CNG exhausts emission modeling: Neural network approach

CNG exhausts emission modeling: Neural network approach

Kirti Bhandari>, Ch Ravi Sekhar, AM Rao and S Gangopadhyay

Transport Planning and Environment Division, Central Road Research Institute, New Delhi 110 020 Received 28February 2005; revised 26lillie 2006; accepted 26July 2006

Journal of Scientific &Industrial Research Vol.65, December 2006, pp, 1000-1007

,I ,'~ .

, i

Traditional statistical regression and Artificial Neural Network (ANN) modeling techniques were applied toassess the emission characteristics of CNG vehicles (cars and three wheelers) tounderstand the influence of explanatory parameters , like vehicle age, vehicle type and air-fuel ratio onemissions of CO, HC, CO2 and

°

ANN model. ANN models are shown better predictive models than the traditional statistical modeling techniques for"

predicting the CNG exhaust emissions (CO. He.CO2, and O2),

Keywords: ANN, CNG, Exhaust emissions, Regression

Introduction

Compressed natural gas (CNG) has emerged as a solution to the depleting crude oil resources as well as to deteriorating urban air quality. CNG (85-99%

methane) has many desirable properties as .a spark ignition engine fuel. Clean burning, cheap and abundant in many parts of the world, natural gas has already played a significant role in Argentina, Canada, Italy, New Zealand and the US. Natural Gas Vehicles (NGV's) have a marked advantage' over conventional diesels in terms of less pollution

(exhaust emission) as follows: carbon monoxide

(CO), 84; nitrogen oxides (NOx), 58; and particulate mater (PM), 97 %. In another study", emission characteristics of in-use CNG vehicles in Delhi

showed notable reduction in CO and hydrocarbons

(BC) emissions from CNG vehicles especially from three wheelers.

Chris Brace.' developed and used successfully three applications of Artificial Neural Network (ANN) prediction of diesel engine exhaust emission.

Shivanagendra & Khare⁴ studied ANN based line source models for vehicular exhaust emission prediction of urban roadways. Dougherty studied identification of driver behavior with advanced, traveler in'f~rmation system. Jankowska6 developed models for NOx 'and C02 emissions for pulverized fuel and circulating fluidized bed combustion boilers.

Present study con-elates different explanatory

"Author for correspondence

E-mail: kirti@urban.env.nagoya-u.ac.jp

parameters such as vehicle age, vehicle type and air"

fuel ratio on emissions of CO, He, CO

2;

Materials and Methods

Study Area and Type ofVehicles

In-use CNG vehicles consisting of three wheelers and cars in Delhi were. tested for their exhaust emission characteristics in April 2004 by using A VL 4000 light Digas exhaust gas analyzer. The tailpipe emissions of CO, HC, CO2, O2 along with lambda (air-fuel ratio) were measured under idle conditions.

Category of vehicles, registration number and year of registration were also recorded. Exhaust emission data of cars and three wheelers has been used for application of traditional regression and ANN modeling.

Development of Neural Network (NN) Model

A multilayer feed-forward (Fig. 1) NN model with single hidden . layer was considered and, back propagation algorithm was selected for training the network7-9• Back propagation algorithm is systematic approach for training multilayer NN. Using. this algorithm, two propagation phases, forward .and backward, are required. Forward pass calculates network output by propagating the input data through.

the network. The network output is then compared with the desired output to calculate the error using a backward pass; during the backward pass connection weights are modified to reduce the target en-or.

BHANDARI et01: CNG EXHAUSTS EMISSION MODELING: NEURAL NETWORK APPROACH 1001

Input Layer Hidden uy.:r(j) OutputLayer

(i) (k)

lnoutParameters

w, WiL

Network Weichrs

Fig. I-Multilayer feed-forward neural network model

configuration \.

Back Propagation Algorithm

Back propagation algorithm used In the present study is as follows:

Step I

Decide network topology and randomize the initial weights.

Step 2Forward Pass

i) Compute the hidden layer neuron activation (OH,,) ... (1)

where F( . ) is the Sigmoid Transfer function F(a)

=1/1+ exp -a, Wji=weights from node i(source) to node j(destination), lpi=input value of node i and pattern p.

ii) Compute output layer neuron activation (00",) 001"

=

= L

where F( . ) is the Sigmoid Transfer function F( a)

=1/1 + exp -a, W'i= weights from node j(source) to node k(destination).

iii) Compute the error for each pattern at each output node E"

... (3)

where Tpk=target output value of node k for pattern p, OOpk= actual output value of node for pattern p.

Step 3Backward Pass

i) Computation of the error signal (D",) at output layer and adjustment of weights between output to hidden nodes

Table I-Training andtestdata sets used forANN models SINo Name ofthe Number of patterns

model Training phase Testing phase

Autos Cars Autos Cars

I CO 295 288 98 51

2 HC 288 269 96 48

3 CO?and 0, 296 287 98 51

D",

Wkj(New) =Wkj (old) + 1] ~)kOHpk+ cx[Wkiold)-

Wki o/d-l) ... (5)

ii)Computation of error signal (D"j) at hidden layer and adjustment of weights between hidden and input nodes.

~)j

=

L

III ···0)

Step 4

All the above steps are repeated with new training patterns. One iteration is over when all training patterns are finished. These iterations continue till the average Mean Squared Error (MSE) with in the tolerable limit.

Neural Network Training

Network training represents acqumng knowledge of predicting emission characteristics of in-use CNG vehicles (three wheelers and cars). After preprocessi ng, datasets (75%) were considered for network training and remaining (25%) were considered for validation for each individual model.

The number of patterns used for network training and validation of each individual model are presented at Table I. For ANN model development, neural solution software'? has been used. This allows the user to select an ANN paradigm with architecture and identify training and test data sets. Training W,,5 stopped when the average MSE reached previously specified minimum value (0.00l). After several trials, optimum topology for each model was found out (Table 2).

Neural Network Testing

Validation process is conducted to verify the network reliability. It is achieved by processing testing sets to the network. These sets should be new

HC model

1002 J SCI IND RES VOL 65 DECEMBER 2006

Parameters CO model

Table 2-Optimum network topology considered for various ANN models

CO?&07model

Auto Car

I.Optimum network topology Number ofinput neurons Number ofhidden neurons Number of output neurons II. Optimum network training Momentum Factor

Number of iterations Ill. Input data types All independent variables IV. Output data types Dependent variables

2 3 I

2 3 1

0.95 10000

0.95 10000

CO

Auto Car Auto Car

2 6 I

2 6 1

2 6 2

2 6 2

0.95 10000

0.95 10000

0.95 10000

0.95 10000

Vehicle age and air fuel ratio(A/F)

HC Table 3--Performance of ANN model vs regression models

Vehicle type Model

Regression co- efficient (R) during ANN model

training phase

Regression coefficienuk) during regression

modeling

Car CO

HC CO2

O2

CO HC CO2

07 Three wheeler

that the network has never been exposed before. For this, 25% of data sets were considere..l. Network output using these sets is compared with t.ie desired output to calculate accuracy rate. If the accuracy rate is low, it means that the network is not properly trained or the hidden nodes considered in the network topology are inadequate. Otherwise the network is considered to be reliable and ready for implementation.

Results and Discussion

Total data sets are randomly separated in training and testing data sets (Table 1).Training data sets were used for ANN model development. After getting required MSE, training phase of NN model is stopped and corresponding synaptic weights between input to hidden and hidden to output layers are used for the validation of NN model, for this testing data has been used. These data sets are unseen to the NN model.

Traditional regression analysis also has been carried out for the same data adopted for ANN model development. Thus ANN models are better predictive

Regression co- efficient (R)during ANN model

testing phase 0.68

0.20 0.67 0.96 0.69 0.12 0.83 0.95

0.71 0.31 0.57 0.(' 0.65 0.12 0.82 0.93

0.40 0.13 0.63 0.89 0.42 0.12 0.80 0.88

model than regression models in all the cases, except for CO² for cars (Table 3).

Significant correlation was observed by considering CO-ANN model for cars as shown by regression coefficient (R) value of 0.71 as compared to the value of 0.65 in case of three wheelers (Fig. 2).

But CO- regression models failed to correlate the significance between input and output variables. This implies that the explanatory variable such as vehicle age and air-fuel ratios (lambda) have significant correlations with the CO emissions.

Tn the case of HC model, both ANN and regression models are unable to correlate between explanatory variable and the prediction of He. This is substantiated by the insignificant R-values of 0.31 in case of cars and 0.12 in case of three wheelers (Fig. 3). This implies that the explanatory variable such as vehicle age and air-fuel ratios (lambda) do not have any significant correlations with the HC emissions.

Incase of CO² (cars), ANN and regression models explains that the explanatory variable such as vehicle

BHANDARI et al:CNG EXHAUSTS EMISSION MODELING: NEURAL NETWORK APPROACH 1003

1~---~---.

... CO Output_ANN

0.8 CO Output_Regression

(a)

. . _..

0.6

"

"

~OA

.•..

00.2

_... .

." .

"

.,.

:

L· .. JiLH" - '. ..

' ~:':: ': ...,

''''"'''1''--=':''' . , .

10 19

28

82

_OAL---...I Exemplar

08

(b) 06·

.\ ..

04

.•.•

::lQ,

.•..

0 21 45 49

-0. ^t'

I )'

-0.4-

-O.J

... CO Output ANN

- - -CO Output Reqression

Exemplar

Fig. 2--Observed andpredicted CO emission for (a) cars (b)three wheelers

1004 ^{J SCI}IND RES VOL 65DECEMBER 2006

(a)

1

r---,

0.9

0.8

--He

... HeOutput_ANN HeOutput_Regression

-

.s-

o

~~~'Vvt~vvtvJJ 'u'bk;'V'V'r ¹

O~I I I I I I I I I I

I

1 10 19 28 37 46 55 64 73 82 91

Exemplar

0.8~---~--~---.

(b) -He

... ·HeOutput_ANN HeOutput_Regression 0.7

0.6 0.5

0J...c...- __

1 _{5 9 13} 17 21

25 29

Exemplar

Fig. 3--Observed and predicted HCemission for(a) cars (b)three wheelers

BHANDARI et a/: CNG EXHAUSTS EMISSION MODELING: NEURAL NETWORK APPROACH 1005

(a) --C02

,nn", CO2Output_ANN

--C02 Output_Regression O.

0.8 0.1'

0.6

-

-

0

0.1

0~ __-+~ __ ~ __ ~~ __ ~~ __ ~ __-7~ __~ __~~ __~ __~~~

1 10 19 28 37 46 55 64 73 82 91

Exemplar

(b)

1.2~---.

--C0²

... 'C02 Output_ANN --C020utput_Regreesion 0.2

oj----+_--~---+--~~I~r---+---+---~---+---+--~----~

1 5 9 13

17 21 25 29

Exemplar

Fig. 4--Observed and predicted CO2 emission for (a) cars (b)three wheelers

1006 ^{JSCI IND RES VOL 6S DECEMBER 2006}

1.2.---.

1

(a)

0.4

0.2 --02

-- - - -_.O2 Output_ANN

-- O2OutputRegression O+----+----~--~----~----~---r----+----+----~----~~

1 ₁₀

19

37

55

73

91

Exemplar

0.9~---.

0.8 0.7 0.6

--0²

-_... O2 Output_ANN -- O2 Output_Regression (b)

.. .

_ 0.5

::lQ.

'5

o

0.3 0.2 0.1

O~--~---+----~--+-_v-~--~~~~--4_--_+----~--~--_+~

1 ₅ 9

13 17 21 25 29 33 37 41 45 49

Exemplar

Fig. 5--Observed andpredicted O2emission for (a) cars (b) three wheelers

BHANDARI et a/:CNG EXHAUSTS EMISSION MODELING: NEURAL NETWORK APPROACH

age and air-fuel ratios (lambda) do not have any significant correlations with CO² emissions. But in the case of CO² (three wheelers), these models identified a significant relation between input and output variables. Inboth the cases, ANN models R values are slightly higher than regression models (Fig. 4).

In prediction of O2 emission, regression models are significant as indicated by the significant R-values of 0.88 in case of cars and three wheelers (Fig. 5).

Compared to this, R-values as given by CO-ANN models are 0.96 in case of cars and 0.93 in case of three wheelers. This indicates that ANN models are highly capable of correlating between input and output variable than regression models.

ANN models are observed more suitable for capturing the highest and lowest values of CNG emissions whereas regression models are unable to do this.

Conclusions

ANN models are better models than the traditional statistical modeling techniques (regression techniques) for predicting CNG exhaust emissions (CO, HC, CO^2, and O^2), Although ANN modeling has been able to predict CO, CO2 and O2 emissions, it has not been able to show any significant correlations for HC emissions(R=0.3 for cars and 0.12 for three wheelers) for the adopted explanatory variables of vehicle age and air-fuel ratio. R values, as given by regression modeling, are also very low (R=0.12 in both vehicle categories) indicating insignificant correlation. The success of NN implementation is dependent not just on the quality of data used for training but also on the topology/structure of the NN adopted.

1007

Acknowledgements

Authors thank Director, CRRI, for permission to publish this paper. Authors also thank Prof P K Sikdar, former Director, CRRI, for his valuable suggestions and guidance.

References

1 Mike F, Noton P, Clark N N & Lyons D W, Evaluation of natural gas versus diesel in medium-duty buses, SAE Technical Paper Series 2000-01-2822, Warendale.

Pennsylvania.

2 Singh A, Sharma N, Sharma K & Bhan C, Emission characteristics of in-use CNG vehicles in Delhi, in Workshop on Land use, Transport and Environment (CIRT Pune in collaboration with Harvard University) 3-4 Dec 2001, 26pp a vai lable at http://www.deas.harvard.edulTransportAsia- /workshop-papers

3 Brace C, Prediction of diesel engine exhaust emission using artificial neural networks, in Neural Networks ^ill System Design (Lucas Electrical and Electronic System, Solihull) 10 J unc (1998), 1-I 1,http://www.bath.ac.uk/-enscjb-/s591.pdf

4 Shiva Nagendra S M& Khare M, Artificial neural network based line source models for vehicular exhaust emission predictions of an urban roadway, Tranportation Res Part D, 9(2004) 199-208.

5 Dougherty M, A review of neural networks applied to transport, Transportation Res Part C, 3(1995),247-260.

6 Jankowska A, Neural models of air pollutants emissions in power units combuston process in AI-METH 2003- Artificial Intelligence Methods, (Gliwice, Poland) 5-7 November (2003),141-144

7 Ravisekhar C H, Master dissertation mode choice analysis using neural network, Department of Civil Engineering, University of Roorkee, Roorkee, India, 1999.

8 Hegazy T. F::z;n P&Moselhi 0, Developing practical neural network applications using back-propogation, J Micro Computers illCivil Engg, Vol. 9 (2), (1994) 145-159.

9 Jain A K & Mao 1, Artificial neural networks: A tutorial,

lEES COIliPUMag, (1996) 31-44.

10 Neural solution software. developed by Neuro Dimension;

www.nd.com