Fuel 193 (2017) 315–321

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Full Length Article

Prediction of elemental composition of coal using proximate analysis

Lan Yi a,b, Jie Feng a, Yu-Hong Qin a, Wen-Ying Li a,b,⇑
a
Key Laboratory of Coal Science and Technology (Taiyuan University of Technology), Ministry of Education and Shanxi Province, Training Base of State Key Laboratory of Coal
Science and Technology Jointly Constructed by Shanxi Province and Ministry of Science and Technology, Taiyuan 030024, PR China
b
State Key Laboratory of Clean Energy Utilization, Zhejiang University, Hangzhou 310027, PR China

h i g h l i g h t s g r a p h i c a l a b s t r a c t

Select data points of four different

rank coals.

Understanding the relation between
ultimate analysis and proximate
analysis.

Validate the fitted correlations with
another set of data.

Offer a valuable tool to set up a coal-
thermal-conversion-process model.

a r t i c l e i n f o a b s t r a c t

Article history: Ultimate analysis is an important property for fuel utilization. The experimental determination of ulti-
Received 20 August 2016 mate analysis is sophisticated, long time consumed, and expensive, on the contrary, the proximate anal-
Received in revised form 8 December 2016 ysis can be run rapidly and easily. A variety of correlations to predict the ultimate analysis of biomass
Accepted 15 December 2016
using the proximate analysis have been appeared, while there exists a few number of correlations to esti-
mate the elemental compositions of coal using proximate analysis in the literature but were focused on
the predicted model or dependent on the heating value of coal. According to the proximate analysis of
Keywords:
four different ranks of coal, this study proposes a series of correlations which are classified to predict car-
Coal
Proximate analysis
bon, hydrogen, and oxygen compositions through using 300 data points and validated further by another
Ultimate analysis set of 40 data points. These correlations have the R2 of 0.95, 0.91, and 0.65 corresponding to the measured
Correlation contents of C, H, and O in anthracite, 0.93, 0.83, and 0.67 of C, H, and O in high-rank bituminous, 0.86,
0.61, and 0.71 of C, H, and O in subbituminous, and 0.92, 0.67, and 0.66 of C, H, and O in lignite, respec-
tively. The main merit of the correlations is the ability to estimate elemental composition of different
rank coals using the proximate analysis and thus offers a valuable tool to set up a coal-thermal-
conversion-process model.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction
⇑ Corresponding author at: Key Laboratory of Coal Science and Technology
(Taiyuan University of Technology), Ministry of Education and Shanxi Province, Utilization technique of coal is a major concern because of
Training Base of State Key Laboratory of Coal Science and Technology Jointly
severe environmental pollution and energy inefficient. To suggest
Constructed by Shanxi Province and Ministry of Science and Technology, Taiyuan
030024, PR China.
the clean and efficient utilization method of coal is essential.
E-mail address: ying@tyut.edu.cn (W.-Y. Li). Physicochemical properties of coal are critical parameters for the

http://dx.doi.org/10.1016/j.fuel.2016.12.044
0016-2361/Ó 2016 Elsevier Ltd. All rights reserved.
316 L. Yi et al. / Fuel 193 (2017) 315–321

utilization technology. For example, the higher heating value degree of coalification. And the content of sulfur varies with depo-
(HHV) that defines coal’s energy content is an important parameter sitional conditions when coal is formed. The nitrogen and sulfur
for designing the combustion system and evaluating pollution compositions are thus not considered in this paper.
compliance [1]. The ultimate analysis of coal is either a necessary The main objective of this study is to develop the correlations
factor to investigate the whole process of chemical conversion for predicting the contents of carbon, hydrogen, and oxygen of coal
and to predict the flow rate of flue gas and air quality in coal com- using proximate analysis.
bustion [2]. While the proximate analysis is a fuel property which
gives the chemical composition of coal and confirms the reason- 2. Methods
able use of coal. Focusing on the ultimate and proximate analyses,
it is possible to predict a large number of fuel parameters. Even if 2.1. Collection and selection of suitable data
there is no direct cause-effect link between composition and prop-
erties, the systematic transformations accompanying coalification The coal ranks are wide and the properties vary with the coal
often permit correlations [3]. rank to a great degree. Thus a great many original proximate and
Several relationships were developed using ultimate and prox- ultimate analyses of different rank coals, such as lignite, subbitu-
imate analyses in the past. Chelgani et al. [4] developed a predic- minous, high-rank bituminous, and anthracite, were chosen from
tion method to estimate coal grindability using multiple Properties, classification and utilization of China coal [21], and mes-
regression and artificial neural network models, taking the data sages of 300 materials whose proximate and ultimate analyses
of the proximate and ultimate analyses. A similar study was con- are listed in Table S1, were gotten in for the sake of the correlations
ducted to compute Hardgrove grindability index for Kentucky coals derivation, comprising 66 data for lignite, 74 data for subbitumi-
[5]. Neavel et al. [6] proposed the simplest equation to estimate the nous, 94 data for high-rank bituminous and 66 data for anthracite.
volatile matter, using ultimate analysis data for low-rank coals. In addition, 40 data were applied for correlations validation, as
Correlations was also generated by Chelgani and his colleagues listed in Table 1. The boundary of C, H, and O in data utilized for
[7]. The goal in that work was to simultaneously predict vitrinite the correlations validation and derivation are the same.
maximum reflectance and higher heating value of coal using ulti-
mate analysis. A work used the combination of proximate and ulti- 2.2. Methodology for fitting correlations
mate analyses with a neural network approach and multivariate
regression analyses to predict maceral composition for Indian coals The proximate analysis on a dry basis includes fixed carbon,
[8]. Attempts have been made to correlate 13C NMR chemical struc- volatile matters and ash. The ultimate analysis includes carbon,
tural analyses with coal properties such as the elemental composi- hydrogen, and oxygen and so on. The volatile matters content shows
tion and volatiles content [9]. By means of a semi-empirical the yield of volatile organic compounds after heating when under
method, Ko et al. [10] evaluated the complex relationship between prescribed conditions. Undergoing a variety of chemical reactions,
the potential maximum amount of tar and elemental analysis. The most of the mineral have shift and formed ash in the condition of
ultimate and proximate analyses are the essential elements to esti- high temperature. The formed ash also includes the unreacted min-
mate heating value of fuel. Various correlations for predicting eral. And the fixed carbon is the residue after deducting the ash from
higher heating value using ultimate and proximate analyses of the coke button. Both the volatile matters and fixed carbon contain
solid fuel were reported [11–13]. Channiwala et al. [1] had devel- carbon, oxygen along with hydrogen. However, the determination
oped the relationship to determine higher heating value based on of fixed carbon content on a dry basis is closely related with the
ultimate analysis of solid, liquid and gaseous fuels. Similarly, Par- ash and volatile matters contents, it is reasonable to assume a rela-
ikh et al. [14] and Komilis et al. [15] had estimated the heating tionship between the ultimate and proximate analyses.
value of biomass and municipal solid waste (MSW) from proximate According to the above analysis, it can be seen that carbon,
analysis data, respectively. All the existing works show that it is hydrogen and oxygen compositions are proportional to the relative
possible to estimate the element composition of fuels with the help contents of fixed carbon, volatile matters and ash, respectively.
of proximate analysis, which requires only common equipment Therefore, it is supposed that carbon (C, wt.%), hydrogen (H, wt.
and can be implemented by any of efficient scientists or engineers. %) and oxygen (O, wt.%) are separately a function of fixed carbon
Knowledge of proximate analysis thus can be used to determine (FC, wt.%), volatile matters (VM, wt.%), and ash (ASH, wt.%). The
element composition of fuels [2,16–19]. Vakkilainen et al. [16] proximate and ultimate analyses are based on a dry basis herein.
developed a correlation to estimate the relationship between ele- Firstly, each dependent variable (i.e., C, H, and O) was plotted as
ment composition and proximate analysis of black liquor. While opposed to each independent variable (i.e., FC, VM, and ASH),
Parikh et al. [17] and Shen et al. [18] proposed correlations to com- which allowed a rapid screening to decide the relative dependence
pute the ultimate analysis of biomass using proximate analysis. on each independent variable and enabled a visual detection of
The data considered by Parikh and his co-workers are the contents correlation forms. Secondly, an individual nonlinear correlation
of carbon, hydrogen, oxygen, volatile matter, and fixed carbon, was created for each independent variable, and a pseudo R2 value
neglecting the effect of ash content on the element composition. was computed to decide the contribution of each correlation. All
Considering this, adding the ash composition, Shen and his col- kinds of equation patterns were tested, yet experience indicated
leagues developed new correlations to estimate the carbon, hydro- that a quadratic polynomial lead to the optimum matching for
gen, and oxygen contents with a better estimation compared to parameters. Thirdly, according to the separate polynomial correla-
those proposed by Parikh et al. [17]. Recently, Nhuchhen [2] distin- tion, the pattern of the correlation was deduced between each ele-
guished raw and torrefied biomass, and then developed new corre- ment composition and the combination of the proximate analysis.
lations to estimate the contents of carbon, hydrogen, and oxygen The detailed steps are described in the following example for ele-
using proximate analysis. For the municipal solid waste, there also ment C of subbituminous.
exists a correlation to predict the element composition from prox-
imate analysis [19]. However, there were no correlations to esti- A case study for element C of subbituminous
mate the ultimate analysis of coal using the proximate analysis From the plots of element C of subbituminous versus each inde-
except the correlation proposed by Li et al. [20], which considered pendent variable in Fig. 1. it was showed the content of C depends
the effect of heating value on element composition. The content of observably on the value of ASH, VM, and FC. Then the patterns of
nitrogen in coal is less to 0.5–1.8%, being haphazard with the the optimum matching equations from the three plots (element
 L. Yi et al. / Fuel 193 (2017) 315–321 317

Table 1
Summary of coal materials data used for correlations validation.

No. Proximate analysis (wt.% by Ultimate analysis (wt.% by dry basis)

dry basis)
C H O N S
ASH VM FC Exp. Cal. Error (%) Exp. Cal. Error (%) Exp. Cal. Error (%)
Anthracite
1 7.23 7.92 84.85 85.14 84.72 0.50 3.66 3.65 0.27 1.67 1.75 4.63 1.30 1.00
2 7.64 7.11 85.25 85.17 84.66 0.60 3.46 3.44 0.55 1.63 1.69 3.93 1.22 0.89
3 7.58 8.26 84.16 84.59 84.17 0.49 3.72 3.70 0.57 1.88 1.83 2.78 1.24 1.00
4 6.15 9.51 84.34 86.39 85.33 1.23 3.88 3.95 1.88 1.80 1.79 0.61 1.24 0.54
5 5.98 11.19 82.83 85.89 85.07 0.95 3.95 4.07 3.10 1.93 1.95 1.07 1.49 0.77
6 8.03 4.99 86.98 86.36 85.51 0.99 2.92 2.76 5.53 1.43 1.49 3.96 0.75 0.51
7 4.14 10.46 85.40 87.74 87.50 0.27 3.90 4.10 5.07 1.69 1.58 6.47 1.67 0.86
8 4.65 13.35 82.00 86.63 86.46 0.20 4.15 4.01 3.46 1.90 1.93 1.63 1.07 1.61
9 5.01 7.33 87.66 89.07 87.52 1.75 3.53 3.65 3.36 1.49 1.49 0.09 0.69 0.20
10 4.66 7.98 87.36 88.94 87.65 1.45 3.86 3.80 1.65 1.58 1.49 5.46 0.80 0.15
High-rank bituminous
1 7.40 27.60 65.00 82.26 82.59 0.41 4.77 4.78 0.15 3.73 3.64 2.39 1.01 0.83
2 5.25 30.33 64.42 83.12 83.80 0.82 4.98 4.94 0.91 3.90 3.83 1.97 1.19 1.54
3 6.24 26.88 66.88 83.17 83.73 0.67 4.77 4.78 0.24 3.64 3.53 2.89 1.42 0.76
4 5.36 29.69 64.95 82.88 83.87 1.19 4.85 4.91 1.41 3.72 3.77 1.29 1.13 2.07
5 7.42 23.02 69.56 81.56 83.35 2.19 4.62 4.57 1.16 3.34 3.24 3.06 1.46 1.59
6 7.81 27.32 64.87 81.80 82.29 0.60 4.99 4.75 4.82 3.70 3.63 1.83 1.12 0.59
7 8.69 24.53 66.78 81.43 81.92 0.61 4.68 4.58 2.22 3.53 3.42 3.21 1.46 0.20
8 9.15 25.31 65.54 80.18 81.39 1.50 4.63 4.59 0.93 3.69 3.50 5.20 1.04 1.30
9 9.47 22.58 67.95 80.52 81.42 1.13 4.52 4.44 1.75 3.27 3.29 0.71 1.39 0.84
10 8.52 23.42 68.06 81.00 82.24 1.53 4.57 4.54 0.84 3.14 3.32 5.92 0.97 1.80
Subbituminous
1 4.38 40.76 54.86 77.01 76.34 0.88 5.13 5.24 2.33 11.21 11.82 5.45 1.33 0.95
2 7.12 36.09 56.79 78.32 77.47 1.08 5.12 5.13 0.18 8.19 8.14 0.61 0.97 0.29
3 8.29 39.12 52.59 76.57 75.25 1.72 5.39 5.09 5.53 8.43 8.92 5.81 1.02 0.30
4 8.00 35.88 56.12 77.37 76.87 0.65 5.17 5.20 0.62 8.33 7.91 4.95 0.86 0.27
5 6.52 39.22 54.26 76.22 76.37 0.20 5.14 5.11 0.63 10.00 9.54 4.63 1.41 0.70
6 8.59 36.13 55.28 76.35 76.24 0.15 5.00 5.18 3.65 8.14 7.98 2.05 1.26 0.65
7 6.30 28.74 64.96 81.24 81.41 0.21 5.02 5.06 0.73 5.63 5.41 3.96 1.31 0.50
8 5.89 30.87 63.24 80.86 80.76 0.12 4.93 5.18 5.10 6.29 6.42 2.08 1.22 0.81
9 6.14 30.83 63.03 79.89 80.59 0.88 5.14 4.94 3.91 6.40 6.33 1.12 1.55 0.87
10 6.68 30.87 62.45 79.28 80.14 1.08 4.92 4.96 0.92 6.03 6.23 3.38 1.33 1.75
Lignite
1 14.00 44.20 41.80 63.55 62.34 1.91 5.25 4.97 5.43 15.74 16.73 6.29 1.20 0.26
2 6.09 47.01 46.90 66.68 67.14 0.69 4.88 4.94 1.16 19.16 20.00 4.40 2.82 0.38
3 1.50 45.70 52.80 69.54 69.96 0.61 4.93 4.87 1.21 22.56 21.72 3.73 1.28 0.10
4 4.10 42.50 53.40 69.05 69.60 0.79 4.80 4.96 3.26 20.04 19.28 3.81 1.63 0.38
5 8.55 34.27 57.18 71.92 70.95 1.34 4.95 4.75 4.03 13.32 13.99 5.00 0.80 0.46
6 6.82 35.35 57.83 72.51 71.83 0.94 5.13 4.88 4.90 13.90 14.61 5.11 1.11 0.52
7 8.71 35.90 55.39 67.66 69.54 2.79 4.71 4.77 1.29 15.88 15.08 5.00 1.13 1.92
8 14.10 44.75 41.15 62.11 62.30 0.31 4.98 5.00 0.40 16.06 16.66 3.73 1.46 1.29
9 5.58 40.84 53.58 69.74 69.30 0.63 5.19 4.95 4.74 17.22 18.09 5.05 1.19 1.08
10 8.55 38.00 53.45 68.98 68.34 0.93 5.05 4.82 4.45 15.38 16.27 5.78 1.07 0.97

C of subbituminous versus ASH, element C versus VM, and element In order to choose the best suitable correlations, two
C versus FC) were added together, leading to Eq. (1): statistical parameters, namely average absolute error (AAE) and
average bias error (ABE), are applied to evaluate the parameters.
C ¼ x1 þ x2 ASH þ x3 ASH2 þ x4 VM þ x5 VM2 þ x6 FC þ x7 FC 2 ð1Þ At the same time, the coefficient of determination (R2) value is
where the xi is an empirical coefficient. All the optimum matching computed to define the degree of goodness of these presented cor-
equations were second-order polynomials, which lead to the modi- relations. The two statistical parameters and R2 are defined as
fied quadratic correlation of Eq. (1). The sum of quadratic error follows:

n  
between the measured result and the predicted result of element C
was minimized by optimizing the coefficients. A thorough computer 1X 
ValueP  ValueM   100%
AAE ¼ ð2Þ
algorithm has been presented to optimize the coefficients using a n i¼1  ValueM 
generalized method of least squares. The same process was repeated
n  
for the remaining dependent parameters, including H and O in sub-
bituminous, C, H, and O in anthracite, C, H, and O in high-rank bitu- 1X ValueP  ValueM
ABE ¼  100% ð3Þ
minous, C, H, and O in lignite. Parallel correlations were obtained. n i¼1 ValueM

2.3. Evaluation of correlations Xn 2

2
ðValueM  ValueP Þ
R ¼1X i¼1
 2 ð4Þ
The correlation is considered to be the optimum matching of a n
i¼1
ValueM  ValueM
specified set of data points if the prediction error tends to zero [22].
318 L. Yi et al. / Fuel 193 (2017) 315–321

results whereas the ABE represents the degree of underestimation

and overestimation. The R2 value used for nonlinear regression
shows how well the correlations means after removing the effect
of the mean of the dependent variable [9,23].

3. Results and discussion

3.1. Derivation of the correlations

The deduced correlations are shown in Eq. (5):

y ¼ a1 þ a2 ASH þ a3 ASH2 þ a4 VM þ a5 VM 2 þ a6 FC þ a7 FC 2 ð5Þ

where y stands for the elements of C, H, and O. The error analysis of

the fitting correlations is listed in Table 2. It can be seen in Table 2
that the coefficients of determination values for carbon correlations
show an obvious higher degree of fitness than those for hydrogen
and oxygen correlations. It may be due to that the carbon content
in coal regularly and evenly changes with the coalification process,
but the hydrogen content in lignite and bituminous has a little
change and the oxygen content in most rank coals will be affected
by the oxidation and weathering. The negative ABE values mean
the correlations faintly underestimate the ultimate analysis, the
opposite results for the positive ABE values. All the negative ABE
values in Table 2 indicate most of the predictions are underesti-
mated, which also explains why there are more negative error val-
ues than positive values in Table 1. The AAE values for oxygen are
obviously greater than those for the other two elements, which
may be due to that the acquisition of original oxygen content data
is by difference and the measuring errors of ash, carbon, hydrogen,
nitrogen, and sulfur are all added to oxygen. The AAE values for oxy-
gen in anthracite and subbituminous are more than 10% and up to
12.31% and 11.85%, respectively.
It is noticed that four correlations for different types of coal
other than one correlation for all coals are proposed herein. As it
is shown in Table S2, the predictions using one correlation for all
materials are less satisfactory than using four different correla-
tions. Ground on one correlation, though the C predictions are
close to the measured values, the errors for hydrogen content are
unstable in the four types of coal and the errors for oxygen content
are even too large to use the calculations. While the predictions
from four correlations show agreeable results. Four different corre-
lations are thus adopted herein.
Table 3 presents the coefficients of the correlations. The quadra-
tic correlation indicates the relation between these two analyses is
not simple linear association. The values of a3 , a5 , and a7 point out
that the quadratic term makes a greater contribution to the corre-
lations. It is remarkable that the sign of all coefficients in the cor-
relations does not mean any dependence relationship among the
variables. The independent variables of FC, VM, and ASH in the cor-
relations are not completely independent. Only two independent
variables are independent. Because on the dry basis, the sum of
FC, VM, and ASH is one hundred, when two of them have values,
the value of third variable can be by difference.
According to the proposed correlations, three groups of three-
Fig. 1. Plots of element C of subbituminous versus (a) ASH, (b) VM, and (c) FC. dimensional diagrams whose Z axis is carbon content, X and Y axis
are two of FC, ASH, and VM, were shown in Fig. 2. This figure intu-
itively displays the relationship between the dependent variable
and independent of the independent variables. As it is shown, an
In Eqs. (2)–(4), subscripts P and M denote the predicted and increase in FC significantly increases C while an increase in ASH
measured element composition results, respectively. ValueM decreases C. The H increases with the increase of VM and decreases
denotes the average measurement. And n is the amount of data with the increase of ASH. The influence of FC to O is more obvious
points applied for the nonlinear regression analysis. The AAE indi- than that of VM and ASH. A summary list of the ranges for all inde-
cates the degree of closeness between the predicted and measured pendent and dependent variables are shown in Table 4.
 L. Yi et al. / Fuel 193 (2017) 315–321 319

Table 2
Error analysis of the fitting correlations.

ABE (%) AAE (%) R2

Anthracite C 0 0.53 0.95
H 1.67 7.81 0.91
O 2.25 12.31 0.65
High-rank bituminous C 0 0.51 0.93
H 0.01 2.52 0.83
O 1.39 9.72 0.67
Subbituminous C 0.01 0.93 0.86
H 0.07 2.16 0.61
O 1.9 11.85 0.71
Lignite C 0.05 1.87 0.92
H 0.23 4 0.67
O 0.7 7.24 0.66

Table 3
Coefficients of quadratic correlations.

Anthracite High-rank bituminous

C H O C H O
a1 3.06540E+05 3.73113E+04 4.74700E+04 1.43398E+05 2.56156E+04 1.64560E+04
a2 3.06661E+03 3.72793E+02 4.75792E+02 1.43461E+03 2.56084E+02 1.64567E+02
a3 2.29439E02 4.80862E03 9.08172E03 1.38486E02 5.64800E03 2.46841E03
a4 3.06631E+03 3.72118E+02 4.75776E+02 1.43578E+03 2.56031E+02 1.64687E+02
a5 4.35868E02 3.05899E02 6.79290E03 1.25349E02 1.01905E03 2.57626E03
a6 3.06336E+03 3.73521E+02 4.73522E+02 1.43388E+03 2.56126E+02 1.64321E+02
a7 1.06054E02 4.20091E03 1.19708E02 1.06647E02 1.69152E05 2.40549E03
Subbituminous Lignite
C H O C H O
a1 4.05212E+04 4.28121E+04 6.83050E+05 5.85580E+03 4.27125E+02 6.33501E+03
a2 4.03482E+02 4.27765E+02 6.83288E+03 5.73735E+01 4.38716E+00 6.21537E+01
a3 8.56030E02 2.07849E02 1.22176E01 2.20361E02 3.80616E03 2.57393E02
a4 4.03029E+02 4.27840E+02 6.83179E+03 5.84226E+01 4.68554E+00 6.33546E+01
a5 1.93969E02 2.51216E03 1.77902E02 1.31499E02 3.93351E03 2.55578E03
a6 4.05885E+02 4.28297E+02 6.82864E+03 5.89345E+01 3.95314E+00 6.51083E+01
a7 1.68029E02 2.10775E03 1.96434E02 1.95368E02 3.86801E03 2.30446E02

3.2. Validation of the correlations correlations. In Fig. 3(a), the predictions of anthracite using Nhuch-
hen’s correlation are close to the straight line while the other exist-
In order to verify the availability of the deduced correlations, 10 ing correlations’ results are badly underestimated. And in Fig. 3(b),
data points of each kind of coal listed in Table 1 were tested, the predictions of bituminous and lignite using Parikh et al.’s cor-
accompanying with percentage errors. The errors for carbon con- relations are much closer to the straight line than the other two
tent in different types of coal all tend to be around zero and for existing correlations. In Fig. 3(c), the results of Shen et al.’s and Par-
hydrogen content are small as well. Although the original mea- ikh et al.’s correlations are also badly overestimated. The result
sured oxygen contents are by difference, the predicted results per- errors using Nhuchhen’s correlations are 5.73%, 28.42% and
form acceptable error ranges. The comparison between predicted 162.14% for C, H, and O, respectively. The errors using Shen
results and measured data of element composition has also been et al.’s correlations are 31.75%, 22.93%, and over 200% for C, H,
vividly shown in Fig. 3. The predicted values being closer to the and O, respectively. The errors using Parikh et al.’s correlations
straight line indicates a better predictive ability of the correlations. are 31.45%, 12.2%, and over 200% for C, H, and O, respectively.
In Fig. 3(a)–(c), it can be seen that the predicted C, H, and O are by While the errors using the proposed correlations are 0.01%,
and large close to the straight line for all four correlations. The 0.62%, and 0.23% for C, H, and O, respectively. The comparison
errors of carbon content are within the error band of ±2%. Several shows that the accuracy of the presented correlations for C, H,
hydrogen content errors reach more than to ±5%, the others are and O is better, which indicates that the application range of these
all within the range of ±4%. The biggest oxygen content error is correlations is much important and any correlation should be used
6.47%. in its own application range.

3.3. Comparison with the correlations from literature 3.4. Utility of the correlations

For further validation of the presented correlations, compar- These developed correlations have been proved the validity and
isons between the predictions of existing correlations and of the merits. Two of the correlations’ applications can be summarized
presented correlations are established, shown in Fig. 3. The data herein: (i) the correlations can be used to estimate the contents of
points listed in Table 1 were examined. The predictions and per- carbon, hydrogen, and oxygen of coal using the proximate analysis.
centage errors for different existing correlations are shown in (ii) They can be used to predict the ultimate analysis of coal char in
Table S3. It can be seen in Fig. 3 that the predicted results of the the performance modeling processes of combustion, gasification
existing correlations are much worse than those of the presented and pyrolysis. Pyrolysis is an analogous process of deoxidation,
320 L. Yi et al. / Fuel 193 (2017) 315–321

Fig. 2. Three-dimension relationship between ultimate analysis and proximate analysis.

Table 4
Range of values used in correlations.

Constituent (wt.%, dry basis) ASH VM FC C H O

Anthracite Maximum 14.22 13.35 91.60 90.80 4.27 2.96
Minimum 4.14 1.87 78.49 79.22 0.91 1.02
High-rank bituminous Maximum 17.60 31.26 82.67 88.46 5.07 4.52
Minimum 2.68 11.46 60.45 78.62 3.76 1.99
Subbituminous Maximum 18.57 40.76 68.27 83.68 5.45 11.21
Minimum 2.92 25.65 49.04 68.58 4.07 4.81
Lignite Maximum 29.08 50.09 59.34 73.91 6.23 24.40
Minimum 1.50 33.48 35.64 49.70 4.25 12.97

dehydrogenation, and carbon-rich [24]. The char obtained from fuel 4. Conclusion
pyrolysis can be considered as a higher-rank fuel. For instance, the
biomass char and coal char can be used as the material together Nonlinear correlations were proposed to predict the element
with biomass or coal for computing the higher heating value composition of coal using a great many data having extensively
(HHV) using proximate and ultimate analyses [1,14], while our varying proximate and ultimate analyses and including four differ-
research using these correlations to analyze the ultimate analysis ent rank coals. Within the stated range of ASH, VM, FC, C, H, and O as
of pyrolysis char is underway. wt.% on a dry basis, these proposed correlations can be expected to
 L. Yi et al. / Fuel 193 (2017) 315–321 321

in lignite, respectively. These correlations were also validated by

another set of data with the four coal types, giving a fairly good pre-
diction effect. The key merit of these correlations is a rapid and easy
method to predict the element composition, when sophisticated
measurement for elemental components is not available.

Acknowledgements

This work was financially supported by the Natural Science

Foundation of China (Grant No.: U1361202; 21206104). Prof. Dr.
Stanislav Vassilev is appreciated for making a thorough
proofreading.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, in

the online version, at http://dx.doi.org/10.1016/j.fuel.2016.12.044.

References

[1] Channiwala S, Parikh P. A unified correlation for estimating HHV of solid, liquid
and gaseous fuels. Fuel 2002;81:1051–63.
[2] Nhuchhen DR. Prediction of carbon, hydrogen, and oxygen compositions of
raw and torrefied biomass using proximate analysis. Fuel 2016;180:348–56.
[3] Mathews JP, Krishnamoorthy V, Louw E, Tchapda AH, Castro-Marcano F, Karri
V, et al. A review of the correlations of coal properties with elemental
composition. Fuel Process Technol 2014;121:104–13.
[4] Chelgani SC, Hower JC, Jorjani E, Mesroghli S, Bagherieh A. Prediction of coal
grindability based on petrography, proximate and ultimate analysis using
multiple regression and artificial neural network models. Fuel Process Technol
2008;89:13–20.
[5] Jorjani E, Hower JC, Chelgani SC, Shirazi MA, Mesroghli S. Studies of
relationship between petrography and elemental analysis with grindability
for Kentucky coals. Fuel 2008;87:707–13.
[6] Neavel RC, Smith SE, Hippo EJ, Miller RN. Interrelationships between coal
compositional parameters. Fuel 1986;65:312–20.
[7] Chelgani SC, Mesroghli S, Hower JC. Simultaneous prediction of coal rank
parameters based on ultimate analysis using regression and artificial neural
network. Int J Coal Geol 2010;83:31–4.
[8] Khandelwal M, Singh T. Prediction of macerals contents of Indian coals from
proximate and ultimate analyses using artificial neural networks. Fuel
2010;89:1101–9.
[9] Genetti D, Fletcher TH, Pugmire RJ. Development and application of a
correlation of 13C NMR chemical structural analyses of coal based on
elemental composition and volatile matter content. Energy Fuels
1999;13:60–8.
[10] Ko GH, Sanchez DM, Peters WA, Howard JB. Correlations for effects of coal type
and pressure on tar yields from rapid devolatilization. Symp (Int) Combust
1989;22:15–24.
[11] Yin CY. Prediction of higher heating values of biomass from proximate and
ultimate analyses. Fuel 2011;90:1128–32.
[12] Sheng C, Azevedo J. Estimating the higher heating value of biomass fuels from
basic analysis data. Biomass Bioenergy 2005;28:499–507.
[13] Mesroghli S, Jorjani E, Chelgani SC. Estimation of gross calorific value based on
coal analysis using regression and artificial neural networks. Int J Coal Geol
2009;79:49–54.
[14] Parikh J, Channiwala S, Ghosal G. A correlation for calculating HHV from
proximate analysis of solid fuels. Fuel 2005;84:487–94.
[15] Komilis D, Evangelou A, Giannakis G, Lymperis C. Revisiting the elemental
composition and the calorific value of the organic fraction of municipal solid
wastes. Waste Manage 2012;32:372–81.
[16] Vakkilainen EK. Estimation of elemental composition from proximate analysis
of black liquor. Pap Puu 2000;82:450–4.
[17] Parikh J, Channiwala S, Ghosal G. A correlation for calculating elemental
composition from proximate analysis of biomass materials. Fuel
2007;86:1710–9.
[18] Shen J, Zhu S, Liu X, Zhang H, Tan J. The prediction of elemental composition of
biomass based on proximate analysis. Energy Convers Manage
Fig. 3. Comparison of the predicted value with the measured data for (a) carbon, (b) 2010;51:983–7.
hydrogen, and (c) oxygen. [19] Chakma S, Vaishya RC, Yadav AK. Modeling chemical compositions of
municipal solid waste. Environ Geotech 2016;3:65–77.
[20] Li C, Liu D, Li F. Study on relevant equations for two kinds of coal quality
analysis. Coal Geol China 2004;16:6–8.
complete reasonable estimations of element composition with the [21] Chen P. Properties, classification and utilization of China
specified average bias error limit of 0, 1.67%, and 2.25% for C, coal. Beijing: Chemical Industry Press; 2007.
[22] Nhuchhen DR, Salam PA. Estimation of higher heating value of biomass from
H, and O in anthracite, 0, 0.01%, and 1.39% for C, H, and O in proximate analysis: a new approach. Fuel 2012;99:55–63.
high-rank bituminous, 0.01%, 0.07%, and 1.9% for C, H, and O [23] Wu J. Applied probability and statistics. Beijing: Higher Education Press; 2001.
in subbituminous, and 0.05%, 0.23%, and 0.7% for C, H, and O [24] Xie KC. Coal structure and its reactivity. Beijing: Science Press; 2002.