Content uploaded by Jan Skvaril
Author content
All content in this area was uploaded by Jan Skvaril on Jan 21, 2021
Content may be subject to copyright.
1876-6102 © 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the scientific committee of the 8th International Conference on Applied Energy.
doi: 10.1016/j.egypro.2017.03.476
Energy Procedia 105 ( 2017 ) 1309 – 1317
ScienceDirect
The 8th International Conference on Applied Energy – ICAE2016
Fast Determination of Fuel Properties in Solid Biofuel
Mixtures by Near Infrared Spectroscopy
Jan Skvaril*, Konstantinos Kyprianidis, Anders Avelin, Monica Odlare, Erik
Dahlquist
Future Energy Center, School of Business, Society and Engineering, Mälardalen University, Västerås 721 23, Sweden
Abstract
This paper focuses on the characterization of highly variable biofuel properties such as moisture content, ash content
and higher heating value by near-infrared (NIR) spectroscopy. Experiments were performed on different biofuel sample
mixtures consisting of stem wood chips, forest residue chips, bark, sawdust, and peat. NIR scans were performed using
a Fourier transform NIR instrument, and reference values were obtained according to standardized laboratory methods.
Spectral data were pre-processed by Multiplicative scatter correction correcting light scattering and change in a path
length for each sample. Multivariate calibration was carried out employing Partial least squares regression while
absorbance values from full NIR spectral range (12,000–4000 cm-1), and reference values were used as inputs. It was
demonstrated that different solid biofuel properties can be measured by means of NIR spectroscopy. The accuracy of
the models is satisfactory for industrial implementation towards improved process control.
© 2016 The Authors. Published by Elsevier Ltd.
Selection and/or peer-review under responsibility of ICAE
Keywords: Ash content; biofuels; higher heating value, moisture content, Near infrared spectroscopy, NIRS.
1. Introduction
Solid biofuels are characterized by highly variable properties which are mostly dependent on the type of
biomass, place of origin, fuel pre-processing and handling techniques, etc. Therefore, its utilization in
thermochemical energy conversion processes i.e. combustion, gasification, pyrolysis and torrefaction
becomes rather challenging, and it may result in increased production of emissions, strong process
instabilities, lower conversion efficiency, and lower final product quality, etc. The knowledge of the actual
fuel composition of the material fed into the energy conversion unit is limited to direct or indirect continuous
moisture content measurements and periodic fuel sampling providing the elementary composition. Since
the operating conditions of a process unit require continuous regulation, this information is not sufficient
for process control and performance optimization, especially when considering strongly heterogeneous
fuels. For this reason, a fast and non-destructive measurements method for fuel property characterization is
needed.
Available online at www.sciencedirect.com
© 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the scientific committee of the 8th International Conference on Applied Energy.
1310 Jan Skvaril et al. / Energy Procedia 105 ( 2017 ) 1309 – 1317
Accurate on-line measurements can be realized by introducing the near-infrared spectroscopy (NIRS)
technique. This inexpensive, contactless and rapid optical method enables qualitative and quantitative
characterization of solid and liquid materials. The method is based on vibrational spectroscopy and
measures interactions between electromagnetic radiation and the chemical bonds of the matter in the near-
infrared (NIR) spectral region. In biomass, electromagnetic radiation interacts with structural molecular
groups such as C-H, O-H, N-H, S-H, C=O, C=H and C=C. The extraction of the information from spectra
is done by applying chemometrics which is a discipline that uses mathematical and statistical methods to
provide maximum chemical information by analyzing NIR spectra. It is usually done by applying spectral
pre-processing techniques which enhance the interesting information included in the spectrum (i.e. chemical
information related to vibration of bonds). The quantitative analysis e.g. multivariate or univariate
regression method such as Partial least squares regression (PLS-R) is then applied in order to determine
quantifiable properties of the material.
The NIRS technique has been successfully used for several decades in the pharmaceutical sector as a
process analytical technology [1], more recently also in the food and feed industry [2] and the pulp and
paper industry [3]. Its application for assessment of biomass and biofuel quality parameters in energy
conversion technology processes is an emerging topic [4]. Lestander and Rhén [5] developed spectroscopy
models for prediction of moisture (material artificially moisturized), ash content and heating value and
achieved reasonable accuracy. Their work was however carried out on rather homogenous ground samples
of Norway spruce. Another interesting study was presented by Yang and Sheng [6]. They used vis-NIR
spectroscopy to measure ash, volatile matter, fixed carbon contents, and heating of three types of biochar
produced from pine wood, cedar wood, and cotton stalk. The assignment of vibrational bands (which is
important for property prediction) was not shown. Eriksson et al. [7] presented a characterization of Scots
pine stump–root biomass as feedstock for gasification. The authors presented the development of a
spectroscopy calibration for predicting the content of moisture and extractives content which is positively
correlated to the energy content of wood. Preece et al. [8] used visible and NIRS data acquired by a hand-
held instrument to predict the higher heating value (HHV) of solid manure samples. The authors concluded
that the resulting error of 1,7% is more than adequate for the industrial application. Posom et al. used NIRS
for the evaluation of moisture content and heating value of Jatropha curcas kernels [9] and Leucaena
leucocephala pellets [10] and achieved promising results. The authors also performed extensive assignment
of vibrational bands based on the plots of X-loading weights and regression coefficients resulting from PLS-
R. All of these studies however were limited to relatively homogeneous or pure materials and did not focus
on fuel mixtures. Therefore the authors argue that following research questions need to be addressed:
x Is it possible to correlate NIR absorbance spectra of solid raw biofuel sample mixtures to reference
values of given fuel properties determined by standard laboratory methods?
x What prediction accuracy can be achieved by means of multivariate linear regression modeling
using NIR absorbance spectra as predictors and reference values i.e. fuel properties as responses?
The objective of this work is to perform spectroscopic measurements on solid raw biofuel samples and
visually analyze acquired NIR spectra. Then, the acquisition of reference values (i.e. fuel properties of
interest) has to be carried out by standard laboratory methods. The biofuel properties included in this study
are moisture content, ash content and heating value. Moreover, the acquired raw NIR spectra were pre-
processed by an appropriate technique. Finally, multivariate statistical regression models describing the
correlation between spectra and reference values were developed. The spectroscopy background was
justified for the developed models. It consisted of the identification of which spectral variables are most
important for prediction of responses i.e. fuel properties and their assignment to reasonable molecular
vibrations of chemical constituents.
Jan Skvaril et al. / Energy Procedia 105 ( 2017 ) 1309 – 1317 1311
Nomenclature
ܾ regression coefficient
FT-NIR Fourier Transform Near Infrared
HHV Higher heating value
MSC Multiplicative scatter correction
݉ number of variables
NIR Near infrared
NIRS Near infrared spectroscopy
݊ number of samples
PLS-R Partial least squares regression
RMSE Root-mean-square error
ܴଶ coefficient of determination
ݕ reference Y value
ݕො predicted Y value
ݕത mean of reference Y values
2. Materials and methods
2.1. Samples collection and experimental data acquisition
There were five biomass types included in this study: (i) stem wood chips (pine and spruce), (ii) bark,
(iii) forest residue chips, (iv) sawdust, and (v) peat. The biomass was collected from local fuel processing
companies in Västmanland region, Sweden. The samples were created by mixing up the different types of
biomass in varying proportions. The mixed substances contained particles of different sizes varying from
5 mm to 50 mm.
NIR spectral data were acquired in diffuse reflectance mode using a Fourier transform-NIR (FT-NIR)
spectrometer equipped with an illumination head. The head was fixed at a distance of 17 cm from the
sample which was moving on a turntable at a velocity of approximately 1 m·s-1. This enabled the replication
of the actual movement of biomass on conveyor belt. NIR absorbance spectra (relative absorbance) were
collected in the range of 12,000–4000 cm-1 with a wavenumber resolution of 8 cm-1. The temperature of
the ambient air during the spectra acquisition was kept at 20±1 °C. The average of 32 scans for each sample
was then used as an input in the multivariate calibration; this was done in order to minimize random noise
in the spectra measurements i.e. to increase the signal to noise ratio.
The determination of reference values i.e. fuel properties was carried out in accordance with standardized
laboratory procedures. The moisture content in mass percentage was determined according to standard EN
ISO 18134-2:2015. The ash content in mass percentage was determined according to standard EN ISO
18122:2015. Both procedures are based on thermogravimetric measurements where samples are thermally
treated at 105° respectively 550 °C. The heating value was determined on moisture free samples using a
bomb calorimeter according to standard EN 14918:2010. The values are reported as higher heating value
(HHV) in MJ/kg.
2.2. Data pre-processing and multivariate calibration
The NIR spectra acquired through the contactless measuring head were affected by light scattering.
Therefore, multiplicative scatter correction (MSC) [11] was applied to the spectra. This pre-processing
1312 Jan Skvaril et al. / Energy Procedia 105 ( 2017 ) 1309 – 1317
method corrects the scatter level in all the spectra of a given dataset to the level of an average spectrum of
the set considered as the ideal sample.
Partial least squares regression (PLS-R) [12,13] was used for the development of the regression models.
This multivariate linear regression technique estimates the relationship between independent X (predictors
i.e. NIR absorbance spectra) and dependent Y variables (responses i.e. fuel properties) and finds latent
variables (i.e. PLS factors) in an independent data matrix that will best predict latent variables in dependent
data matrix while maximizing X and Y covariance. Among other parameters, PLS-R results in the linear
regression equation Eq. (1) including regression coefficients for each of X variables predicting given Y
variable.
ܻൌܾܾଵܺଵܾଶܺଶڮܾܺ
(1)
Regression coefficients and also X loading weights (parameter obtained from PLS decomposition of
original matrices) have a crucial importance for understanding which X variables are important in
describing Y variables (responses) [10].
Moreover, the developed models were validated applying leave-one-out full cross-validation, which is
appropriate for relatively small datasets.
Each one of the individual PLS-R models developed (i.e. one for each variable) was evaluated using the
coefficient of determination R2 from the parity plot calculated according to Eq. (2) for each predicted
variable Y.
ܴଶൌͳെσሺ
ݕെݕොሻଶ
ୀଵ
σሺ
ݕെݕതሻଶ
ୀଵ
(2)
Furthermore, the root means square error (RMSE) was calculated for each PLS-R model according to
Eq. (3). This parameter expresses the difference between values predicted by the model and reference
values.
ܴܯܵܧൌඨσሺ
ݕොെݕሻଶ
ୀଵ݊െͳ
(3)
The coefficient of determination R2 and RMSE were calculated from the parity plot. For the calibration
it is noted as R2cal and SEC and for the cross-validation as R2cv and RMSECV.
3. Results and discussion
3.1. Visual inspection and spectral pre-processing
Original raw spectra received from FT-NIR instrument are shown in Fig. 1a. The spectroscopy
measurements were performed in the diffuse-reflectance regime but the results are received as relative
absorbance. Raw spectra obtained in this manner typically have a baseline shift; it appears due to different
particle sizes of a solid material influencing spectral path length. Application of spectral pretreatment is
therefore required. Scatter level was corrected using MSC as shown in Fig. 1b.
Jan Skvaril et al. / Energy Procedia 105 ( 2017 ) 1309 – 1317 1313
Fig. 1. (a) Raw spectra (9000 – 4000 cm-1) ; (b) Spectra pre-processed by MSC
3.2. Multivariate calibration
Multivariate calibrations were developed using PLS1 algorithm [14]. Individual PLS-R models
including the full NIR spectral region (12,000–4000 cm-1) as X variables were developed for each Y variable
(response). The results from the reference methods used for determining moisture content, ash content and
HHV are presented in Table 1.
Table 1. Descriptive statistics of reference data set
Parameter
݊
ݔҧ
ݏ
ܯ݅݊
ܯܽݔ
Moisture content (wt%)
83
45,21
6,49
31,91
61,53
Ash content (wt%)
37
2,00
0,79
0,33
3,27
Higher heating value (MJ/kg)
20
20,45
0,56
19,66
21,72
Multivariate calibration results are reported in Table 2 (figures of merit). The resulting parity plots, i.e.
values of predicted versus reference results, are illustrated in Fig. 3. The best performing model was for
prediction of moisture content and was based on 5 PLS factors. The ability of the NIRS technique to
effectively predict moisture content builds on the fact that there are strong interactions between
electromagnetic radiation and molecules of water that are detectable in the NIR spectral region. The model
predicting ash content was based on 5 PLS factors. However, the prediction performance of the model is
lower compared to that for moisture. The model predicting higher heating value is based on 7 PLS factors,
therefore it is considered as rather more complex. Generally, the heating value is strongly affected by
moisture content of the fuel, however, in our case, the calibration was based on reference values determined
on moisture free samples. Therefore, the varying heating value over the range of samples was related to the
actual composition of the combustible matter.
a)
b)
1314 Jan Skvaril et al. / Energy Procedia 105 ( 2017 ) 1309 – 1317
Table 2. Prediction performance of PLS-R models
Predicted parameter
Rank
(Number of
PLS-R factors)
Spectral
pre-processing
ܴ
ଶ
ܴܯܵܧܥ
ܴ
ଶ
ܴܯܵܧܥܸ
Moisture content (wt%)
5
MSC
0,90
1,96
0,88
2,18
Ash content (wt%)
5
MSC
0,80
0,39
0,69
0,44
Higher heating value (MJ/kg)
7
MSC
0,92
0,14
0,85
0,22
All developed models show reasonable accuracy and demonstrate the potential for implementing online
for improved process control. Before the actual industrial application, the PLS-R models can be
complemented by a larger number of samples and validated by means of external validation using an
independent dataset. Future improvements in prediction performance could also be achieved by employing
non-linear regression methods or testing other pre-processing techniques and their combinations. The
robustness of the model could be enhanced by selecting different reduced spectral ranges or several
individual X variables as multivariate calibration inputs [15], rather than considering the full NIR spectrum.
Fig. 2. Parity plots; (a) model predicting moisture content; (b) model predicting ash content; (c) model predicting higher heating value
3.3. Important model predictors
The regression coefficient plot and X-loading weights plot for PLS-R model predicting moisture content is
illustrated in Fig. 3. X variables (predictors) for which the regression coefficients and X-loading weights
reach critical points (i.e. peaks or troughs) are important in describing Y variables in PLSR-R model.
Typically, the presence of a critical point indicates that the vibration of a specific chemical bond occurring
at a corresponding wavenumber (i.e. frequency) has a significant influence on the prediction of the Y
variable (i.e. given fuel property) [10]. The actual position of the critical point in both plots can be however
biased to a certain extent due to the use of a particular spectral pre-processing and calibration technique
[16]. It can make the assignment of vibrational bands in some cases particularly challenging. The critical
points for the regression coefficients for the model predicting moisture content are at: 5280, 5238, 5084,
4875 and 4170 cm-1. Furthermore, critical points for the X- loading weights are at: 5265 and 4937 cm-1 for
PLS factor 1; 5250, 5034, 4729 and 4390 cm-1 for PLS factor 2 and 7055, 5265, 5118 and 4162 cm-1 for
PLS factor 3. The critical point at approximately 5238 cm-1 can be assigned to the bond vibration for water
(combination of O-H antisymmetric stretching and O-H deformation vibration) as well as the critical point
at approx. 5034 cm-1 (combination of O-H stretching and O-H deformation vibration).
30
35
40
45
50
55
60
30 35 40 45 50 55 60
Predicted (Moisture wt%)
Reference (Moisture wt%)
Calibration
Cross-Validation
±5 wt%
0
0,5
1
1,5
2
2,5
3
3,5
0 0,5 1 1,5 2 2,5 3 3,5
Predicted (Ash wt%)
Reference (Ash wt%)
Calibration
Cross-Validation
±2,5 wt%
19
19,5
20
20,5
21
21,5
22
19 19,5 20 20,5 21 21,5 22
Predicted (HHV MJ/kg)
Reference (HHV MJ/kg)
Calibration
Cross-Validation
±2,5%
(@ 20 MJ/kg)
a)
b)
c)
Jan Skvaril et al. / Energy Procedia 105 ( 2017 ) 1309 – 1317 1315
Fig. 3. Moisture content predictors; (a) regression coefficients plot; (b) X loading weights plot
The regression coefficient plot and X-loading weight plot for the PLS-R model predicting ash content is
illustrated in Fig. 4. The critical points for the regression coefficients for the model predicting ash content
are at 5230, 5053, 4640, 4204 and 4158 cm-1. Furthermore, the critical points for the X- loading weights
are at: 7070, 5265, 4952, 4736 and 4409 cm-1 for PLS factor 2; 5250, 5049 and 4200 cm-1 for PLS factor 3.
The critical point at approx. 4204 cm-1 can be assigned to the bond vibration of calcium carbonate CaCO3
(2nd overtone of asymmetric stretching vibration) which is often present in biomass ash. For further band
assignment, the knowledge of the ash chemical composition is necessary, which was not part of the present
study.
Fig. 4. Ash content predictors; (a) regression coefficients plot; (b) X loading weights plot
The regression coefficient plot and X-loading weight plot for the PLS-R model predicting higher heating
value is illustrated in Fig. 5. The critical points of the regression coefficients for the model predicting HHV
are at 5257, 5049, 4613, 4613 and 4192 cm-1. Moreover, the critical points of the X- loading weights are
at: 7055, 5257, 5018, 4736 and 4262 cm-1 for PLS factor 2; 6889, 5273, 4995, 4660 cm-1 for PLS factor 3.
Fig. 5. Higher heating value predictors; (a) regression coefficients plot; (b) X loading weights plot
The critical point at approximately 7055 cm-1 can be assigned to the bond vibration of lignin
(combination of 1st overtone C-H stretching and C-H bending vibration). The same holds for the critical
-0,12
-0,08
-0,04
0
0,04
0,08
0,12
40005000600070008000
Regression coefficient
Wavenumber (cm
-1
)
5280
4875
5238
5084
4170
-0,12
-0,08
-0,04
0
0,04
0,08
0,12
40005000600070008000
X-loading weights
Wavenumber (cm
-1
)
Factor 1
Factor 2
Factor 3
4390
4729
5118
4162
4937
5034
5265
7055
-0,04
-0,02
0
0,02
0,04
40005000600070008000
Regression coefficient
Wavenumber (cm
-1
)
4640
5053
5230
4204
4158
-0,12
-0,08
-0,04
0
0,04
0,08
0,12
40005000600070008000
X-loading weights
Wavenumber (cm
-1
)
Factor 1
Factor 2
Factor 3
7070
5265
4952
4200
5250
5049
4736
4409
-3
-2
-1
0
1
2
3
40005000600070008000
Regression coefficient
Wavenumber (cm
-1
)
4192
4355
4613
5257
5049
-0,12
-0,08
-0,04
0
0,04
0,08
0,12
40005000600070008000
X-loading weights
Wavenumber (cm
-1
)
Factor 1
Factor 2
Factor 3
6889
7055
5257
5018
4660
5273
4995
4736
4262
a)
b)
a)
b)
a)
b)
1316 Jan Skvaril et al. / Energy Procedia 105 ( 2017 ) 1309 – 1317
point at approximately 6889 cm-1 (1st overtone O-H stretching vibration). The critical point at approximately
5257 cm-1 can be assigned to the bond vibration of hemicellulose (2nd overtone C=O stretching vibration).
The critical point at approximately 4734 cm-1 can be assigned to the bond vibration of cellulose
(combination of O-H deformation and O-H stretching vibration). The same holds for the critical point at
approximately 4262 cm-1 (combination of C-H stretching and C-H deformation vibration and/or 2nd
overtone C-H deformation vibration). Moreover, the critical point at approximately 4660 cm-1 can be
assigned to the bond vibration of lignin and extractives (combination of Carom-H stretching and C=C
stretching vibration). As can be observed from the results, the prediction capability of the model for higher
heating value is related to the actual composition of the combustible matter (i.e. content of lignin, cellulose,
hemicellulose, etc.).
4. Conclusions
This work presents experimental measurements and multivariate spectroscopic calibration of near
infrared (NIR) spectra and different fuel properties determined by standardized procedures.
According to the results presented, the following conclusions can be drawn:
x Multiplicative scatter correction (MSC) presents satisfactory results when used as a spectral pre-
processing technique correcting light scattering and change in a path length for each sample.
x The best figures of merit were achieved for the model predicting moisture content (i.e. R2CV=0,88;
RMSECV=2,18). This is due to the strong interaction of electromagnetic radiation with water molecules
resulting in the formation of strong absorption bands in NIR spectra.
x The figures of merit for the model predicting ash content (i.e. R2CV=0,69; RMSECV=0,44) show
somewhat lower numbers than those for moisture. Nevertheless, such a model is still satisfactory for use
in a process control (soft sensor).
x The model for predicting the higher heating value for moisture free samples, exhibits relatively good
figures of merit (i.e. R2CV=0,85; RMSECV=0,22); the prediction ability is related to the actual
composition of a combustible matter (i.e. content of lignin, cellulose, hemicellulose, etc.).
In summary, it has been succesfully demonstrated that different solid biofuel properties can be determined
by means of near infrared spectroscopy. This technique enables fast and non-destructive measurements
which may be implemented for improved control of thermochemical energy conversion processes leading
towards increased efficiency and reduction of emissions.
Acknowledgements
The authors gratefully acknowledge financial support from the Swedish Knowledge Foundation
(Stiftelsen för kunskaps- och kompetensutveckling, KK-stiftelsen). In addition, the authors would like to
acknowledge assistance from Nemir Gabriel and Emad Al Hamrami for performing part of NIR scans
and reference laboratory measurements. This project has received funding from the European Union’s
Horizon 2020 research and innovation programme under grant agreement No 723523”.
References
[1] Bakeev KA. Process analytical technology: spectroscopic tools and implementation strategies for the chemical and
pharmaceutical industries. John Wiley & Sons; 2010.
[2] Huang H, Yu H, Xu H, Ying Y. Near infrared spectroscopy for on/in-line monitoring of quality in foods and beverages: A
review. Journal of Food Engineering 2008;87:303–13. doi:10.1016/j.jfoodeng.2007.12.022.
[3] Tsuchikawa S, Kobori H. A review of recent application of near infrared spectroscopy to wood science and technology.
Journal of Wood Science 2015:213–20. doi:10.1007/s10086-015-1467-x.
[4] Chadwick DT, McDonnell KP, Brennan LP, Fagan CC, Everard CD. Evaluation of infrared techniques for the assessment
of biomass and biofuel quality parameters and conversion technology processes: A review. Renewable and Sustainable
Energy Reviews 2014;30:672–81. doi:10.1016/j.rser.2013.11.006.
Jan Skvaril et al. / Energy Procedia 105 ( 2017 ) 1309 – 1317 1317
[5] Lestander TA, Rhén C. Multivariate NIR spectroscopy models for moisture, ash and calorific content in biofuels using bi-
orthogonal partial least squares regression. Analyst 2005;130:1182–9.
[6] Yang H, Sheng K. Characterization of Biochar Properties Affected by Different Pyrolysis Temperatures Using Visible-
Near-Infrared Spectroscopy. ISRN Spectroscopy 2012;2012:1–7. doi:10.5402/2012/712837.
[7] Eriksson D, Weiland F, Hedman H, Stenberg M, Öhrman O, Lestander T a., et al. Characterization of Scots pine stump-
root biomass as feed-stock for gasification. Bioresource Technology 2012;104:729–36.
doi:10.1016/j.biortech.2011.10.102.
[8] Preece SLM, Auvermann BW, MacDonald JC, Morgan CLS. Predicting the heating value of solid manure with visible
and near-infrared spectroscopy. Fuel 2013;106:712–7. doi:10.1016/j.fuel.2012.10.006.
[9] Posom J, Sirisomboon P. Evaluation of the moisture content of Jatropha curcas kernels and the heating value of the oil-
extracted residue using near-infrared spectroscopy. Biosystems Engineering 2015;130:52–9.
doi:10.1016/j.biosystemseng.2014.12.003.
[10] Posom J, Shrestha A, Saechua W, Sirisomboon P. Rapid non-destructive evaluation of moisture content and higher
heating value of Leucaena leucocephala pellets using near infrared spectroscopy. Energy 2016;107:464–72.
doi:10.1016/j.energy.2016.04.041.
[11] Fearn T, Riccioli C, Garrido-Varo A, Guerrero-Ginel JE. On the geometry of SNV and MSC. Chemometrics and
Intelligent Laboratory Systems 2009;96:22–6. doi:10.1016/j.chemolab.2008.11.006.
[12] Wold S, Sjöström M, Eriksson L. PLS-regression: A basic tool of chemometrics. Chemometrics and Intelligent
Laboratory Systems 2001;58:109–30. doi:10.1016/S0169-7439(01)00155-1.
[13] Geladi P, Kowalski BR. Partial least-squares regression: a tutorial. Analytica Chimica Acta 1986;185:1–17.
doi:10.1016/0003-2670(86)80028-9.
[14] Fichet B, Piccolo D, Verde R, Vichi M. Classification and Multivariate Analysis for Complex Data Structures. 2011.
doi:10.1007/978-3-642-13312-1.
[15] Mehmood T, Liland KH, Snipen L, Sæbø S. A review of variable selection methods in Partial Least Squares Regression.
Chemometrics and Intelligent Laboratory Systems 2012;118:62–9. doi:10.1016/j.chemolab.2012.07.010.
[16] Via BK, Zhou C, Acquah G, Jiang W, Eckhardt L. Near infrared spectroscopy calibration for wood chemistry: Which
chemometric technique is best for prediction and interpretation? Sensors (Switzerland) 2014;14:13532–47.
doi:10.3390/s140813532.
Biography
Jan Skvaril, is a doctoral candidate and lecturer at the School of Business, Society and
Engineering at Mälardalen University in Sweden. He received his M.Sc. (Ing.) degree in
Power Engineering in 2008 and M.Sc. (Ing.) in Company Management and Economics in
2009. His research is focused on combustion in biomass-fired industrial steam boilers and
online determination of fuel properties by near infrared spectroscopy (NIRS).
