577 10 of 12

Supplementary Materials: The following supporting information can be downloaded at: https:
//www.mdpi.com/article/10.3390/pr10030577/s1, Table S1: Topological indices of working fluids.
Author Contributions: Conceptualization, Y.Q. and J.R.; methodology, S.R.; software, Z.H.; valida-
tion, Y.Q., S.R. and J.R.; formal analysis, J.R.; investigation, J.R.; resources, Y.Q.; data curation, Y.Q.;
writing—original draft preparation, Y.Q. and J.R.; writing—review and editing, Y.Q., S.R. and J.R.;
project administration, Y.Q.; funding acquisition, Y.Q. and J.R. All authors have read and agreed to
the published version of the manuscript.
Funding: This work is supported by the National Natural Science Foundation of China (No. 51876015).
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Acknowledgments: The authors wish to acknowledge Yu Liu (guest editor) and anonymous reviewers.
Conflicts of Interest: The authors declare no conflict of interest.

Nomenclature

AAD absolute average deviation

DT decision tree
GCM group contribution method
GWP global warming potential
ML machine learning
MLP multilayer perceptron
MF molecular fingerprint
ODP ozone depletion potential
ORC organic Rankine cycle
QSPR quantitative structure property relationship
R2 coefficient of determination
RMSE root mean square error
RF random forest
SMILES simplified molecular input line entry specification
SVR support vector regression
TI topological index

References
1. Wang, X.; Li, X.; Li, Q.; Liu, L.; Liu, C. Performance of a solar thermal power plant with direct air-cooled supercritical carbon
dioxide Brayton cycle under off-design conditions. Appl. Energy 2020, 261, 114359. [CrossRef]
2. Wang, S.; Liu, C.; Zhang, C.; Xu, X.; Li, Q. Thermo-economic evaluations of dual pressure organic Rankine cycle (DPORC) driven
by geothermal heat source. J. Renew. Sustain. Energy 2018, 10, 063901. [CrossRef]
3. Chen, X.; Liu, C.; Li, Q.; Wang, X.; Wang, S. Dynamic behavior of supercritical organic Rankine cycle using zeotropic mixture
working fluids. Energy 2020, 191, 116576. [CrossRef]
4. Wang, S.; Liu, C.; Li, Q.; Liu, L.; Zhang, C. Selection principle of working fluid for organic Rankine cycle based on environmental
benefits and economic performance. Appl. Therm. Eng. 2020, 178, 115598. [CrossRef]
5. Mahmoudi, A.; Fazli, M.; Morad, M.R. A recent review of waste heat recovery by Organic Rankine Cycle. Appl. Therm. Eng. 2018,
143, 660–675. [CrossRef]
6. Samudra, A.; Sahinidis, N.V. Design of Secondary Refrigerants: A Combined Optimization-Enumeration Approach. In Design for
Energy and the Environment; El Halwagi, M.M., Linninger, A.A., Eds.; CRC Press: Boca Raton, FL, USA, 2010; pp. 879–886.
7. Liu, B.T.; Chien, K.H.; Wang, C.C. Effect of working fluids on organic Rankine cycle for waste heat recovery. Energy 2004, 29,
1207–1217. [CrossRef]
8. Kay, W.B.; Pak, S.C. Determination of the critical constants of high-boiling hydrocarbons Experiments with gallium as a containing
liquid. J. Chem. Thermodyn. 1980, 12, 673–681. [CrossRef]
9. Kleinrahm, R.; Wagner, W. Measurement and correlation of the equilibrium liquid and vapour densities and the vapour pressure
along the coexistence curve of methane. J. Chem. Thermodyn. 1986, 18, 739–760. [CrossRef]
10. Su, W.; Zhao, L.; Deng, S. Group contribution methods in thermodynamic cycles: Physical properties estimation of pure working
fluids. Renew. Sustain. Energy Rev. 2017, 79, 984–1001. [CrossRef]
Processes 2022, 10, 577 11 of 12

11. Reid, R.C.; Sherwood, T.K.; Street, R.E. The properties of gases and liquids. Phys. Today 1959, 18, 739–760. [CrossRef]
12. Vejahati, F.; Nikoo, M.B.; Mokhatab, S.; Towler, B.F. Simple Correlation Estimates Critical Properties of Alkanes. Pet. Sci. Technol.
2007, 25, 1115–1123. [CrossRef]
13. Klincewicz, K.M.; Reid, R.C. Estimation of critical properties with group contribution methods. AIChE J. 1984, 30, 137–142.
[CrossRef]
14. Riedel, L. Additives Verfahren zur Abschätzung der kritischen Temperatur aus dem normalen Siedepunkt. Chem. Ing. Tech. 1952,
24, 353–357. [CrossRef]
15. Lydersen, L.A. Estimation of Critical Properties of Organic Compounds Vol. 2; Engineering Experiment Station Report 3; College of
Engineering, University of Wisconsin: Madison, WI, USA, 1955; p. 12S.
16. Joback, K.G.; Reid, R.C. Estimation of pure-component properties from group-contribution. Chem. Eng. Commun. 1987, 57,
233–243. [CrossRef]
17. Constantinou, L.; Gani, R. New group contribution method for estimating properties of pure compounds. AIChE J. 1994, 40,
1697–1710. [CrossRef]
18. Marrero-Morejón, J.; Pardillo-Fontdevila, E. Estimation of pure compound properties using group-interaction contributions.
AIChE J. 1999, 45, 615–621. [CrossRef]
19. Wang, Q.; Ma, P.; Jia, Q.; Xia, S. Position Group Contribution Method for the Prediction of Critical Temperatures of Organic
Compounds. J. Chem. Eng. Data 2008, 53, 1103–1109. [CrossRef]
20. Frenkel, D.; Smit, B. Understanding Molecular Simulation: From Algorithms to Applications, 2nd ed.; Academic Press: New York, NY,
USA, 2002.
21. Gabriele, R. Molecular Simulation Studies on the Vapor–Liquid Phase Equilibria of Binary Mixtures of R-1234yf and R-1234ze(E)
with R-32 and CO2 . J. Chem. Eng. Data 2013, 58, 1867–1873.
22. Dong, X.; Gong, M.; Li, X.; Wu, J.; Yang, Z. Molecular modeling and simulation of vapor–liquid equilibrium of the refrigerant
R152a and its mixture R152a+R32. Fluid Phase Equilibria 2015, 394, 93–100.
23. Cai, S.; Li, Q.; Liu, C.; Zhou, Y. Evaporation of R32/R152a mixtures on the Pt surface: A molecular dynamics study. Int. J. Refrig.
2020, 113, 156–163. [CrossRef]
24. Li, Q.; Xiao, Y.; Shi, X.; Song, S. Rapid Evaporation of Water on Graphene/Graphene-Oxide: A Molecular Dynamics Study.
Nanomaterials 2017, 7, 265. [CrossRef] [PubMed]
25. Hu, J.; Liu, C.; Li, Q.; Shi, X. Molecular simulation of thermal energy storage of mixed CO2 /IRMOF-1 nanoparticle nanofluid. Int.
J. Heat Mass Transf. 2018, 125, 1345–1348. [CrossRef]
26. Huo, E.; Liu, C.; Xu, X.; Li, Q.; Dang, C. The oxidation decom position mechanisms of HFO-1336mzz(Z) as an environmentally
friendly refrigerant in O2 /H2 O environment. Energy 2019, 185, 1154–1162. [CrossRef]
27. Maleki, A.A.H.; Mahariq, I. Machine learning-based approaches for modeling thermophysical properties of hybrid nanofluids: A
comprehensive review. J. Mol. Liq. 2021, 322, 114843. [CrossRef]
28. Mitchell, J. Machine learning methods in chemoinformatics. Wiley Interdiscip. Rev. Comput. Mol. Sci. 2014, 4, 468–481. [CrossRef]
[PubMed]
29. Venkatraman, V.; Evjen, S.; Knuutila, H.K.; Fiksdahl, A.; Alsberg, B.K. Predicting ionic liquid melting points using machine
learning. J. Mol. Liq. 2018, 264, 318–326. [CrossRef]
30. Deng, S.; Su, W.; Zhao, L. A neural network for predicting normal boiling point of pure refrigerants using molecular groups and a
topological index. Int. J. Refrig. 2016, 63, 63–71. [CrossRef]
31. Zolfaghari, H.; Yousefi, F. Thermodynamic properties of lubricant/refrigerant mixtures using statistical mechanics and artificial
intelligence. Int. J. Refrig. 2017, 80, 130–144. [CrossRef]
32. Gao, N.A.; Wang, X.B.; Xuan, Y.A.; Chen, G. An artificial neural network for the residual isobaric heat capacity of liquid HFC and
HFO refrigerants. Int. J. Refrig. 2019, 98, 381–387. [CrossRef]
33. Gharagheizi, F.; Eslamimanesh, A.; Mohammadi, A.H.; Richon, D. Determination of Critical Properties and Acentric Factors of
Pure Compounds Using the Artificial Neural Network Group Contribution Algorithm. J. Chem. Eng. Data 2011, 56, 2460–2476.
[CrossRef]
34. Kohberger, R. Similarity and Clustering in Chemical Information Systerns. Technometrics 1990, 32, 359–360. [CrossRef]
35. Domenico, A.; Daniela, T.; Kamel, M.; Felice, M.G.; Orazio, N. Prediction of Acute Oral Systemic Toxicity Using a Multifingerprint
Similarity Approach. Toxicol. Sci. 2019, 167, 484–495.
36. Yi, D.A.; Mc, B.; Chao, G.A.; Peng, Z.C.; Jw, A. Molecular fingerprint-based machine learning assisted QSAR model development
for prediction of ionic liquid properties. J. Mol. Liq. 2021, 326, 115212.
37. Bahadori, A.; Tavalaeian, M.; Soleimani, R.; Lee, M.; Hashemkhani, M. Prediction of the binary surface tension of mixtures
containing ionic liquids using Support Vector Machine algorithms. J. Mol. Liq. 2015, 211, 534–552.
38. Huo, Y.; Bouffard, F.; Joós, G. Decision tree-based optimization for flexibility management for sustainable energy microgrids.
Appl. Energy 2021, 290, 116772. [CrossRef]
39. Liaw, A.; Wiener, M. Classification and Regression by randomForest. R News 2002, 2, 18–22.
40. Cheshmberah, F.; Fathizad, H.; Parad, G.A.; Shojaei, S. Comparison of RBF and MLP neural network performance and regression
analysis to estimate carbon sequestration. Int. J. Environ. Sci. Technol. 2020, 17, 3891–3900. [CrossRef]
Processes 2022, 10, 577 12 of 12

41. Calm, J.M.; Hourahan, G.C. Physical, safety, and environmental data for current and alternative refrigerants. In Proceedings of
the 23rd International Congress of Refrigeration (ICR2011), Prague, Czech Republic, 21–26 August 2011.
42. Dong, J.; Cao, D.-S.; Miao, H.-Y.; Liu, S.; Deng, B.-C.; Yun, Y.-H.; Wang, N.-N.; Lu, A.-P.; Zeng, W.-B.; Chen, A.F. ChemDes: An
integrated web-based platform for molecular descriptor and fingerprint computation. J. Cheminform. 2015, 7, 60.
43. Schultz, H.P.; Schultz, T.P.; Schultz, E.B. Topological Organic Chemistry. 9. Graph Theory and Molecular Topological Indices of
Stereoisomeric Organic Compounds. J. Chem. Inf. Comput. Sci. 1995, 35, 864–870. [CrossRef]
44. Haruo, H. Topological Index. A Newly Proposed Quantity Characterizing the Topological Nature of Structural Isomers of
Saturated Hydrocarbons. Bull. Chem. Soc. Jpn. 1971, 44, 2332–2339.
45. Steinbeck, C.; Han, Y.; Kuhn, S.; Horlacher, O.; Willighagen, E. The Chemistry Development Kit (CDK): An Open-Source Java
Library for Chemo- and Bioinformatics. ChemInform 2003, 34, 493–500. [CrossRef]