• No results found

Artificial neural network modeling of methane emissions at Istanbul Kemerburgaz-Odayeri landfill site

N/A
N/A
Protected

Academic year: 2022

Share "Artificial neural network modeling of methane emissions at Istanbul Kemerburgaz-Odayeri landfill site"

Copied!
7
0
0

Loading.... (view fulltext now)

Full text

(1)

Artificial neural network modeling of methane emissions at Istanbul Kemerburgaz-Odayeri landfill site

H Kurtulus Ozcan1,*, Osman N Ucan2, Ulku Sahin1, Mehmet Borat1 and Cuma Bayat1

1Istanbul University, Engineering Faculty, Environmental Eng. Dept. 34 320, Avcilar, Istanbul, Turkey

2Istanbul University, Engineering Faculty, Electrical-Electronics Eng. Dept. 34 320, Avcilar, Istanbul, Turkey Received 29 April 2005; revised 13 October 2005; accepted 23 November 2005

This study reports on Istanbul Kemerburgaz-Odayeri (Turkey) solid waste landfills, models CH4, CO2, CO, atmospheric temperature parameters of this area, and predicts CH4 using Artificial Neural Networks (ANN). Here, ANN structure employs 4 input, 10 hidden and 1 output neurons. In order to evaluate performance of ANN model, statistical performance indices between real and estimated data have been measured and the correlation is found as 0.983 and 0.806 for training and testing respectively.

Keywords: Artificial neural network, Carbon dioxide, Istanbul, Landfill gas, Methane IPC Code: G06N3/02; C02F11/04

Introduction

Solid waste landfills are quite different from one another due to their heterogenic structures and one cannot find a single equation of decomposition rate or constant because there exist many decomposable matters in landfill1. Biological decomposition results in the landfill gases production. Initially, decomposition process is aerobic due to the existence of oxygen. When there is no oxygen left, anaerobic conditions arise and the organic components decompose after a chain of reactions2. Some factors that greatly affect the biological decomposition of solid wastes are, the nutrient ingredients of the waste, temperature, moisture, pH, particle dimensions, density and the composition of buried wastes1,3,4. In landfill gases (CH4, CO2, CO, H2S, N2, NH3, and O2), CH4 (60%) and CO2 (40%) are major gases, resulting from the anaerobic degradation of degradable domestic solid wastes2,5,6. Methane reduction should be a major objective in any mitigation strategy as emissions need only be reduced by 10-15 percent to stabilize the global atmospheric burden, while CO2

emissions would have to be reduced by 60-80 percent to achieve stabilization7. Landfills contributed a portion of the total increase in the atmospheric concentration of CH4 (1 % per year) for 1984-19928.

As a landfill gases, CH4 depicts explosive properties even at low concentration (5-15 %) in air. In O2

limited case, even if CH4 reaches these concentration values, the risk of explosion is highly decreased9.

Artificial Neural Network (ANN) is used in various engineering fields and, demonstrated remarkable success10. ANN models are computer programs that are designed to emulate human knowledge processing, speech, prediction, classification, and control11. ANN is a cellular information processing system designed and developed on the basis of the perceived notion of the human brain and its neural system12. In air pollution modeling, neural network (NN)-based models have been applied to predict various pollutant concentrations. Chelani et al13 constructed a three-layer NN model with a hidden recurrent layer to predict SO2 concentration at three sites in Delhi. In their study, a multivariate regression model was also used for comparison with the results obtained by using NN model. Sahin et al14 applied Multi-Layer Perceptron NN model to predict daily CO concentrations using meteorological variables as predictors for the European part of Istanbul, Turkey.

Viotti et al15 used ANN to forecast short and middle long-term concentration levels for Benzene, NOx, CO and ozone. Abdul-Wahab & Al-Alawi11 applied NN to predict ozone concentrations as a function of meteorological conditions and various air quality parameters. The results of their study indicate that the ANN is a promising method for air pollution

_________________

*Author for correspondence

Tel: +90 212 4737070/17726; Fax: +90 212 4737180 E-mail: hkozcan@istanbul.edu.tr

(2)

OZCAN et al: ANN MODELING OF CH4 EMISSIONS AT ISTANBUL LANDFILL SITE 129 modeling. Other ANN based researches in

environmental air pollution modeling can be found in Gardner & Dorling16 studies. This paper, proposes a multi-layered ANN structure for future prediction of CH4 in Istanbul Kemerburgaz-Odayeri (Turkey) landfill area that is the main landfill area of Istanbul.

Materials and Methods

Study Area

Solid wastes in European part of Istanbul are disposed in Kemerburgaz-Odayeri solid waste landfill area17,18 (Fig. 1). The LMSXi gas measurement device, produced by Gas Data, was used in CH4, CO2, and O2 measurements. This device determines CO2 and CH4 by infrared measurement technique (IR sensor), while O2 is processed with electrochemical sensors. Measurements were made twice a month at 7 different points of this area during July 2002-April 2003.

Artificial Neural Networks

In present ANN model (Fig. 2), input {i}, hidden {k} and output {1} neurons are used in parameter estimation of landfill area. In ANN, {x1, x2, x3, … , xi

} are input and y output parameters, whereas {w1(1,1)…

w1(i, k)} and {w2(1,1), ,….w2(k,1)} are input and output weight coefficients, respectively, which will be trained to find an optimum solution. The artificial model of a neuron consists of following three elements:

1 A set of synapses or connection links, each of which is characterized by a weight or strength of its own. Specially, a signal xj at the input of synapse j connected to neuron k is multiplied by the synaptic weight wkj. Unlike a synapse in the brain, the synaptic weight of an artificial neuron may lie in a range that includes negative as well as positive values.

2 An adder for summing the input signals, weighted by the respective synapses of the neuron.

3 An activation function or transfer functions for limiting the amplitude of the output of a neuron.

The neuron model can also include an externally applied bias, bk, which has effect of increasing or lowering the net input of the activation function depending on whether it is positive or negative, respectively. Mathematically, neuron k will be described by the following equations:

=

m

j

j j k

k w x

u , …(1)

where {x1,….,xm} are input signals, and {wk,1,….wk,m } are synaptic weights of neuron k. Activation function, f(net), defines output of a neuron which considerably influences behavior of the network,

k

k b

u

net= + …(2)

) (net f

yk = …(3)

where, bk is threshold value and f is activation function. Three basic types of activation function are generally used in ANN as follows:









>

>

=

2 1 2

1 2

1 2 1

0

` 1 ) (

v v v v

v

f …(4)

2 Threshold Function





<

= ≥

0 0 0

) 1

( v

v iv v if

f …(5)

3 Sigmoid Function

e av

v

f

= + 1 ) 1

( …(6)

In present model, NN is trained and tested using MATLAB 6.0. A 3-layer NN that consists of an input layer, output layer and one hidden layer is used (Fig. 2). The monitoring data (CH4, CO2, O2 and atmospheric temperature) belonging to 1-year period (July 2002 – April 2003), was designed to meet the requirements of training and testing the NN. This data base, in its original time series form, is divided into training and test sets taking the odd numbered patterns as training data and even numbered ones as test data. In training and testing set, y presents CH4 concentrations in vol % (Fig. 2). In simulations, various ANN models are tested changing the number of neurons in the hidden layer between 2 and 30. All the data are normalized into the range {–1.0, 1.0}.

This is carried out by determining the maximum and minimum values of each variable over the whole data period.

(3)

Fig.1 Location of Kemerburgaz-Odayeri solid waste landfill area in Istanbul and landfill site layout

(4)

OZCAN et al: ANN MODELING OF CH4 EMISSIONS AT ISTANBUL LANDFILL SITE 131

( )

(

max min

)

1.0

* min

2

=

X X

X

Xnorm X …(7)

Statistical Performance Indices

In this study, following five statistical performance indices were computed: i) Mean bias error (Bias); ii) Mean absolute error (MAE); iii) Root mean square error (RMSE); iv) Correlation coefficient (r); and v) Index of agreement (d). Bias is the degree of correspondence between mean prediction and mean observation. MAE is mean absolute value of the residuals. Lower numbers of Bias and MAE are the best, and values of bias < 0 indicate under- forecasting. The weighting of (P-O) by its square tends to inflate RMSE, particularly when extreme values are present. With respect to a good model, RMSE should approach zero. Evaluation can also be undertaken by considering measures of agreement, such as the Pearson product moment correlation coefficient (r) values. The index of agreement, bounded, relative measure that is capable to measure the degree of which predictions are error-free. The denominator accounts for the model’s deviation from the mean of the observations as well as to the observations deviation from their mean. In a good model, d and r should approach one. All these indices are formulated as follows:

=

=

N

1 i

i

i O )

N (P

Bias 1 …(8)

=

=

N

1 i

i

i O

N P

MAE 1 …(9)

=

=

N

1 i

2 i

i O )

N (P

RMSE 1 …(10)

( )

( )

=

=

= N

1 i

2 i N

1 i

2 i i

O O

P O 1

r …(11)

=

=

+

= N

1 i

2 i i

N

1 i

2 i i

) O O O P (

) O (P 1

d

…(12)

where, Oi and Pi are the observed and predicted pollution values, respectively. In i ={1, 2,., N} days, Ō is the mean of the observed times series and N is the total observation.

Results and Discussion

Minimum, medium and maximum values of parameters that used in model were measured (Table 1). A correlation was observed between {CH4, CO2, O2, Temperature (T)} values (Fig. 3) and optimum regression equations were calculated (Table 2).

Various combinations of ANN structures were investigated and optimum case was found as {4, 10, 1}

corresponding to 4 input, 10 hidden and 1 output neuron.

Here, learning rate was found to be 0.1 and Sum- Squared Error (SSE) was 0.19 (Fig. 4). Training stopped after training of (1.14 E+6) epochs. Differences between actual and estimated values were calculated using Eqs (8-12), and following results were found: Bias, 3.04;

MAE, 7.98; RMSE, 10.95; R, 0.81; and d, 0.88.

In training, the approximation of ANN is almost 100 % (Fig. 5). Using weight coefficients of the training set, extraordinary performance was found indicating that ANN can be a promising technique for parameter estimation of landfill areas.

Conclusions

This study estimated CH4 values of Istanbul Kemerburgaz-Odayeri (Turkey) landfill area using

Fig.2 Structure of the three layers ANN

Table 1 Minimum, medium and maximum values of parameters (July 2002-April 2003)

Parameter Minimum Medium Maximum

Methane (CH4), vol % 1.1 21.77 60 Carbon dioxide (CO2),

vol %

0.8 16 40

Oxygen (O2),vol % 0 12.62 20.2

Temperature (T), °C 4 16.7 33

Table 2 Regression equations of various parameters

Parameter Regression Equation (R2)

(CH4)(t+1) -(CO2)(t) (CH4)(t+1) = 0,0098[(CO2)(t)]2 + 1,1476(CO2) (t) + 1,5449

0,8674 (CH4)(t+1) -(O2)(t) (CH4)(t+1) = 0,0359[(O2)(t)]2 -

3,6974(O2)(t) + 63,865

0,8631 (CH4)(t+1) -(CH4)(t) (CH4)(t+1) = -0,0008[(CH4)(t)]2 +

1,0479(CH4)(t) + 1,097

0,897 (CH4)(t+1) - T (CH4)(t+1) = 26,047e-0,0264[T(t)]

0,063

(5)

ANN approach for the measured data using four parameters (CH4, CO2, CO and atmospheric temperature) as inputs during one year. The ratio concentration of CH4 gas is altering (1-60 %), thus it is a complex prediction problem. Air pollutants have

been modeled in earlier ANN studies. Spelmann 199919 applied NN model to predict ozone concentrations for London, Harwell and Birmingham cities. In his study, correlation coefficients (r) were found 0.77, 0.72 and 0.53 respectively. Gariazzo &

Fig.3 Correlation curves between input parameters and estimated CH4 value of ANN

Fig. 4 Sum-Squared Error and Learning Rate of ANN model

(6)

OZCAN et al: ANN MODELING OF CH4 EMISSIONS AT ISTANBUL LANDFILL SITE 133

Tirabassi 200320 also modeled air pollutants using ANN. Present ANN model results have better performance (r=0.90) when compared with earlier studies10,14,16,19,20. As a result, ANN based approaches can be considered as a compromising approach in landfill gas prediction problem.

Acknowledgements

Authors gratefully thank ISTAC (Istanbul Metropolitan Municipality Environment Protection and Waste Material Recycling Industry and Trade J.S.Co.) for technical support.

References

1 McBean E A, Rovers F A & Farquhar G J, Solid Waste Landfill Engineering and Design (Prentice Hall PTR, New Jersey) 1995, 380.

2 Tchobanoglous G, Theisen H & Vigil S, Integrated Solid Waste Management (McGraw-Hill Inc) 1993, 392.

3 El-Fadel M & Massould M, Emissions from landfills, a methodology comparative assessment, Environ Technol, 21 (2000) 965-978.

4 El-Fadel M, Findikakis A N & Leckie J O, Environmental impacts of solid waste landfilling, J Environ Manage, 50 (1997) 1-25.

5 Allen R M, Braithwaite A & Hills C, Trace organic compounds in landfill gas at seven U.K. waste disposal sites, Environ Sci Technol, 31 (1997) 1054-1061.

6 Demir G, Ozcan H K, Nemlioglu S, Sezgin N, Borat M & Bayat C, Gas emissions of Yakacik closed solid waste landfill site in Istanbul, Fresen Environ Bull, 13 (2004) 974-978.

7 Blaha D, Barlett K, Czepiel P, Harriss R & Crill P, Natural and anthropogenic methane Sources in New England, Atmos Environ, 33 (1999) 243-255.

8 Klusman W R & Dick J C, Seasonal variability in CH4

Emissions from a Landfill in a cool semiarid climate, J Air &

Waste Manage Assoc,50 (2000) 1632-1636.

9 Mosher W B, Czepiel M P, Haris C R, Shorter H J, Kolb E C, McManus B J, Allwine E & Lamb K B, Methane emissions at nine landfill sites in the northeastern United States, Environ Sci Technol, 33 (1999) 2088-2094.

Fig.5 Comparison of ANN outputs and observed values

(7)

10 Elkamel A, Abdul-Wahab S A, Bouhamra W & Alper E, Measurement and prediction of ozone levels around a heavily industrialized area: A neural network approach, Adv Environ Res, 5 (2001) 47-59.

11 Wahab-Abdul S A & Al-Alawi S M, Assesment and prediction of trospheric ozone concentration levels using artifical neural networks, Environ Model Softw, 17 (2002) 219-228.

12 Huang Y F, Huang G H, Dong M Z & Feng G M, Development of an artifical neural network model for predicting minimum miscibility pressure in CO2 flooding, J Petrol Sc Eng, 37 (2003) 83-95.

13 Chelani A B, Chalapati R C V, Phadke K M & Hasan M Z, Prediction of sulphur dioxide concentration using artificial neural networks, Environ Model Softw, 17 (2002) 161-168.

14 Sahin U, Ucan O N, Soyhan B & Bayat C, Modeling of CO distribution in Istanbul using artificial neural networks, Fresen Environ Bull, 13 (2004) 839-845.

15 Viotti P & Genova P D, Atmospheric urban pollution:

applications of artificial neural network (ANN) to the city of Perugia, Ecol Model, 148 (2002) 27-46.

16 Gardner M W & Dorling S R, Artificial neural networks (The Multilayer Perceptron) – A review of applications in the atmospheric sciences, Atmos Environ, 32 (1998) 2627-2636.

17 Borat M, Nemlioğlu S, Demir G, Sezgin N & Bayat C, Comparison of landfill gas emissions of some closed and active solid waste landfill sites in Istanbul, 1st National Environmental Problems Symp, (Eds. Demircioglu N, Ocak S & Kocadagistan E) (in Turkish), 2002, 460-468.

18 Demir A, Ozkaya B & Bilgili S B, Effect of leachate recirculation on methane production and storage capacity in landfill, Fresen Environ Bull, 12 (2003) 29-38.

19 Spellman G, An application of artificial neural networks to the prediction of surface ozone concentrations in the United Kingdom” Appl Geography, 19 (1999) 123-136.

20 Tirabassi T, Pelliccioni A & Garazzo C, Coupling of neural network and dispersion models: A novel methodology for air pollution models, J Environ Pollut, 20 (2003) 136-146.

References

Related documents

Feature-based parametric Solid Modeling system represents the recent advance of computer geometric modeling. It is used as the foundation of solid modeling software's like

The total area of landfill is 7 acres. Compound wall for the whole area is proposed excluding the fence of adjoining the estate. A drain is proposed to collect and drain the

lt is stated that Municipal Committee Bandipora is continuously and intermittently dumping Municipal Solid Waste in the Wullar wetland at Nussu Zalwan area in violation

The garbage collected is being properly disposed and dumped at the SWT Plant in accordance with Solid waste Management Rules, 2016 and,.. without breaching

Traditional statistical regression and Artificial Neural Network (ANN) modeling techniques were applied toassess the emission characteristics of CNG vehicles (cars and three

This Chapter has provided the groundwork for prediction of the breakdown voltage of five insulating materials namely White Minilex, Leatheroid Paper, Glass Cloth,

There are several methods for disposal of municipal and industrial plastic waste as shown in figure 2 landfill, incineration (energy recovery), true material

Figure 4.10 Variation in Carbon Nitrogen Ratio with respect to height 80 Figure 4.11 Topography of C&amp;D waste (At 15 th Day) 81 Figure 4.12 Topography of C&amp;D waste (At