- Email: [email protected]
Uncertainty estimation by Bayesian approach in thermochemical conversion of walnut hull and lignite coal blends
Bioresource Technology 232 (2017) 87–92
Contents lists available at ScienceDirect
Bioresource Technology journal homepage: www.elsevier.com/locate/biortech
Uncertainty estimation by Bayesian approach in thermochemical conversion of walnut hull and lignite coal blends Musa Buyukada Abant Izzet Baysal University, Environmental Engineering Department, 14052 Bolu, Turkey
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
Walnut hull was utilized in (co-)
combustion with lignite coal. Thermal behaviors of (co-)
combustion process was investigated. MNLR technique was used to predict
the mass loss percentage. Bayesian approach was found quite
effective in identification of uncertainties.
a r t i c l e
i n f o
Article history: Received 5 January 2017 Received in revised form 2 February 2017 Accepted 4 February 2017 Available online 8 February 2017 Keywords: Walnut hull Co-combustion Non-linear regression Uncertainty analysis Monte Carlo simulation Bayesian approach
a b s t r a c t The main purpose of the present study was to incorporate the uncertainties in the thermal behavior of walnut hull (WH), lignite coal, and their various blends using Bayesian approach. First of all, thermal behavior of related materials were investigated under different temperatures, blend ratios, and heating rates. Results of ultimate and proximate analyses showed the main steps of oxidation mechanism of (co-)combustion process. Thermal degradation started with the (hemi-)cellulosic compounds and finished with lignin. Finally, a partial sensitivity analysis based on Bayesian approach (Markov Chain Monte Carlo simulations) were applied to data driven regression model (the best fit). The main purpose of uncertainty analysis was to point out the importance of operating conditions (explanatory variables). The other important aspect of the present work was the first performance evaluation study on various uncertainty estimation techniques in (co-)combustion literature. Ó 2017 Elsevier Ltd. All rights reserved.
1. Introduction Fossil fuels were the most common used materials in thermalbased electricity production plants. One of these fossil fuels, lignite coal, could be stated as an important primary energy source in the world and it has been generally used in these plants. Although the abundant reserve in the world, lignite coal has been also stated as one of the most important reasons of global warming based on CO2 emissions, and also air pollution. There exist many studies that
E-mail address: [email protected] http://dx.doi.org/10.1016/j.biortech.2017.02.021 0960-8524/Ó 2017 Elsevier Ltd. All rights reserved.
point out this situation in related literature (e.g. Yildiz et al., 2016; Kurekci, 2016; Cepeliogullar and Putun, 2014; Kayahan and Ozdogan, 2016). Considering these hazardous effects, decreasing CO2 emission based on combustion of fossil fuels has become a necessity to be addressed. Many researchers have suggested usage of fossil fuels especially lignite coal with biomass in co-combustion process. So that, many advantages can be gained such as reducing CO2, protecting the air pollution, and also decreasing both of greenhouse gas emissions and fossil fuel consumption. Co-combustion provides these advantages with a cost-effective way as many researchers have already stated (e.g. Hu et al., 2015; Mi et al., 2016). The main goal of co-
88
M. Buyukada / Bioresource Technology 232 (2017) 87–92
combustion of fossil fuels and biomass underlies the phenomena that providing a further stable flame and also a better handle of the burning process depending on increasing the volatile matter content while increasing the mass loss percentage and decreasing the ash content (e.g. Yildiz et al., 2016; Iegorov et al., 2017). Turkey places in quite important geography in the world. As a result of this situation, Turkey has important underground and aboveground resources such as lignite coal, copper, alumina, hazelnut, walnut, peanut, ant etc. (e.g. Ozer et al., 2007; Tanyildizi, 2011; Uzuner and Cekmecelioglu, 2015; Buyukada, 2016; Ozkal and Yener, 2016; Yildiz et al., 2016). These materials have been used in several types of industries for related purposes for example energy and electricity production. Huge amount of hull as biomass sources have been produced as a result of industrial food production. These hull have no alternative uses except for production of domestic animal feed and storage of them causes a disposal problem because of the large volumes. Unfortunately, there are a few studies on utilization and also recovery of this biomass although this potential. Thus, investigation of WH as one of important agricultural waste of Turkey in co-combustion process with lignite coal as one of the most important fossil fuel of Turkey using a thermogravimetric method has been on demand considering this issue. Nowadays, probabilistic uncertainty analyses have also gained an importance in statistical modeling studies. It can be considered a simple and partial sensitivity analysis. There are a lot of techniques to incorporate the uncertainties in proposed model such as likelihood (e.g. Chaudhary and Hantush, 2017), maximum a posterior (e.g. Pan and Pandey, 2016), Monte Carlo (e.g. Evrendilek et al., 2016), Bayesian approach (e.g. Pan and Pandey, 2016; Zhang et al., 2016), and etc. The effect of each explanatory variable on response variable is saved while the explanatory variables keeping constant. This may generally help in identifying the uncertainty in a linear model but it is not applicable for non-linear models because of the larger deviation. The solution is just possible to apply a suitable framework for identification of the uncertainties in model predictors (Evrendilek et al., 2016; Pan and Pandey, 2016). It is certainly clear that there are very few studies on this subject in the related literature. There are a lot of studies on co-combustion of biomass and coal but approximately none of them use WH and lignite coal. Thermochemical conversion of WH and its co-combustion with lignite coal can cotribute to enhancing air quality in Turkey. Additionally, statistical approaches based on uncertainty estimation can be useful for the improvement of design parameters and also researchers. Therefore, the main objectives of the present study is to investigate effects of blend ratio, temperature and heating rate on mass loss percentage (MLP) of (co-) combustion of WH and lignite coal (1), to guide the researchers by stating the importance of uncertainty estimation in determination of the levels of operating conditions using MC simulation in the absence and presence of Bayesian approach (MCMC) (2).
C70WH30, C80WH20, and C90WH10 (%, wt to wt). After the temperature arrangement to 25 °C had been done with a heating rate of 20 °C min 1 in the atmosphere, the temperature was raised to 1000 °C with various heating rates of 10, 20 and 30 °C min 1 (10.00 ± 0.5 mg of samples). The air flow was determined as 20 mL min 1 and it was kept constant. The mass loss percentage (MLP, %) and the mass loss rate (MLR, % min 1) of the samples were analyzed under non-isothermal conditions. All the runs were performed with three replicates to determine the standard deviation and to prevent the experimental errors. 2.3. Uncertainty estimation Mass loss percentage (MLP, %, response to co-combustion process) as a function of three explanatory variables of (1) blend ratio (wt%), (2) heating rate (°C min 1), and (3) temperature (°C) was investigated by MNLR modeling technique considering the adjusted regression coefficient (R2adj) and predictive regression coefficient (R2pred). Predictors in the best-fit MNLR model were chosen by following the best-subset procedure. The main goal of the bestsubset procedure was to choose the proposed MNLR model which having the maximum values of adjusted and predicted R2 values with the lowest number of predictors (except for intercept), the lowest multicollinearity (considering variation inflation factor, VIF), and the lowest autocorrelation (considering Durbin Watson factor, D-W). Monte Carlo (MC) and its modified form based on Bayesian approach (Markov Chain Monte Carlo, MCMC) simulations were used to analyze the uncertainties in predictors of proposed MNLR model. Thereby, a comparison study and also a partial sensitivity analysis were performed on estimation of uncertainty. Minitab 17 (Minitab PA), Matlab 2012b (Matlab, USA), and Microsoft Excel-based Model Risk (trial version, Vose Software) were used for MNLR, MC, and MCMC analyses, respectively. 3. Results and discussion 3.1. Proximate analyses
Lignite coal samples were provided from Chemical Engineering Department of Fırat University, Elazig, Turkey. Walnut hull samples were provided from a local domestic feed production plant, Giresun, Turkey. A similar pretreatment procedure was applied to samples that Buyukada (2017) had performed.
Volatile content, harmful gas emissions (SOx and NOx), and ash amount (including fixed carbon) of pure lignite coal, pure WH, and their various blend ratios were investigated by proximate analyses and related results were given in Table 1. The first result of proximate analyses showed that volatile content of coal and WH were 28% and 56%, respectively. This finding showed that WH can be removed at a lower temperature than lignite coal. The second finding was about reduction of harmful gas release. Utilization of WH with lignite coal in a co-combustion process can be resulted as a reduction in SOx and NOx production according to combustion of the pure lignite coal. This claim was justified by the following third finding that was about ash contents. When considering the ash contents of WH and coal (as 4% and 48%, respectively on dry basis), these data supports significantly the expectation that the blends of WH and lignite coal may be beneficial for reduction of waste combustion. Furthermore, higher ash amounts may also decrease the efficiency of the combustion process considering the formation of aggregates and decrease of the coefficients of mass and heat transfer. All the findings in this part were in good accordance with the results of similar studies (Buyukada, 2016; Kayahan and Ozdogan, 2016; Yildiz et al., 2016; Atimtay et al., 2017; Chen et al., 2017; Yao et al., 2017).
2.2. Co-combustion process
3.2. Ultimate analyses
Co-combustion experiments were performed by the lignite coal (C), walnut hull (WH), and their various blend ratios of C60WH40,
Results of elemental findings were obtained by ultimate analyses and given in Table 1. As a result of ultimate analyses, 58% and
2. Materials and methods 2.1. Samples
89
M. Buyukada / Bioresource Technology 232 (2017) 87–92 Table 1 Results of ultimate and proximate analyses of lignite coal and walnut hull on dried basis. Sample
Ultimate analyses (wt%)
Lignite coal Walnut hull *
Proximate analyses (wt%)
Heating values (x10 3 kcal kg 1)
C
H
O
N
S
Moisture
Volatile matter
Ash
Fixed carbon*
Lower
Higher
68.20 61.40
11.08 4.31
6.35 34.06
1.61 0.2
12.76 0.03
11.70 13.57
28.71 56.04
48.04 4.06
11.55 26.33
3.5 3.8
3.7 4.1
Calculated by difference.
61% of C content, 11% and 4% of H content, and 6% and 34% of O content were determined for lignite coal and WH, respectively. Additionally, lower and higher heating values were estimated as 3.5 and 3.7 kcal kg 1, and 3.8 and 4.1 kcal kg 1 for the same order. These results lead a decrease both in lower and higher heating value, thus it may cause an increase in the thermal efficiency during combustion simultaneously. Results of similar studies showed that there was a good agreement between the findings of this part of study and related literature (Atimtay et al., 2017; Chen et al., 2017; Yao et al., 2017).
cellulose structures (between 180 and 375 °C). Thus, the second step also showed the release of volatile compounds. This caused to form of char products. The last and the third step was lignin (as the most stable chemical component in WH) degradation (between 400 and 540 °C). These findings stated the thermal behaviors of pure WH, C, and their various blends clearly. Additionally, this step can be also stated as an indicator that showed the biomass-based char formation depending on lignin degradation. (Li et al., 2016; Yildiz et al., 2016; Kayahan and Ozdogan, 2016; Mi et al., 2016; Myllari et al., 2017; Parvez and Wu, 2017).
3.3. Mass loss percentage (%) and rate (% min 1)
3.4. Thermal behaviors and mechanism steps
Mass loss percentage and rate of (co-)combustion of WH and lignite coal were illustrated in Fig. 1. The certain differences were detected in the thermal behavior of WH and lignite coal samples. These differences can be listed as follows: The first step was removal of low molecular compounds including moisture (to 180 °C). This step was followed by the decomposition of (hemi-)
The main levels of oxidation of co-combustion of WH and lignite coal was pointed out in Fig. 1. Degradation of (hemi-) cellulosic compounds can be clearly seen between the temperatures of 240 and 320 °C. The other level was the release of volatile organic matters (up to 550 °C). Furthermore, devolatilization and co-combustion of blends were the main role of heating rate. This phenomena underlied the significant effect of the heating rate on acceleration of devolatilization (Fig. 2). Effect of increasing heating rate caused to form a shift to a wider temperature range in
Mass Loss Percentage (%)
100
Pure lignite coal (C) C60WH40 C70WH30 C80WH20 C90WH10 WH
80 60 40 20 0 0
200
0
200
(b)
400 600 Temperature (°C)
Temperature (°C) 400 600
800
1000
800
1000
(a) 100 Mass Loss Percentage (%)
(a)
Lignite coal - 10°C/min 80
Lignite coal - 20°C/min
60
Lignite coal - 30°C/min
40
WH - 10°C/min WH - 20°C/min
20
WH - 30°C/min
0 0
200
400 600 800 Temperature (°C)
(b)
0 0
200
Temperature (°C) 400 600
800
1000
0
-5 -10 -15
Pure lignite coal (C) C60WH40 C70WH30 C80WH20 C90WH10 WH
Mass Loss Rate (%/min)
Mass Loss Rate (%/min)
1000
-5 Lignite coal - 10°C/min -10
Lignite coal - 20°C/min Lignite coal - 30°C/min
-15
WH - 10°C/min WH - 20°C/min
-20
-20
WH - 30°C/min
-25 Fig. 1. Effect of blend ratio and temperature on mass loss (a) percentage (b) rate (under a constant heating rate of 20 °C min 1).
Fig. 2. Effect of heating rate and temperature on mass loss (a) percentage (b) rate.
M. Buyukada / Bioresource Technology 232 (2017) 87–92
combustion process in Fig. 2. This can be explained by formation of shifts on the devolatilization peak. The higher heating rate provided a shorter time period to reach a given temperature. The similar findings were reported for both thermal degradation and oxidation steps by researchers (e.g. Buyukada, 2016, 2017; Kayahan and Ozdogan, 2016; Mi et al., 2016; Yildiz et al., 2016; Atimtay et al., 2017). This showed that the findings of the present study were in good accordance with the related literature. 3.5. Multiple non-linear regression modeling To predict the mass loss percentage of co-combustion of WH and lignite coal blends, a MNLR model was derived with the highest adjusted and predicted regression coefficients. Because,
Table 2 Results of proposed multiple non-linear regression model (R2 of 99.9%, R2adj of 99.5%, R2pred of 97.5%, S of 0.52, DW of 1.33). Source**
p
Model* Intercept Heating rate* (HR, °C min Temperature (T, °C) Blend ratio* (BR, %WH) HR*T HR*BR BR*T HR2,* T2,* BR2,* T3 T4
<0.05 1
)
<0.05 =0.0002 <0.05 =0.73 =0.007 =0.002 <0.05 <0.05 <0.05 =0.0028 =0.015
VIF
Coefficients
1.01 1.01 1.01 10.33 1.07 1.07 1.01 1.03 1.01 13.51 13.55
127.28 2.12 1.24 1.67 0.11 1.05 0.88 17.12 24.35 11.34 0.0000023 0.0000001
(a)
20
30
60 72 Parameter values
86
44.4
44.6 44.8 Parameter values
50
50.2
43.0
43.2 43.4 Parameter values
43.6
43.8
Min
Median
Max
Blend ratio1 (%C) Heating rate2 (°C min 1) Temperature2 (°C)
75 20
±15 ±10
60 10
80 20
90 30
515
±490
25
600
1000
0
Standard deviations
Categorical variable (dimensionless). Numeric variable.
(c)
Temperature
50
Heating rate Blend ratio
2.5
Distribution function
100
44.2
10
2
Mean
Frequency 5 7.5
1
Explanatory variables
94
Frequency 5 7.5
Table 3 Descriptive statistics of explanatory variables of co-combustion process.
(b)
46
10
*
2.5
p value is lower than 0.0001. D–W: Durbin Watson statistic, R2: Regression coefficient, R2adj: Adjusted regression coefficient, R2pred: Predicted regression coefficient, S: Standard error, VIF: Variation inflation factor. ** Predictors in proposed MNLR model were selected by following the best-subset procedure.
adjusted and predicted regression coefficients were the only way to explain the variation in response variable and also predict it, respectively. The predictors of proposed MNLR model were given in Table 2. As it is seen, adjusted and predicted regression coefficients were determined as 0.997 and 0.975, respectively. Although these statistical parameters demonstrated the powerful side of derived MNLR model significantly, they were not enough to state the proposed MNLR model totally. For this issue, some additional statistical parameters should be given for further explanation of proposed MNLR model. Durbin Watson (DW) factor and variation inflation factor (VIF) can be stated as indicators that showed the autocorrelation and multicollinearity, respectively. To meet the criteria that underlies the assumption in design of MNLR model, Durbin Watson and variation inflation factors must be in a range of 0 and 2, and 0 and 10, respectively (Buyukada and Evrendilek, 2016; Pan and Pandey, 2016; Buyukada, 2017). As it is seen in Table 2, DW and VIF values were estimated as 1.33 and lower than 10, respectively. This situation also showed that the proposed MNLR model met the assumption criteria, and it can be certainly
Frequency 0 1000 2000 3000
90
0
0 0
20000
40000 60000 Randomly generation
80000
100000
Fig. 3. Result of Monte Carlo simulations on cumulative distributive functions of explanatory variables of co-combustion process.
42.8
Fig. 4. Results of MLP predictions by (a) MNLR (b) MC simulation (c) MCMC simulation.
M. Buyukada / Bioresource Technology 232 (2017) 87–92
used in prediction of mass loss percentage of co-combustion of WH and lignite coal as response variable. Although the proposed MNLR model seemed to be competent, it also contained unexplained parts in itself that must be addressed. For this issue, an uncertainty estimation technique must be used. So, probabilistic uncertainty analysis based on MC simulations in the absence and presence of Bayesian approach (MCMC) as one of the main novel contributors of the present study has been on demand. 3.6. Monte Carlo simulations The necessity of this approach could be listed as follows: to identify the uncertainties in selected predictors, because these uncertainties could not be captured through by deterministic ways (1), and to demonstrate the importance of operating parameters on
91
MLP (2). To apply this approach, randomly generated predictors were used to run the proposed MNLR model. In another terms, the MNLR model was also chosen according prediction accuracy and stochastic predictors. This choosing criteria depended on the absence of any meaningful MNLR model (Table 3). Distributive functions for probabilistic changes in blend ratio, heating rate, and temperature were illustrated in Fig 3. The good agreement was found between mean (obtained experimentally) and Monte Carlo (obtained by simulations) values of mass loss percentage (Fig 3). 3.7. Markov Chain Monte Carlo simulations Fig. 4 showed the conditional predictions for each individual co-varieties that obtained by keeping all the other parameters
Fig. 5. Conditional predictors for (a) MC (b) MCMC simulations.
92
M. Buyukada / Bioresource Technology 232 (2017) 87–92
constant to their mean values. Mean values of proposed MNLR model, MC, and also MCMC simulations were determined as 54.5, 45.5, and 44.7, respectively (Fig. 4). The differences in these mean values demonstrated the uncertainties in derived MNLR model. As shown in Fig. 5, most of the temperature points were quite closely distributed in the whole range of the abscissa. The main lines in Fig. 5 showed the minimum and maximum levels of each explanatory variables while the dotted line between main lines (min and max) showed the ideal places (that any uncertainty should not be occupied under ideal conditions). The dots on the dotted line showed the experimental results and the dots around this line showed the deviation from ideality. Fig. 5 demonstrated the uncertainties in linear and quadratic effects of temperature in proposed MNLR model clearly. Quite close main lines at the beginning level of these explanatory variables indicated an uncertainty that should be identified. Differently, this situation was approximately not applicable for heating rate, and it was moderate applicable for blend ratio (Fig. 5). Therefore the uncertainties in the conditional predictions were smaller near the data-points. The same situation was not applicable for heating rate because it approximately covered the whole range of the abscissa uniformly and therefore it had similar levels of uncertainty predictions throughout. Even though the situation of blend ratio was not clear and it had a random distribution of the data points, it can be said that the uncertainties in the conditional predictions was less than temperature’s. Fig. 5 pointed out that conducting further experiments with lower temperature and wider range of blend ratio values was absolutely necessary. It was very important to improve the predictions. Heating rate was needed to be varied much in order to get more confident predictions. Thus, it was so clear and also necessary to identify the different experimental designs for cocombustion of WH and lignite coal could be conducted and the data points could be used to significantly improve the predictive power of the models. The findings about uncertainty estimation seemed to be in great accordance with the results of similar studies and this situation demonstrated the importance of uncertainty analyses on energy related processes (e.g. Buyukada, 2016, 2017; Evrendilek et al., 2016; Pan and Pandey, 2016; Zhang et al., 2016; Chaudhary and Hantush, 2017).
4. Conclusions Results of the thermogravimetric analyses showed that increasing WH amount in blend ratio caused a decrease in ash amount. When the position of Turkey in global walnut production were taken into consideration, the findings of present study may be useful for thermal plants. (MC-)MC simulations were applied to (co-) combustion of WH, lignite coal, and their various blends to incorporate the uncertainties in predictors of data-driven regression model. Results pointed out that deterministic approaches was inadequate to identify the uncertainties. Furthermore, Bayesian approach was found as fairly powerful technique to capture through the stochastic uncertainties in the predictors of MNLR model. Acknowledgement The author would like to acknowledge Dr. G.A. Evrendilek, the supervisor of YENIGIDAM Research Center (AIBU, Bolu, Turkey), for her kindly helps in laboratory analysis and E. Aydogmus from
Chemical Engineering Department of Firat University (FU, Elazig, Turkey) for him invaluable helps in providing lignite coals. References Atimtay, A.T., Kayahan, U., Unlu, A., Engin, B., Varol, M., Olgun, H., Atakul, H., 2017. Co-firing of pine chips in 750 kWth circulating fludized bed combustion system. Bioresour. Technol. 224, 601–610. Buyukada, M., 2016. Co-combustion of peanut hull and coal blends: Artificial neural networks modeling, particle swarm optimization and Monte Carlo Simulation. Bioresour. Technol. 216, 280–286. Buyukada, M., 2017. Probabilistic uncertainty analysis based on Monte Carlo simulations of co-combustion of hazelnut hull and coal blends: data driven modeling and response surface optimization. Bioresour. Technol. 225, 106–112. Buyukada, M., Evrendilek, F., 2016. Modeling efficiency of dehydrated sunflower seed cake as a novel biosorbent to remove a toxic azo dye. Chem. Eng. Commun. 203 (6), 746–757. Cepeliogullar, O., Putun, A.E., 2014. Products characterization study of a slow pyrolysis of biomass-plastic mixtures in a fixed-bed reactor. J. Anal. App. Pyrol. 110 (1), 363–374. Chaudhary, A., Hantush, M.M., 2017. Bayesian Monte Carlo and maximum likelihood approach for uncertainty estimation and risk management. Application to lake oxygen recovery model. Water Res. 108, 301–311. Chen, J., Liu, J., He, Y., Huang, L., Sun, S., Sun, J., Chang, K., Kuo, J., Huang, S., Ning, X., 2017. Investigation of co-combustion characteristics of sewage sludge and coffee grounds mixtures using thermogravimetric analysis coupled to artificial neural networks modeling. Bioresour. Technol. 225, 234–245. Evrendilek, G.A., Avsar, Y.K., Evrendilek, F., 2016. Modelling stochastic variability and uncertainty in aroma active compounds of PEF-treated peach nectar as a function of physical and sensory properties, and treatment time. Food Chem. 190, 634–642. Hu, S., Ma, X., Lin, Y., Yu, Z., Fang, S., 2015. Thermogravimetric analysis of the cocombustion of paper mill sludge and municipal solid waste. Energy Convers. Manage. 99, 112–118. Iegorov, R., Valiullin, T., Strizhak, P., 2017. Co-combustion of the coal-water slurry doped by combustible and non-combustible nano-particles. Appl. Therm. Energy 113, 1012–1023. Kayahan, U., Ozdogan, S., 2016. Oxygen enriched combustion and co-combustion of lignites and biomass in a 30 kWth circulating fluidized bed. Energy 116, 317– 328. Kurekci, N.A., 2016. Determination of optimum insulation thickness for building walls by using heating and cooling degree-day values of all Turkey’s provincial centers. Energy Build. 118, 197–213. Li, M., Wang, J., Yang, Y., Xie, G., 2016. Alkali-based pretreatments distinctively extract lignin and pectin for enhancing biomass saccharification by alterating cellulose features in sugar-rich Jerusalem artichoke stem. Bioresour. Technol. 208, 31–41. Mi, B., Liu, Z., Hu, W., Wei, P., Jiang, Z., Fei, B., 2016. Investigating pyrolysis and combustion characteristics of torrefied bamboo, torrefied wood and their blends. Bioresour. Technol. 209. 55-55. Myllari, F., Karjalanien, P., Taipale, R., Aalto, P., Hayrinen, A., Rautiainen, J., et al., 2017. Physical and chemical characteristics of flue-gas particles in a large pulverized fuel-fired power plant boiler during co-combustion of coal and wool pellets. Combust. Flame 176, 554–566. Ozer, D., Dursun, G., Ozer, A., 2007. Methylene blue adsorption from aqueous solution by dehydrated peanut hull. J. Hazard. Mater. 144 (1–2), 171–179. Ozkal, S.G., Yener, M.E., 2016. Supercritical carbon dioxide extraction of flaxseed oil: effect of extraction parameters and mass transfer modeling. J. Supercrit. Fluid 112, 76–80. Pan, I., Pandey, D.S., 2016. Incorporating uncertainty in data driven regression models of fluidized bed gasification: a Bayesian approach. Fuel Proces. Technol. 142, 305–314. Parvez, A.M., Wu, T., 2017. Characteristics and interactions between coal and carbonaceous wastes during co-combustion. J. Energy Inst. 90, 12–20. Tanyildizi, M.S., 2011. Modeling of adsorption isotherms and kinetics of reactive dye from aqueous solution by peanut hull. Chem. Eng. J. 168 (3), 1234–1240. Uzuner, S., Cekmecelioglu, D., 2015. Enhanced pectinase production by optimizing fermentation conditions of Bacillus subtilis growing on hazelnut shell hydrolyzate. J. Molec. Catal. B Enzym. 113, 62–67. Yao, Z., Ma, X., Wang, Z., Chen, L., 2017. Characteristics of co-combustion and kinetic study on hydrochar with oil shale: a thermogravimetric analysis. Appl. Therm. Energy 110, 1420–1427. Yildiz, Z., Uzun, H., Ceylan, S., Topcu, Y., 2016. Application of artificial neural networks to co-combustion hazelnut husk-lignite coal blends. Bioresour. Technol. 200, 42–47. Zhang, J., Li, Y., Huang, G., Chen, X., Bao, A., 2016. Assessment of parameter uncertainty in hydrological model using a Markov-Chain-Monte-Carlo-based multilevel-factorial-analysis method. J. Hyrdol. 538, 471–486.
Contents lists available at ScienceDirect
Bioresource Technology journal homepage: www.elsevier.com/locate/biortech
Uncertainty estimation by Bayesian approach in thermochemical conversion of walnut hull and lignite coal blends Musa Buyukada Abant Izzet Baysal University, Environmental Engineering Department, 14052 Bolu, Turkey
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
Walnut hull was utilized in (co-)
combustion with lignite coal. Thermal behaviors of (co-)
combustion process was investigated. MNLR technique was used to predict
the mass loss percentage. Bayesian approach was found quite
effective in identification of uncertainties.
a r t i c l e
i n f o
Article history: Received 5 January 2017 Received in revised form 2 February 2017 Accepted 4 February 2017 Available online 8 February 2017 Keywords: Walnut hull Co-combustion Non-linear regression Uncertainty analysis Monte Carlo simulation Bayesian approach
a b s t r a c t The main purpose of the present study was to incorporate the uncertainties in the thermal behavior of walnut hull (WH), lignite coal, and their various blends using Bayesian approach. First of all, thermal behavior of related materials were investigated under different temperatures, blend ratios, and heating rates. Results of ultimate and proximate analyses showed the main steps of oxidation mechanism of (co-)combustion process. Thermal degradation started with the (hemi-)cellulosic compounds and finished with lignin. Finally, a partial sensitivity analysis based on Bayesian approach (Markov Chain Monte Carlo simulations) were applied to data driven regression model (the best fit). The main purpose of uncertainty analysis was to point out the importance of operating conditions (explanatory variables). The other important aspect of the present work was the first performance evaluation study on various uncertainty estimation techniques in (co-)combustion literature. Ó 2017 Elsevier Ltd. All rights reserved.
1. Introduction Fossil fuels were the most common used materials in thermalbased electricity production plants. One of these fossil fuels, lignite coal, could be stated as an important primary energy source in the world and it has been generally used in these plants. Although the abundant reserve in the world, lignite coal has been also stated as one of the most important reasons of global warming based on CO2 emissions, and also air pollution. There exist many studies that
E-mail address: [email protected] http://dx.doi.org/10.1016/j.biortech.2017.02.021 0960-8524/Ó 2017 Elsevier Ltd. All rights reserved.
point out this situation in related literature (e.g. Yildiz et al., 2016; Kurekci, 2016; Cepeliogullar and Putun, 2014; Kayahan and Ozdogan, 2016). Considering these hazardous effects, decreasing CO2 emission based on combustion of fossil fuels has become a necessity to be addressed. Many researchers have suggested usage of fossil fuels especially lignite coal with biomass in co-combustion process. So that, many advantages can be gained such as reducing CO2, protecting the air pollution, and also decreasing both of greenhouse gas emissions and fossil fuel consumption. Co-combustion provides these advantages with a cost-effective way as many researchers have already stated (e.g. Hu et al., 2015; Mi et al., 2016). The main goal of co-
88
M. Buyukada / Bioresource Technology 232 (2017) 87–92
combustion of fossil fuels and biomass underlies the phenomena that providing a further stable flame and also a better handle of the burning process depending on increasing the volatile matter content while increasing the mass loss percentage and decreasing the ash content (e.g. Yildiz et al., 2016; Iegorov et al., 2017). Turkey places in quite important geography in the world. As a result of this situation, Turkey has important underground and aboveground resources such as lignite coal, copper, alumina, hazelnut, walnut, peanut, ant etc. (e.g. Ozer et al., 2007; Tanyildizi, 2011; Uzuner and Cekmecelioglu, 2015; Buyukada, 2016; Ozkal and Yener, 2016; Yildiz et al., 2016). These materials have been used in several types of industries for related purposes for example energy and electricity production. Huge amount of hull as biomass sources have been produced as a result of industrial food production. These hull have no alternative uses except for production of domestic animal feed and storage of them causes a disposal problem because of the large volumes. Unfortunately, there are a few studies on utilization and also recovery of this biomass although this potential. Thus, investigation of WH as one of important agricultural waste of Turkey in co-combustion process with lignite coal as one of the most important fossil fuel of Turkey using a thermogravimetric method has been on demand considering this issue. Nowadays, probabilistic uncertainty analyses have also gained an importance in statistical modeling studies. It can be considered a simple and partial sensitivity analysis. There are a lot of techniques to incorporate the uncertainties in proposed model such as likelihood (e.g. Chaudhary and Hantush, 2017), maximum a posterior (e.g. Pan and Pandey, 2016), Monte Carlo (e.g. Evrendilek et al., 2016), Bayesian approach (e.g. Pan and Pandey, 2016; Zhang et al., 2016), and etc. The effect of each explanatory variable on response variable is saved while the explanatory variables keeping constant. This may generally help in identifying the uncertainty in a linear model but it is not applicable for non-linear models because of the larger deviation. The solution is just possible to apply a suitable framework for identification of the uncertainties in model predictors (Evrendilek et al., 2016; Pan and Pandey, 2016). It is certainly clear that there are very few studies on this subject in the related literature. There are a lot of studies on co-combustion of biomass and coal but approximately none of them use WH and lignite coal. Thermochemical conversion of WH and its co-combustion with lignite coal can cotribute to enhancing air quality in Turkey. Additionally, statistical approaches based on uncertainty estimation can be useful for the improvement of design parameters and also researchers. Therefore, the main objectives of the present study is to investigate effects of blend ratio, temperature and heating rate on mass loss percentage (MLP) of (co-) combustion of WH and lignite coal (1), to guide the researchers by stating the importance of uncertainty estimation in determination of the levels of operating conditions using MC simulation in the absence and presence of Bayesian approach (MCMC) (2).
C70WH30, C80WH20, and C90WH10 (%, wt to wt). After the temperature arrangement to 25 °C had been done with a heating rate of 20 °C min 1 in the atmosphere, the temperature was raised to 1000 °C with various heating rates of 10, 20 and 30 °C min 1 (10.00 ± 0.5 mg of samples). The air flow was determined as 20 mL min 1 and it was kept constant. The mass loss percentage (MLP, %) and the mass loss rate (MLR, % min 1) of the samples were analyzed under non-isothermal conditions. All the runs were performed with three replicates to determine the standard deviation and to prevent the experimental errors. 2.3. Uncertainty estimation Mass loss percentage (MLP, %, response to co-combustion process) as a function of three explanatory variables of (1) blend ratio (wt%), (2) heating rate (°C min 1), and (3) temperature (°C) was investigated by MNLR modeling technique considering the adjusted regression coefficient (R2adj) and predictive regression coefficient (R2pred). Predictors in the best-fit MNLR model were chosen by following the best-subset procedure. The main goal of the bestsubset procedure was to choose the proposed MNLR model which having the maximum values of adjusted and predicted R2 values with the lowest number of predictors (except for intercept), the lowest multicollinearity (considering variation inflation factor, VIF), and the lowest autocorrelation (considering Durbin Watson factor, D-W). Monte Carlo (MC) and its modified form based on Bayesian approach (Markov Chain Monte Carlo, MCMC) simulations were used to analyze the uncertainties in predictors of proposed MNLR model. Thereby, a comparison study and also a partial sensitivity analysis were performed on estimation of uncertainty. Minitab 17 (Minitab PA), Matlab 2012b (Matlab, USA), and Microsoft Excel-based Model Risk (trial version, Vose Software) were used for MNLR, MC, and MCMC analyses, respectively. 3. Results and discussion 3.1. Proximate analyses
Lignite coal samples were provided from Chemical Engineering Department of Fırat University, Elazig, Turkey. Walnut hull samples were provided from a local domestic feed production plant, Giresun, Turkey. A similar pretreatment procedure was applied to samples that Buyukada (2017) had performed.
Volatile content, harmful gas emissions (SOx and NOx), and ash amount (including fixed carbon) of pure lignite coal, pure WH, and their various blend ratios were investigated by proximate analyses and related results were given in Table 1. The first result of proximate analyses showed that volatile content of coal and WH were 28% and 56%, respectively. This finding showed that WH can be removed at a lower temperature than lignite coal. The second finding was about reduction of harmful gas release. Utilization of WH with lignite coal in a co-combustion process can be resulted as a reduction in SOx and NOx production according to combustion of the pure lignite coal. This claim was justified by the following third finding that was about ash contents. When considering the ash contents of WH and coal (as 4% and 48%, respectively on dry basis), these data supports significantly the expectation that the blends of WH and lignite coal may be beneficial for reduction of waste combustion. Furthermore, higher ash amounts may also decrease the efficiency of the combustion process considering the formation of aggregates and decrease of the coefficients of mass and heat transfer. All the findings in this part were in good accordance with the results of similar studies (Buyukada, 2016; Kayahan and Ozdogan, 2016; Yildiz et al., 2016; Atimtay et al., 2017; Chen et al., 2017; Yao et al., 2017).
2.2. Co-combustion process
3.2. Ultimate analyses
Co-combustion experiments were performed by the lignite coal (C), walnut hull (WH), and their various blend ratios of C60WH40,
Results of elemental findings were obtained by ultimate analyses and given in Table 1. As a result of ultimate analyses, 58% and
2. Materials and methods 2.1. Samples
89
M. Buyukada / Bioresource Technology 232 (2017) 87–92 Table 1 Results of ultimate and proximate analyses of lignite coal and walnut hull on dried basis. Sample
Ultimate analyses (wt%)
Lignite coal Walnut hull *
Proximate analyses (wt%)
Heating values (x10 3 kcal kg 1)
C
H
O
N
S
Moisture
Volatile matter
Ash
Fixed carbon*
Lower
Higher
68.20 61.40
11.08 4.31
6.35 34.06
1.61 0.2
12.76 0.03
11.70 13.57
28.71 56.04
48.04 4.06
11.55 26.33
3.5 3.8
3.7 4.1
Calculated by difference.
61% of C content, 11% and 4% of H content, and 6% and 34% of O content were determined for lignite coal and WH, respectively. Additionally, lower and higher heating values were estimated as 3.5 and 3.7 kcal kg 1, and 3.8 and 4.1 kcal kg 1 for the same order. These results lead a decrease both in lower and higher heating value, thus it may cause an increase in the thermal efficiency during combustion simultaneously. Results of similar studies showed that there was a good agreement between the findings of this part of study and related literature (Atimtay et al., 2017; Chen et al., 2017; Yao et al., 2017).
cellulose structures (between 180 and 375 °C). Thus, the second step also showed the release of volatile compounds. This caused to form of char products. The last and the third step was lignin (as the most stable chemical component in WH) degradation (between 400 and 540 °C). These findings stated the thermal behaviors of pure WH, C, and their various blends clearly. Additionally, this step can be also stated as an indicator that showed the biomass-based char formation depending on lignin degradation. (Li et al., 2016; Yildiz et al., 2016; Kayahan and Ozdogan, 2016; Mi et al., 2016; Myllari et al., 2017; Parvez and Wu, 2017).
3.3. Mass loss percentage (%) and rate (% min 1)
3.4. Thermal behaviors and mechanism steps
Mass loss percentage and rate of (co-)combustion of WH and lignite coal were illustrated in Fig. 1. The certain differences were detected in the thermal behavior of WH and lignite coal samples. These differences can be listed as follows: The first step was removal of low molecular compounds including moisture (to 180 °C). This step was followed by the decomposition of (hemi-)
The main levels of oxidation of co-combustion of WH and lignite coal was pointed out in Fig. 1. Degradation of (hemi-) cellulosic compounds can be clearly seen between the temperatures of 240 and 320 °C. The other level was the release of volatile organic matters (up to 550 °C). Furthermore, devolatilization and co-combustion of blends were the main role of heating rate. This phenomena underlied the significant effect of the heating rate on acceleration of devolatilization (Fig. 2). Effect of increasing heating rate caused to form a shift to a wider temperature range in
Mass Loss Percentage (%)
100
Pure lignite coal (C) C60WH40 C70WH30 C80WH20 C90WH10 WH
80 60 40 20 0 0
200
0
200
(b)
400 600 Temperature (°C)
Temperature (°C) 400 600
800
1000
800
1000
(a) 100 Mass Loss Percentage (%)
(a)
Lignite coal - 10°C/min 80
Lignite coal - 20°C/min
60
Lignite coal - 30°C/min
40
WH - 10°C/min WH - 20°C/min
20
WH - 30°C/min
0 0
200
400 600 800 Temperature (°C)
(b)
0 0
200
Temperature (°C) 400 600
800
1000
0
-5 -10 -15
Pure lignite coal (C) C60WH40 C70WH30 C80WH20 C90WH10 WH
Mass Loss Rate (%/min)
Mass Loss Rate (%/min)
1000
-5 Lignite coal - 10°C/min -10
Lignite coal - 20°C/min Lignite coal - 30°C/min
-15
WH - 10°C/min WH - 20°C/min
-20
-20
WH - 30°C/min
-25 Fig. 1. Effect of blend ratio and temperature on mass loss (a) percentage (b) rate (under a constant heating rate of 20 °C min 1).
Fig. 2. Effect of heating rate and temperature on mass loss (a) percentage (b) rate.
M. Buyukada / Bioresource Technology 232 (2017) 87–92
combustion process in Fig. 2. This can be explained by formation of shifts on the devolatilization peak. The higher heating rate provided a shorter time period to reach a given temperature. The similar findings were reported for both thermal degradation and oxidation steps by researchers (e.g. Buyukada, 2016, 2017; Kayahan and Ozdogan, 2016; Mi et al., 2016; Yildiz et al., 2016; Atimtay et al., 2017). This showed that the findings of the present study were in good accordance with the related literature. 3.5. Multiple non-linear regression modeling To predict the mass loss percentage of co-combustion of WH and lignite coal blends, a MNLR model was derived with the highest adjusted and predicted regression coefficients. Because,
Table 2 Results of proposed multiple non-linear regression model (R2 of 99.9%, R2adj of 99.5%, R2pred of 97.5%, S of 0.52, DW of 1.33). Source**
p
Model* Intercept Heating rate* (HR, °C min Temperature (T, °C) Blend ratio* (BR, %WH) HR*T HR*BR BR*T HR2,* T2,* BR2,* T3 T4
<0.05 1
)
<0.05 =0.0002 <0.05 =0.73 =0.007 =0.002 <0.05 <0.05 <0.05 =0.0028 =0.015
VIF
Coefficients
1.01 1.01 1.01 10.33 1.07 1.07 1.01 1.03 1.01 13.51 13.55
127.28 2.12 1.24 1.67 0.11 1.05 0.88 17.12 24.35 11.34 0.0000023 0.0000001
(a)
20
30
60 72 Parameter values
86
44.4
44.6 44.8 Parameter values
50
50.2
43.0
43.2 43.4 Parameter values
43.6
43.8
Min
Median
Max
Blend ratio1 (%C) Heating rate2 (°C min 1) Temperature2 (°C)
75 20
±15 ±10
60 10
80 20
90 30
515
±490
25
600
1000
0
Standard deviations
Categorical variable (dimensionless). Numeric variable.
(c)
Temperature
50
Heating rate Blend ratio
2.5
Distribution function
100
44.2
10
2
Mean
Frequency 5 7.5
1
Explanatory variables
94
Frequency 5 7.5
Table 3 Descriptive statistics of explanatory variables of co-combustion process.
(b)
46
10
*
2.5
p value is lower than 0.0001. D–W: Durbin Watson statistic, R2: Regression coefficient, R2adj: Adjusted regression coefficient, R2pred: Predicted regression coefficient, S: Standard error, VIF: Variation inflation factor. ** Predictors in proposed MNLR model were selected by following the best-subset procedure.
adjusted and predicted regression coefficients were the only way to explain the variation in response variable and also predict it, respectively. The predictors of proposed MNLR model were given in Table 2. As it is seen, adjusted and predicted regression coefficients were determined as 0.997 and 0.975, respectively. Although these statistical parameters demonstrated the powerful side of derived MNLR model significantly, they were not enough to state the proposed MNLR model totally. For this issue, some additional statistical parameters should be given for further explanation of proposed MNLR model. Durbin Watson (DW) factor and variation inflation factor (VIF) can be stated as indicators that showed the autocorrelation and multicollinearity, respectively. To meet the criteria that underlies the assumption in design of MNLR model, Durbin Watson and variation inflation factors must be in a range of 0 and 2, and 0 and 10, respectively (Buyukada and Evrendilek, 2016; Pan and Pandey, 2016; Buyukada, 2017). As it is seen in Table 2, DW and VIF values were estimated as 1.33 and lower than 10, respectively. This situation also showed that the proposed MNLR model met the assumption criteria, and it can be certainly
Frequency 0 1000 2000 3000
90
0
0 0
20000
40000 60000 Randomly generation
80000
100000
Fig. 3. Result of Monte Carlo simulations on cumulative distributive functions of explanatory variables of co-combustion process.
42.8
Fig. 4. Results of MLP predictions by (a) MNLR (b) MC simulation (c) MCMC simulation.
M. Buyukada / Bioresource Technology 232 (2017) 87–92
used in prediction of mass loss percentage of co-combustion of WH and lignite coal as response variable. Although the proposed MNLR model seemed to be competent, it also contained unexplained parts in itself that must be addressed. For this issue, an uncertainty estimation technique must be used. So, probabilistic uncertainty analysis based on MC simulations in the absence and presence of Bayesian approach (MCMC) as one of the main novel contributors of the present study has been on demand. 3.6. Monte Carlo simulations The necessity of this approach could be listed as follows: to identify the uncertainties in selected predictors, because these uncertainties could not be captured through by deterministic ways (1), and to demonstrate the importance of operating parameters on
91
MLP (2). To apply this approach, randomly generated predictors were used to run the proposed MNLR model. In another terms, the MNLR model was also chosen according prediction accuracy and stochastic predictors. This choosing criteria depended on the absence of any meaningful MNLR model (Table 3). Distributive functions for probabilistic changes in blend ratio, heating rate, and temperature were illustrated in Fig 3. The good agreement was found between mean (obtained experimentally) and Monte Carlo (obtained by simulations) values of mass loss percentage (Fig 3). 3.7. Markov Chain Monte Carlo simulations Fig. 4 showed the conditional predictions for each individual co-varieties that obtained by keeping all the other parameters
Fig. 5. Conditional predictors for (a) MC (b) MCMC simulations.
92
M. Buyukada / Bioresource Technology 232 (2017) 87–92
constant to their mean values. Mean values of proposed MNLR model, MC, and also MCMC simulations were determined as 54.5, 45.5, and 44.7, respectively (Fig. 4). The differences in these mean values demonstrated the uncertainties in derived MNLR model. As shown in Fig. 5, most of the temperature points were quite closely distributed in the whole range of the abscissa. The main lines in Fig. 5 showed the minimum and maximum levels of each explanatory variables while the dotted line between main lines (min and max) showed the ideal places (that any uncertainty should not be occupied under ideal conditions). The dots on the dotted line showed the experimental results and the dots around this line showed the deviation from ideality. Fig. 5 demonstrated the uncertainties in linear and quadratic effects of temperature in proposed MNLR model clearly. Quite close main lines at the beginning level of these explanatory variables indicated an uncertainty that should be identified. Differently, this situation was approximately not applicable for heating rate, and it was moderate applicable for blend ratio (Fig. 5). Therefore the uncertainties in the conditional predictions were smaller near the data-points. The same situation was not applicable for heating rate because it approximately covered the whole range of the abscissa uniformly and therefore it had similar levels of uncertainty predictions throughout. Even though the situation of blend ratio was not clear and it had a random distribution of the data points, it can be said that the uncertainties in the conditional predictions was less than temperature’s. Fig. 5 pointed out that conducting further experiments with lower temperature and wider range of blend ratio values was absolutely necessary. It was very important to improve the predictions. Heating rate was needed to be varied much in order to get more confident predictions. Thus, it was so clear and also necessary to identify the different experimental designs for cocombustion of WH and lignite coal could be conducted and the data points could be used to significantly improve the predictive power of the models. The findings about uncertainty estimation seemed to be in great accordance with the results of similar studies and this situation demonstrated the importance of uncertainty analyses on energy related processes (e.g. Buyukada, 2016, 2017; Evrendilek et al., 2016; Pan and Pandey, 2016; Zhang et al., 2016; Chaudhary and Hantush, 2017).
4. Conclusions Results of the thermogravimetric analyses showed that increasing WH amount in blend ratio caused a decrease in ash amount. When the position of Turkey in global walnut production were taken into consideration, the findings of present study may be useful for thermal plants. (MC-)MC simulations were applied to (co-) combustion of WH, lignite coal, and their various blends to incorporate the uncertainties in predictors of data-driven regression model. Results pointed out that deterministic approaches was inadequate to identify the uncertainties. Furthermore, Bayesian approach was found as fairly powerful technique to capture through the stochastic uncertainties in the predictors of MNLR model. Acknowledgement The author would like to acknowledge Dr. G.A. Evrendilek, the supervisor of YENIGIDAM Research Center (AIBU, Bolu, Turkey), for her kindly helps in laboratory analysis and E. Aydogmus from
Chemical Engineering Department of Firat University (FU, Elazig, Turkey) for him invaluable helps in providing lignite coals. References Atimtay, A.T., Kayahan, U., Unlu, A., Engin, B., Varol, M., Olgun, H., Atakul, H., 2017. Co-firing of pine chips in 750 kWth circulating fludized bed combustion system. Bioresour. Technol. 224, 601–610. Buyukada, M., 2016. Co-combustion of peanut hull and coal blends: Artificial neural networks modeling, particle swarm optimization and Monte Carlo Simulation. Bioresour. Technol. 216, 280–286. Buyukada, M., 2017. Probabilistic uncertainty analysis based on Monte Carlo simulations of co-combustion of hazelnut hull and coal blends: data driven modeling and response surface optimization. Bioresour. Technol. 225, 106–112. Buyukada, M., Evrendilek, F., 2016. Modeling efficiency of dehydrated sunflower seed cake as a novel biosorbent to remove a toxic azo dye. Chem. Eng. Commun. 203 (6), 746–757. Cepeliogullar, O., Putun, A.E., 2014. Products characterization study of a slow pyrolysis of biomass-plastic mixtures in a fixed-bed reactor. J. Anal. App. Pyrol. 110 (1), 363–374. Chaudhary, A., Hantush, M.M., 2017. Bayesian Monte Carlo and maximum likelihood approach for uncertainty estimation and risk management. Application to lake oxygen recovery model. Water Res. 108, 301–311. Chen, J., Liu, J., He, Y., Huang, L., Sun, S., Sun, J., Chang, K., Kuo, J., Huang, S., Ning, X., 2017. Investigation of co-combustion characteristics of sewage sludge and coffee grounds mixtures using thermogravimetric analysis coupled to artificial neural networks modeling. Bioresour. Technol. 225, 234–245. Evrendilek, G.A., Avsar, Y.K., Evrendilek, F., 2016. Modelling stochastic variability and uncertainty in aroma active compounds of PEF-treated peach nectar as a function of physical and sensory properties, and treatment time. Food Chem. 190, 634–642. Hu, S., Ma, X., Lin, Y., Yu, Z., Fang, S., 2015. Thermogravimetric analysis of the cocombustion of paper mill sludge and municipal solid waste. Energy Convers. Manage. 99, 112–118. Iegorov, R., Valiullin, T., Strizhak, P., 2017. Co-combustion of the coal-water slurry doped by combustible and non-combustible nano-particles. Appl. Therm. Energy 113, 1012–1023. Kayahan, U., Ozdogan, S., 2016. Oxygen enriched combustion and co-combustion of lignites and biomass in a 30 kWth circulating fluidized bed. Energy 116, 317– 328. Kurekci, N.A., 2016. Determination of optimum insulation thickness for building walls by using heating and cooling degree-day values of all Turkey’s provincial centers. Energy Build. 118, 197–213. Li, M., Wang, J., Yang, Y., Xie, G., 2016. Alkali-based pretreatments distinctively extract lignin and pectin for enhancing biomass saccharification by alterating cellulose features in sugar-rich Jerusalem artichoke stem. Bioresour. Technol. 208, 31–41. Mi, B., Liu, Z., Hu, W., Wei, P., Jiang, Z., Fei, B., 2016. Investigating pyrolysis and combustion characteristics of torrefied bamboo, torrefied wood and their blends. Bioresour. Technol. 209. 55-55. Myllari, F., Karjalanien, P., Taipale, R., Aalto, P., Hayrinen, A., Rautiainen, J., et al., 2017. Physical and chemical characteristics of flue-gas particles in a large pulverized fuel-fired power plant boiler during co-combustion of coal and wool pellets. Combust. Flame 176, 554–566. Ozer, D., Dursun, G., Ozer, A., 2007. Methylene blue adsorption from aqueous solution by dehydrated peanut hull. J. Hazard. Mater. 144 (1–2), 171–179. Ozkal, S.G., Yener, M.E., 2016. Supercritical carbon dioxide extraction of flaxseed oil: effect of extraction parameters and mass transfer modeling. J. Supercrit. Fluid 112, 76–80. Pan, I., Pandey, D.S., 2016. Incorporating uncertainty in data driven regression models of fluidized bed gasification: a Bayesian approach. Fuel Proces. Technol. 142, 305–314. Parvez, A.M., Wu, T., 2017. Characteristics and interactions between coal and carbonaceous wastes during co-combustion. J. Energy Inst. 90, 12–20. Tanyildizi, M.S., 2011. Modeling of adsorption isotherms and kinetics of reactive dye from aqueous solution by peanut hull. Chem. Eng. J. 168 (3), 1234–1240. Uzuner, S., Cekmecelioglu, D., 2015. Enhanced pectinase production by optimizing fermentation conditions of Bacillus subtilis growing on hazelnut shell hydrolyzate. J. Molec. Catal. B Enzym. 113, 62–67. Yao, Z., Ma, X., Wang, Z., Chen, L., 2017. Characteristics of co-combustion and kinetic study on hydrochar with oil shale: a thermogravimetric analysis. Appl. Therm. Energy 110, 1420–1427. Yildiz, Z., Uzun, H., Ceylan, S., Topcu, Y., 2016. Application of artificial neural networks to co-combustion hazelnut husk-lignite coal blends. Bioresour. Technol. 200, 42–47. Zhang, J., Li, Y., Huang, G., Chen, X., Bao, A., 2016. Assessment of parameter uncertainty in hydrological model using a Markov-Chain-Monte-Carlo-based multilevel-factorial-analysis method. J. Hyrdol. 538, 471–486.