- Email: [email protected]
Ultimate analysis and heating value prediction of straw by near infrared spectroscopy
Waste Management 29 (2009) 1793–1797
Contents lists available at ScienceDirect
Waste Management journal homepage: www.elsevier.com/locate/wasman
Ultimate analysis and heating value prediction of straw by near infrared spectroscopy C. Huang, L. Han *, Z. Yang, X. Liu College of Engineering, China Agricultural University, P.O. Box 232, No. 17, Qinghua Donglu, Haidian District, Beijing 100083, China
a r t i c l e
i n f o
Article history: Accepted 27 November 2008 Available online 12 January 2009
a b s t r a c t Ultimate analysis and heating value determination are two of the most important routine analyses for exploiting agricultural wastes for energy conversion. The use of near infrared spectroscopy (NIRS) was investigated as an alternative method to predict the carbon, hydrogen, and nitrogen content and the heating value of straw. A total of 222 straw samples, collected from 24 provinces of China, were used for NIRS calibration and validation in this study. The R2v and standard error of predictions in independent validation were, respectively, 0.97 and 0.37% for C, 0.77 and 0.17% for H, 0.87 and 0.10% for N and 0.96 and 181 J/g for heating value. A multiple linear regression (MLR) model was also built to predict the heating value from the contents of C, H and N. The MLR equation gave good prediction (standard error of prediction = 224 J/g) when evaluated using the same validation set as the NIRS. Therefore, rapid analysis of straw can be achieved through the constructed equations, saving analytical time and cost. Ó 2008 Elsevier Ltd. All rights reserved.
1. Introduction Great achievements have been made in using agricultural waste for energy purposes during the last few decades. Initially driven by the intention to decrease the aggravation of the greenhouse effect caused by combustion of fossil fuels, straw resource utilization brings more benefits to social, economic and environmental sustainability. More than 600 million tons of straw, equivalent to about 300 million tons of standard coal, were produced annually from China’s agricultural production (Han et al., 2002). Efficient energy conversion and utilization from straw resources will play an important role in maintaining China’s energy security (Olz et al., 2007). Although, approaches towards energy conversion are diversified, the characterization of biomass from the point of view of energy utilization is usually conducted using proximate analysis, ultimate analysis and heating value. The proximate analysis gives moisture, volatile matter, fixed carbon and ash contents, and ultimate analysis gives the composition of the biomass in wt.% of carbon, hydrogen and oxygen, as well as sulfur and nitrogen. The heating value, defined as the amount of heat released from the biomass combustion, is one of the most important properties for design calculations of conversion systems for biomass. Biomass samples vary in composition, leading to different performances in combustion and other conversion processes (Allica * Corresponding author. Tel.: +86 1062736480; fax: +86 1062736778. E-mail addresses: [email protected] (C. Huang), [email protected] (L. Han). 0956-053X/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.wasman.2008.11.027
et al., 2001). For engineering estimations, ultimate data and heating value are critical for proper design and operation of conversion facilities (Jenkins et al., 1998). Conventional laboratory measurements of biomass composition are usually tedious and time consuming; consequently, technologies that provide prompt quantitative analysis become quite necessary. Near infrared spectroscopy (NIRS) is a spectroscopic method utilizing the near infrared region of the electromagnetic spectrum (from about 800 to 2500 nm). Intensities of the absorption at each wavelength in the near infrared region are measured for presented samples. Chemometric methods are then applied to correlate spectral data and investigated properties. NIRS has been extensively used for quantitative and qualitative analysis in agricultural, food and pharmaceutical industries, and has proved to be a strong technology that determines the constituents of samples easily, rapidly and nondestructively. Ultimate analysis of coal by NIRS, in which predictive models for C, H and N were constructed, has been recently tested (Kaihara et al., 2002; Andres and Bona, 2005). NIRS predictions of carbon and nitrogen have also been reported in the field of soil and compost science, as well as several cases involving plant material (Gilson et al., 1999). The relationship between ultimate data and heating value of biomass resources has been frequently described by predictive models (Demirbas, 1997; Sheng and Azevedo, 2005). Since predictive models already exist for other biomass resources, the present study focused on model development for straw samples only. NIRS was firstly applied for prediction of C, H, N and heating value for a rapid characterization of straw. Then, the relationship between heating value and percentage of C, H, N
1794
C. Huang et al. / Waste Management 29 (2009) 1793–1797
ascending in reference data into validation set, while the remaining samples comprised the calibration set.
Table 1 Statistical results of C, H, N and heating value of rice straw and wheat straw. Rice straw (n = 172)
C (%) H (%) N (%) CV (J/g)
Wheat straw (n = 50)
Range
Mean
SD
Range
Mean
SD
35.20–40.95 4.69–6.19 0.56–1.46 14,191–16,886
38.45 5.28 0.88 15,685
1.14 0.30 0.17 502
40.14–44.97 5.53–6.05 0.13–0.62 16,175–18,041
42.89 5.82 0.36 17281
1.03 0.11 0.13 406
Table 2 Mean, standard deviation (SD) and range for carbon, hydrogen, nitrogen and heating value in straw samples. Calibration
C (%) H (%) N (%) Heating value (J/ g)
Validation
2.2. Spectroscopic analysis Spectra were obtained on a Foss NIRSystem 6500 spectrometer from 400 to 2498 nm with 2-nm increments. Samples were presented in a rectangular quarter cup and scanned at 20–22 °C controlled by an air conditioner. In order to reduce random variability caused by sample loading, samples were repacked and scanned three times and the absorbance of each repack was measured. The average of each set of repeated reading was calculated before the calibration step. 2.3. Reference analysis
Range
Mean
SD
Range
Mean
SD
35.20–44.69 4.69–6.19 0.13–1.46 14,191– 18,041
39.38 5.40 0.77 16,019
2.11 0.35 0.27 810
36.42–44.63 4.82–6.05 0.17–1.40 14,598– 17,953
39.45 5.41 0.78 16,045
2.11 0.35 0.27 808
was determined, and a MLR model predicting heating value from C, H and N was built. The two heating value models were compared to evaluate their accuracy. 2. Materials and methods 2.1. Samples preparation A total of 222 straw samples, collected from 24 provinces of China, were used in the development of NIRS models for prediction of C, H, N and heating value. The samples consisted of rice straw and wheat straw and varied considerably in harvest time, varieties of grain, places of origin, climate where grown, soil types and storage conditions. After being oven dried at 70 °C for 24 h, the samples were ground in a ZM100 mill (Rersch GmbH & Company, Germany) through 1.0-mm sieve, and kept in airtight ziplock bags before spectra collection and reference analysis. The entire set of samples was split into calibration and validation sets, by taking every fourth sample from the samples set
The carbon, hydrogen and nitrogen concentrations in the straw samples were measured using an Elementar Vario EL III analyzer (Elementar Analysensysteme GmbH, Germany) according to CEN standard methods (CEN/TS 15104). Heating value was determined by CEN method (CEN/TS 14918) using an IKA C200 oxygen bomb calorimeter (IKA Analysentechnik GmbH, Heitersheim, Germany). C, H and N were reported in weight percentage on a dry basis, and heating value in J/g on dry basis. 2.4. Calibration and validation The WinISI II software (Infrasoft International LLC, USA) was used for data processing. Calibration equations were developed by performing principal component regression (PCR), partial least square regression (PLS) and modified partial least square regression (MPLS) on spectra data and reference data. The maximum number of principal components (or PLS factors) were set at 16, based on the rule of 1 per 10 samples of the calibration set. Cross validation was employed to avoid over-fitting. Scattering correction was made by standard normal variate (SNV), de-trend (D), standard multiplicative scatter correction (SMSC) and weighted multiplicative scatter correction. Additionally, two derivative treatments of the spectra were compared, (1,4,4,1) and (2,4,4,1), where the first number indicates the order of the derivative, the second number is the gap in data points over which the derivative is calculated, and the third and the fourth number refer to the
Fig. 1. NIR Spectra of straw samples.
1795
C. Huang et al. / Waste Management 29 (2009) 1793–1797 Table 3 Calibration and validation statistics for selected equations of C, H, N and heating value. Calibration
C (%) H (%) N (%) Heating value (J/g)
Validation
R2
SEC
R2cv
SECV
R2v
SEP
RPD
0.98 0.85 0.93 0.97
0.30 0.13 0.07 152
0.97 0.78 0.90 0.96
0.34 0.16 0.08 173
0.97 0.77 0.87 0.96
0.37 0.17 0.10 181
5.64 2.07 2.76 4.47
vel of 0.05. The inclusion of each variable in the proposed model was based on the t-criterion (Nikolaou et al., 2004). The goodness of the model fit was evaluated through examination of various statistical parameters, including regression coefficient (R2), the F statistic, and the a value for linear regression (Uyak et al., 2007). All the procedures, including equation construction and statistics, were executed using Excel 2003 software. 3. Results and discussion 3.1. Reference data
number of the data points used in first and second smoothing, respectively. Statistic parameters as standard error of cross validation (SECV) and the determination coefficient of calibration (R2) were employed to select optimum calibration. The validation set was then introduced to assess the calibration. Besides a high determination coefficient of validation (R2v ) and low standard error of prediction (SEP), relative prediction deviation (RPD, the ratio of the standard deviation of the reference data (SD) to the SEP) was also calculated to evaluate the performance of calibration (Malley et al., 2005). 2.5. MLR equation development Multiple linear regression was adopted to build the equation to predict heating value of straw samples from C, H and N percentages. For the purpose of comparison, the calibration set used in the NIRS equation of heating value was applied to train the MLR model, and the validation set of the NIRS equation was used to examine the prediction performance of the MLR model. Significance evaluations of variables were conducted at a significance le-
Compositional difference between rice straw and wheat straw was expressed with range, mean and standard deviation of C, H, N and heating value in Table 1. It is proved by ANOVA analysis that the N of rice straw was significantly higher than that of wheat straw (p < 0.01); and the C, H and heating value of rice straw were significantly lower than those of wheat straw (p < 0.01). The mean, range and standard deviation of reference data are shown in Table 2. There was a large standard deviation in carbon content between samples while H, N, and heating value were less variable. The means of the four properties are considered to lie within normal ranges and agreed with literature values (Jenkins et al., 1998; Demirbas, 2004; Liao et al., 2004). 3.2. Sample spectra Near infrared spectra of 222 straw samples were illustrated in Fig. 1. The whole spectra were divided into two parts, divided by the discontinuities at 1100 nm where the sensor changed. The spectra range of 400–1100 nm consists of visible light region and
Heating value (J/g)
18000
17000
16000
15000
14000 35
37
39
41
43
45
17000
16000
15000
14000 4. 50
5. 00
C (%)
5. 50
H (%)
18000
Heating value (J/g)
Heating value (J/g)
18000
17000
16000
15000
14000 0. 00
0. 50
1. 00
1. 50
N (%) Fig. 2. Heating value is plotted against the contents of carbon, hydrogen and nitrogen.
6. 00
6. 50
1796
C. Huang et al. / Waste Management 29 (2009) 1793–1797
Table 4 Statistical test result of MLR models for heating value prediction. Regression
Statistics
R R2 Adjusted R2 Standard error of the estimate Observations
0.9629 0.9271 0.9257 221 160 df
Sum of square
Mean square
F
Significance
3 156 159
96,659,781 7,597,705 104,257,485
32,219,927 48,703
662
1.84E88
short wave near infrared region, while 1100–2498 nm belong to the long wave near infrared region. Spectra of 400–1100 nm region were more scattered than those of 1100–2498 nm, and contained more irregular interlacement, decreasing the freedom of spectra interpretability. By contrast, the spectra in 1100–2498 nm region were gathered and comprised more informative peaks. Based on description above, the 1100–2498 nm region was selected for the construction of the NIRS models. The spectra showed absorption bands at 1194 nm related with C– H stretch second overtone, at 1724 nm, 1766 nm related with C–H first overtone, at 1930 nm related with O–H stretch/HOH deformation combination, at 2100 nm related with O–H bond/C–O stretch combination, at 2280 nm related with C–H stretch/CH2 deformation, and at 2236 related with C–H stretch/C–H deformation. 3.3. NIRS equations 3.3.1. Influence of spectra preprocessing and regression A total of 54 of regression equations were obtained from the six scatter correction options, the three derivative treatments and the three regression methods. Overall comparisons were made to examine whether preprocessing and regression methods influence the performance of predictive equations. SECV ranged from 0.33% to 0.50% for C equations, 0.16% to 0.22% for H equations, 0.08% to 0.11% for N equations, and 173 to 202 J/g for heating value equations. The performance of equations for each property was similar, independent of mathematical preprocessing and regression methods. 3.3.2. Equation accuracy Table 3 shows the statistical results obtained in the calibration and validation. It can be seen that R2cv and R2v of C, N, and heating value equations were higher than 0.90 and 0.85, respectively. Very small prediction errors (SEP) were observed compared with the mean values of corresponding components. RPD, a commonly used index for evaluating the performance of NIRS equations, were all higher than 2.0, supporting the applicability of the equations. Particularly, the RPD of C and heating value equations were 5.64 and 4.47, respectively, indicating excellent NIRS equations. 3.4. MLR equation for heating value Straw is primarily composed of elements C, H, O, N, S and Cl, of which C and H make the most positive contribution to heating value, and O makes a negative contribution. The correlation between heating value and C, H, N was calculated in this study. The heating value was plotted against the contents of C, H and N contents (Fig. 2). The heating value increased with increasing of C and H contents, but decreased with an increase of N content. Pearson correlation (Wilcox, 2003) between heating value and carbon (r = 0.94), hydrogen (r = 0.65), nitrogen (r = 0.59) were all significant at a 0.01 level.
Multiple regression analysis was applied to the data in order to find equations that would describe the relationship between heating values and C, H, N contents of straw samples. Prediction models obtained consisted of variables which were statistically significant (p < 0.05), and had the form:
Heating value ¼ 356:08 þ ð432:80 CÞ ð297:73 HÞ þ ð287:45 NÞ Although C, H and N were combustible elements, the Pearson correlation and the regression coefficients in the equation were not all positive. This was caused by the correlation between independent variables such as those between C and H (r = 0.72), as well as those between C and N (r = 0.65). Consequently, the equation can not explain how much each element contributes to the heating value, but as a whole performs well in predicting heating value. The statistical results of the MLR model are presented in Table 4. The correlation coefficient (R) indicates the extent of linear correlation between heating value and C, H and N; R2 explains the percentage of variation in heating value caused by C, H and N; and R2adj eliminated the influence of sample numbers on the R2. All of them were above 0.90, which showed the high correlation between heating value and C, H, N contents. The F value (F = 662) of the equation was extremely large, indicating that the MLR equation was significant. One of the assumptions of analysis of variance (ANOVA) is that variances of the observations in the individual groups are equal, a situations referred to as homogeneity of variance. Ideally, residuals should be nearly constant and independent of dependent variables. A scatter plot of standard residuals against predicted heating value was showed in Fig. 3. The standard residues were randomly distributed, and most of the data points lay within ±2. That suggested that the ANOVA analysis conducted in the previous section was reliable. 3.5. Independent validation for heating value equations Although the procedures and costs of the development of the equations differed, both NIRS and MLR equations predict heating value accurately when evaluated on independent validation. As shown in Fig. 4 the determination coefficients of NIRS and MLR equations were 0.9562 and 0.9273, respectively. The accuracy of
Standard Residue
ANOVA Regression Residual Total
4
0 14000
16000
18000
-4
Heating value (J/g) Fig. 3. Residue plot for MLR heating value equation.
1797
C. Huang et al. / Waste Management 29 (2009) 1793–1797
18000
18000
2
17000
MLR value (J/g)
NIRS value (J/g)
2
R v =0.9562 SEP=181 J/g
16000
15000
14000 14000
15000
16000
17000
18000
Lab value (J/g)
17000
R v =0.9273 SEP=224 J/g
16000
15000
14000 14000
15000
16000
17000
18000
lab value (J/g)
Fig. 4. Independent validation of NIRS equation and MLR equation for predicting heating value.
the NIRS equation (standard error of prediction = 181 J/g) was a little better than that of the MLR equation (standard error of prediction = 224 J/g). NIRS prediction presented greater ease of use and maintainability than prediction of the MLR equation for potential online analysis. However, robustness of the NIRS was greatly influenced by environmental conditions such as temperature and air humidity. 4. Conclusion NIRS prediction of C, H, N contents and heating value of straw samples presented satisfactory accuracy, which is believed to be reliable for quantitative analysis. Particularly, the NIRS equations for C and heating value predictions, judging from the RPD value, are successful for quantitative analysis and therefore eligible for engineering application. The MLR equation for heating value prediction also presents good performance. By the constructed equations, rapid analysis of straw can be achieved in an easier and less costly approach, and satisfactory accuracy of the models allows further possibility to monitor biological or thermochemical conversion processes. References Allica, J.H., Mitre, A.J., Bustamante, J.G., Itoiz, G., Blanco, F., Alkorta, I., Garbisu, C., 2001. Straw quality for its combustion in a straw-fired power plant. Biomass and Bioenergy 21, 249–258. Andres, J.M., Bona, M.T., 2005. Analysis of coal by diffuse reflectance near infrared spectroscopy. Analytica Chimica Acta 35, 123–132.
CEN/TS 15104:2005. Solid Biofuels – Determination of total content of carbon, hydrogen and nitrogen – Instrumental methods. CEN/TS 14918:2005. Solid Biofuels – Method for the determination of calorific value. Demirbas, A., 1997. Calculation of higher heating values of biomass fuels. Fuel 76 (5), 431–434. Demirbas, A., 2004. Combustion characteristics of different biomass fuels. Progress in Energy and Combustion Science 30, 219–230. Gilson, D., Houssard, C., Joffre, R., 1999. Using near infrared reflectance spectroscopy to predict carbon, nitrogen and phosphorus content in heterogeneous plant material. Oecologia 118, 173–182. Han, L., Yan, Q., Liu, X., Hu, J., 2002. Straw resources and their utilization in China. Transaction of the CSAE 18 (3), 87–91. Jenkins, B.M., Baxter, L.L., Miles Jr., T.R., Miles, T.R., 1998. Combustion properties of biomass. Fuel Processing Technology 54, 17–46. Kaihara, M., Takahashi, T., Akazawa, T., Sato, T., Takahashi, S., 2002. Application of near infrared spectroscopy to rapid analysis of coal. Spectroscopy Letters 35 (3), 369–376. Liao, C., Wu, C., Yan, Y., Huang, H., 2004. Chemical elemental characteristics of biomass fuels in China. Biomass and Bioenergy 27, 119–130. Malley, D.F., McClure, C., Martin, P.D., Buckley, K., McCaughey, W.P., 2005. Compositional analysis of cattle manure during composting using a fieldportable near-infrared spectrometer. Communications in Soil Science and Plant Analysis 36, 455–475. Nikolaou, A.D., Golfinopoulos, S.K., Arhonditsis, G.B., Kolovoyiannis, V., Lekkas, T.D., 2004. Modeling the formation of chlorination by-products in river waters with different quality. Chemosphere 55, 409–420. Olz, S., Sims, R. and Kirchner, N., 2007. Contribution of Renewables to Energy Security. IEA Information Paper 2007. Sheng, C., Azevedo, J.L.T., 2005. Estimating the higher heating value of biomass fuels from basic analysis data. Biomass and Bioenergy 28, 499–507. Uyak, V., Ozdemir, K., Toroz, I., 2007. Multiple linear regression modeling of disinfection by-products formation in Istanbul drinking water reservoirs. Science of the Total Environment. doi:10.1016/j.scitotenv.2007.02.041. Wilcox, R.R., 2003. Modern insights about Pearson’s correlation and least squares regression. International Journal of Selection and Assessment 9 (2), 195–205.
Contents lists available at ScienceDirect
Waste Management journal homepage: www.elsevier.com/locate/wasman
Ultimate analysis and heating value prediction of straw by near infrared spectroscopy C. Huang, L. Han *, Z. Yang, X. Liu College of Engineering, China Agricultural University, P.O. Box 232, No. 17, Qinghua Donglu, Haidian District, Beijing 100083, China
a r t i c l e
i n f o
Article history: Accepted 27 November 2008 Available online 12 January 2009
a b s t r a c t Ultimate analysis and heating value determination are two of the most important routine analyses for exploiting agricultural wastes for energy conversion. The use of near infrared spectroscopy (NIRS) was investigated as an alternative method to predict the carbon, hydrogen, and nitrogen content and the heating value of straw. A total of 222 straw samples, collected from 24 provinces of China, were used for NIRS calibration and validation in this study. The R2v and standard error of predictions in independent validation were, respectively, 0.97 and 0.37% for C, 0.77 and 0.17% for H, 0.87 and 0.10% for N and 0.96 and 181 J/g for heating value. A multiple linear regression (MLR) model was also built to predict the heating value from the contents of C, H and N. The MLR equation gave good prediction (standard error of prediction = 224 J/g) when evaluated using the same validation set as the NIRS. Therefore, rapid analysis of straw can be achieved through the constructed equations, saving analytical time and cost. Ó 2008 Elsevier Ltd. All rights reserved.
1. Introduction Great achievements have been made in using agricultural waste for energy purposes during the last few decades. Initially driven by the intention to decrease the aggravation of the greenhouse effect caused by combustion of fossil fuels, straw resource utilization brings more benefits to social, economic and environmental sustainability. More than 600 million tons of straw, equivalent to about 300 million tons of standard coal, were produced annually from China’s agricultural production (Han et al., 2002). Efficient energy conversion and utilization from straw resources will play an important role in maintaining China’s energy security (Olz et al., 2007). Although, approaches towards energy conversion are diversified, the characterization of biomass from the point of view of energy utilization is usually conducted using proximate analysis, ultimate analysis and heating value. The proximate analysis gives moisture, volatile matter, fixed carbon and ash contents, and ultimate analysis gives the composition of the biomass in wt.% of carbon, hydrogen and oxygen, as well as sulfur and nitrogen. The heating value, defined as the amount of heat released from the biomass combustion, is one of the most important properties for design calculations of conversion systems for biomass. Biomass samples vary in composition, leading to different performances in combustion and other conversion processes (Allica * Corresponding author. Tel.: +86 1062736480; fax: +86 1062736778. E-mail addresses: [email protected] (C. Huang), [email protected] (L. Han). 0956-053X/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.wasman.2008.11.027
et al., 2001). For engineering estimations, ultimate data and heating value are critical for proper design and operation of conversion facilities (Jenkins et al., 1998). Conventional laboratory measurements of biomass composition are usually tedious and time consuming; consequently, technologies that provide prompt quantitative analysis become quite necessary. Near infrared spectroscopy (NIRS) is a spectroscopic method utilizing the near infrared region of the electromagnetic spectrum (from about 800 to 2500 nm). Intensities of the absorption at each wavelength in the near infrared region are measured for presented samples. Chemometric methods are then applied to correlate spectral data and investigated properties. NIRS has been extensively used for quantitative and qualitative analysis in agricultural, food and pharmaceutical industries, and has proved to be a strong technology that determines the constituents of samples easily, rapidly and nondestructively. Ultimate analysis of coal by NIRS, in which predictive models for C, H and N were constructed, has been recently tested (Kaihara et al., 2002; Andres and Bona, 2005). NIRS predictions of carbon and nitrogen have also been reported in the field of soil and compost science, as well as several cases involving plant material (Gilson et al., 1999). The relationship between ultimate data and heating value of biomass resources has been frequently described by predictive models (Demirbas, 1997; Sheng and Azevedo, 2005). Since predictive models already exist for other biomass resources, the present study focused on model development for straw samples only. NIRS was firstly applied for prediction of C, H, N and heating value for a rapid characterization of straw. Then, the relationship between heating value and percentage of C, H, N
1794
C. Huang et al. / Waste Management 29 (2009) 1793–1797
ascending in reference data into validation set, while the remaining samples comprised the calibration set.
Table 1 Statistical results of C, H, N and heating value of rice straw and wheat straw. Rice straw (n = 172)
C (%) H (%) N (%) CV (J/g)
Wheat straw (n = 50)
Range
Mean
SD
Range
Mean
SD
35.20–40.95 4.69–6.19 0.56–1.46 14,191–16,886
38.45 5.28 0.88 15,685
1.14 0.30 0.17 502
40.14–44.97 5.53–6.05 0.13–0.62 16,175–18,041
42.89 5.82 0.36 17281
1.03 0.11 0.13 406
Table 2 Mean, standard deviation (SD) and range for carbon, hydrogen, nitrogen and heating value in straw samples. Calibration
C (%) H (%) N (%) Heating value (J/ g)
Validation
2.2. Spectroscopic analysis Spectra were obtained on a Foss NIRSystem 6500 spectrometer from 400 to 2498 nm with 2-nm increments. Samples were presented in a rectangular quarter cup and scanned at 20–22 °C controlled by an air conditioner. In order to reduce random variability caused by sample loading, samples were repacked and scanned three times and the absorbance of each repack was measured. The average of each set of repeated reading was calculated before the calibration step. 2.3. Reference analysis
Range
Mean
SD
Range
Mean
SD
35.20–44.69 4.69–6.19 0.13–1.46 14,191– 18,041
39.38 5.40 0.77 16,019
2.11 0.35 0.27 810
36.42–44.63 4.82–6.05 0.17–1.40 14,598– 17,953
39.45 5.41 0.78 16,045
2.11 0.35 0.27 808
was determined, and a MLR model predicting heating value from C, H and N was built. The two heating value models were compared to evaluate their accuracy. 2. Materials and methods 2.1. Samples preparation A total of 222 straw samples, collected from 24 provinces of China, were used in the development of NIRS models for prediction of C, H, N and heating value. The samples consisted of rice straw and wheat straw and varied considerably in harvest time, varieties of grain, places of origin, climate where grown, soil types and storage conditions. After being oven dried at 70 °C for 24 h, the samples were ground in a ZM100 mill (Rersch GmbH & Company, Germany) through 1.0-mm sieve, and kept in airtight ziplock bags before spectra collection and reference analysis. The entire set of samples was split into calibration and validation sets, by taking every fourth sample from the samples set
The carbon, hydrogen and nitrogen concentrations in the straw samples were measured using an Elementar Vario EL III analyzer (Elementar Analysensysteme GmbH, Germany) according to CEN standard methods (CEN/TS 15104). Heating value was determined by CEN method (CEN/TS 14918) using an IKA C200 oxygen bomb calorimeter (IKA Analysentechnik GmbH, Heitersheim, Germany). C, H and N were reported in weight percentage on a dry basis, and heating value in J/g on dry basis. 2.4. Calibration and validation The WinISI II software (Infrasoft International LLC, USA) was used for data processing. Calibration equations were developed by performing principal component regression (PCR), partial least square regression (PLS) and modified partial least square regression (MPLS) on spectra data and reference data. The maximum number of principal components (or PLS factors) were set at 16, based on the rule of 1 per 10 samples of the calibration set. Cross validation was employed to avoid over-fitting. Scattering correction was made by standard normal variate (SNV), de-trend (D), standard multiplicative scatter correction (SMSC) and weighted multiplicative scatter correction. Additionally, two derivative treatments of the spectra were compared, (1,4,4,1) and (2,4,4,1), where the first number indicates the order of the derivative, the second number is the gap in data points over which the derivative is calculated, and the third and the fourth number refer to the
Fig. 1. NIR Spectra of straw samples.
1795
C. Huang et al. / Waste Management 29 (2009) 1793–1797 Table 3 Calibration and validation statistics for selected equations of C, H, N and heating value. Calibration
C (%) H (%) N (%) Heating value (J/g)
Validation
R2
SEC
R2cv
SECV
R2v
SEP
RPD
0.98 0.85 0.93 0.97
0.30 0.13 0.07 152
0.97 0.78 0.90 0.96
0.34 0.16 0.08 173
0.97 0.77 0.87 0.96
0.37 0.17 0.10 181
5.64 2.07 2.76 4.47
vel of 0.05. The inclusion of each variable in the proposed model was based on the t-criterion (Nikolaou et al., 2004). The goodness of the model fit was evaluated through examination of various statistical parameters, including regression coefficient (R2), the F statistic, and the a value for linear regression (Uyak et al., 2007). All the procedures, including equation construction and statistics, were executed using Excel 2003 software. 3. Results and discussion 3.1. Reference data
number of the data points used in first and second smoothing, respectively. Statistic parameters as standard error of cross validation (SECV) and the determination coefficient of calibration (R2) were employed to select optimum calibration. The validation set was then introduced to assess the calibration. Besides a high determination coefficient of validation (R2v ) and low standard error of prediction (SEP), relative prediction deviation (RPD, the ratio of the standard deviation of the reference data (SD) to the SEP) was also calculated to evaluate the performance of calibration (Malley et al., 2005). 2.5. MLR equation development Multiple linear regression was adopted to build the equation to predict heating value of straw samples from C, H and N percentages. For the purpose of comparison, the calibration set used in the NIRS equation of heating value was applied to train the MLR model, and the validation set of the NIRS equation was used to examine the prediction performance of the MLR model. Significance evaluations of variables were conducted at a significance le-
Compositional difference between rice straw and wheat straw was expressed with range, mean and standard deviation of C, H, N and heating value in Table 1. It is proved by ANOVA analysis that the N of rice straw was significantly higher than that of wheat straw (p < 0.01); and the C, H and heating value of rice straw were significantly lower than those of wheat straw (p < 0.01). The mean, range and standard deviation of reference data are shown in Table 2. There was a large standard deviation in carbon content between samples while H, N, and heating value were less variable. The means of the four properties are considered to lie within normal ranges and agreed with literature values (Jenkins et al., 1998; Demirbas, 2004; Liao et al., 2004). 3.2. Sample spectra Near infrared spectra of 222 straw samples were illustrated in Fig. 1. The whole spectra were divided into two parts, divided by the discontinuities at 1100 nm where the sensor changed. The spectra range of 400–1100 nm consists of visible light region and
Heating value (J/g)
18000
17000
16000
15000
14000 35
37
39
41
43
45
17000
16000
15000
14000 4. 50
5. 00
C (%)
5. 50
H (%)
18000
Heating value (J/g)
Heating value (J/g)
18000
17000
16000
15000
14000 0. 00
0. 50
1. 00
1. 50
N (%) Fig. 2. Heating value is plotted against the contents of carbon, hydrogen and nitrogen.
6. 00
6. 50
1796
C. Huang et al. / Waste Management 29 (2009) 1793–1797
Table 4 Statistical test result of MLR models for heating value prediction. Regression
Statistics
R R2 Adjusted R2 Standard error of the estimate Observations
0.9629 0.9271 0.9257 221 160 df
Sum of square
Mean square
F
Significance
3 156 159
96,659,781 7,597,705 104,257,485
32,219,927 48,703
662
1.84E88
short wave near infrared region, while 1100–2498 nm belong to the long wave near infrared region. Spectra of 400–1100 nm region were more scattered than those of 1100–2498 nm, and contained more irregular interlacement, decreasing the freedom of spectra interpretability. By contrast, the spectra in 1100–2498 nm region were gathered and comprised more informative peaks. Based on description above, the 1100–2498 nm region was selected for the construction of the NIRS models. The spectra showed absorption bands at 1194 nm related with C– H stretch second overtone, at 1724 nm, 1766 nm related with C–H first overtone, at 1930 nm related with O–H stretch/HOH deformation combination, at 2100 nm related with O–H bond/C–O stretch combination, at 2280 nm related with C–H stretch/CH2 deformation, and at 2236 related with C–H stretch/C–H deformation. 3.3. NIRS equations 3.3.1. Influence of spectra preprocessing and regression A total of 54 of regression equations were obtained from the six scatter correction options, the three derivative treatments and the three regression methods. Overall comparisons were made to examine whether preprocessing and regression methods influence the performance of predictive equations. SECV ranged from 0.33% to 0.50% for C equations, 0.16% to 0.22% for H equations, 0.08% to 0.11% for N equations, and 173 to 202 J/g for heating value equations. The performance of equations for each property was similar, independent of mathematical preprocessing and regression methods. 3.3.2. Equation accuracy Table 3 shows the statistical results obtained in the calibration and validation. It can be seen that R2cv and R2v of C, N, and heating value equations were higher than 0.90 and 0.85, respectively. Very small prediction errors (SEP) were observed compared with the mean values of corresponding components. RPD, a commonly used index for evaluating the performance of NIRS equations, were all higher than 2.0, supporting the applicability of the equations. Particularly, the RPD of C and heating value equations were 5.64 and 4.47, respectively, indicating excellent NIRS equations. 3.4. MLR equation for heating value Straw is primarily composed of elements C, H, O, N, S and Cl, of which C and H make the most positive contribution to heating value, and O makes a negative contribution. The correlation between heating value and C, H, N was calculated in this study. The heating value was plotted against the contents of C, H and N contents (Fig. 2). The heating value increased with increasing of C and H contents, but decreased with an increase of N content. Pearson correlation (Wilcox, 2003) between heating value and carbon (r = 0.94), hydrogen (r = 0.65), nitrogen (r = 0.59) were all significant at a 0.01 level.
Multiple regression analysis was applied to the data in order to find equations that would describe the relationship between heating values and C, H, N contents of straw samples. Prediction models obtained consisted of variables which were statistically significant (p < 0.05), and had the form:
Heating value ¼ 356:08 þ ð432:80 CÞ ð297:73 HÞ þ ð287:45 NÞ Although C, H and N were combustible elements, the Pearson correlation and the regression coefficients in the equation were not all positive. This was caused by the correlation between independent variables such as those between C and H (r = 0.72), as well as those between C and N (r = 0.65). Consequently, the equation can not explain how much each element contributes to the heating value, but as a whole performs well in predicting heating value. The statistical results of the MLR model are presented in Table 4. The correlation coefficient (R) indicates the extent of linear correlation between heating value and C, H and N; R2 explains the percentage of variation in heating value caused by C, H and N; and R2adj eliminated the influence of sample numbers on the R2. All of them were above 0.90, which showed the high correlation between heating value and C, H, N contents. The F value (F = 662) of the equation was extremely large, indicating that the MLR equation was significant. One of the assumptions of analysis of variance (ANOVA) is that variances of the observations in the individual groups are equal, a situations referred to as homogeneity of variance. Ideally, residuals should be nearly constant and independent of dependent variables. A scatter plot of standard residuals against predicted heating value was showed in Fig. 3. The standard residues were randomly distributed, and most of the data points lay within ±2. That suggested that the ANOVA analysis conducted in the previous section was reliable. 3.5. Independent validation for heating value equations Although the procedures and costs of the development of the equations differed, both NIRS and MLR equations predict heating value accurately when evaluated on independent validation. As shown in Fig. 4 the determination coefficients of NIRS and MLR equations were 0.9562 and 0.9273, respectively. The accuracy of
Standard Residue
ANOVA Regression Residual Total
4
0 14000
16000
18000
-4
Heating value (J/g) Fig. 3. Residue plot for MLR heating value equation.
1797
C. Huang et al. / Waste Management 29 (2009) 1793–1797
18000
18000
2
17000
MLR value (J/g)
NIRS value (J/g)
2
R v =0.9562 SEP=181 J/g
16000
15000
14000 14000
15000
16000
17000
18000
Lab value (J/g)
17000
R v =0.9273 SEP=224 J/g
16000
15000
14000 14000
15000
16000
17000
18000
lab value (J/g)
Fig. 4. Independent validation of NIRS equation and MLR equation for predicting heating value.
the NIRS equation (standard error of prediction = 181 J/g) was a little better than that of the MLR equation (standard error of prediction = 224 J/g). NIRS prediction presented greater ease of use and maintainability than prediction of the MLR equation for potential online analysis. However, robustness of the NIRS was greatly influenced by environmental conditions such as temperature and air humidity. 4. Conclusion NIRS prediction of C, H, N contents and heating value of straw samples presented satisfactory accuracy, which is believed to be reliable for quantitative analysis. Particularly, the NIRS equations for C and heating value predictions, judging from the RPD value, are successful for quantitative analysis and therefore eligible for engineering application. The MLR equation for heating value prediction also presents good performance. By the constructed equations, rapid analysis of straw can be achieved in an easier and less costly approach, and satisfactory accuracy of the models allows further possibility to monitor biological or thermochemical conversion processes. References Allica, J.H., Mitre, A.J., Bustamante, J.G., Itoiz, G., Blanco, F., Alkorta, I., Garbisu, C., 2001. Straw quality for its combustion in a straw-fired power plant. Biomass and Bioenergy 21, 249–258. Andres, J.M., Bona, M.T., 2005. Analysis of coal by diffuse reflectance near infrared spectroscopy. Analytica Chimica Acta 35, 123–132.
CEN/TS 15104:2005. Solid Biofuels – Determination of total content of carbon, hydrogen and nitrogen – Instrumental methods. CEN/TS 14918:2005. Solid Biofuels – Method for the determination of calorific value. Demirbas, A., 1997. Calculation of higher heating values of biomass fuels. Fuel 76 (5), 431–434. Demirbas, A., 2004. Combustion characteristics of different biomass fuels. Progress in Energy and Combustion Science 30, 219–230. Gilson, D., Houssard, C., Joffre, R., 1999. Using near infrared reflectance spectroscopy to predict carbon, nitrogen and phosphorus content in heterogeneous plant material. Oecologia 118, 173–182. Han, L., Yan, Q., Liu, X., Hu, J., 2002. Straw resources and their utilization in China. Transaction of the CSAE 18 (3), 87–91. Jenkins, B.M., Baxter, L.L., Miles Jr., T.R., Miles, T.R., 1998. Combustion properties of biomass. Fuel Processing Technology 54, 17–46. Kaihara, M., Takahashi, T., Akazawa, T., Sato, T., Takahashi, S., 2002. Application of near infrared spectroscopy to rapid analysis of coal. Spectroscopy Letters 35 (3), 369–376. Liao, C., Wu, C., Yan, Y., Huang, H., 2004. Chemical elemental characteristics of biomass fuels in China. Biomass and Bioenergy 27, 119–130. Malley, D.F., McClure, C., Martin, P.D., Buckley, K., McCaughey, W.P., 2005. Compositional analysis of cattle manure during composting using a fieldportable near-infrared spectrometer. Communications in Soil Science and Plant Analysis 36, 455–475. Nikolaou, A.D., Golfinopoulos, S.K., Arhonditsis, G.B., Kolovoyiannis, V., Lekkas, T.D., 2004. Modeling the formation of chlorination by-products in river waters with different quality. Chemosphere 55, 409–420. Olz, S., Sims, R. and Kirchner, N., 2007. Contribution of Renewables to Energy Security. IEA Information Paper 2007. Sheng, C., Azevedo, J.L.T., 2005. Estimating the higher heating value of biomass fuels from basic analysis data. Biomass and Bioenergy 28, 499–507. Uyak, V., Ozdemir, K., Toroz, I., 2007. Multiple linear regression modeling of disinfection by-products formation in Istanbul drinking water reservoirs. Science of the Total Environment. doi:10.1016/j.scitotenv.2007.02.041. Wilcox, R.R., 2003. Modern insights about Pearson’s correlation and least squares regression. International Journal of Selection and Assessment 9 (2), 195–205.