Más contenido relacionado
La actualidad más candente (20)
Similar a Artificial neural network based nitrogen oxides emission prediction and (20)
Más de IAEME Publication (20)
Artificial neural network based nitrogen oxides emission prediction and
- 1. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
491
ARTIFICIAL NEURAL NETWORK BASED NITROGEN OXIDES
EMISSION PREDICTION AND OPTIMIZATION IN THERMAL
POWER PLANT
Preeti Manke1
, Sharad Tembhurne2
1
(Computer Science & Engineering, Institute of Technology, Korba, India)
2
(Operation & Maintenance, National Thermal Power Corporation Limited, Korba, India)
ABSTRACT
Currently, emission monitoring is done via analytical instruments which are very
expensive to install and maintain. The power generating industry is undergoing an
unprecedented reform. This paper describes an efficient approach to predict nitrogen oxides
emission from a 500 MW coal fired thermal power plant with optimized combustion
parameters. The oxygen concentration in flue gas, coal properties, coal flow, boiler load, air
distribution scheme, flue gas outlet temperature and nozzle tilt were studied. Artificial neural
networks (ANNs) have been used in a broad range of applications including: pattern
recognition, optimization, prediction and automatic control. The parametric field experiment
data were used to build artificial neural network (ANN). The coal combustion parameters
were used as inputs and nitrogen oxides as output of the model. The predicted values of the
model for full load condition were verified with the actual values. The predicted values are
94% closer to actual operating values with lowest Root mean square error (RMSE) 0.4908
and highest correlation factor 0.9627. The optimum level of input operating conditions for
low nitrogen oxides emission was determined by simulated annealing (SA) approach. These
optimum operating parameters can help us to ensure complete combustion, less emission with
increased boiler life.
Keywords: Boiler, Neural network, NOx emission, Thermal Power plant
I INTRODUCTION
Power plants, chemical plants, government utilities and petroleum refineries typically
produce output more than the plant’s capacity in order to meet demands. Coal is the major
INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING
& TECHNOLOGY (IJCET)
ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online)
Volume 4, Issue 3, May-June (2013), pp. 491-502
© IAEME: www.iaeme.com/ijcet.asp
Journal Impact Factor (2013): 6.1302 (Calculated by GISI)
www.jifactor.com
IJCET
© I A E M E
- 2. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
492
fossil fuel in India and continues to play a pivotal role in the energy sector. Coal meets about
61% of the commercial energy needs and about 72% of the electricity produced in India
comes from coal [1]. The coal combustion process produces various pollutants, such as
oxides of carbon (COx), oxides of sulphur (SOx), oxides of nitrogen (NOx) and particulates.
The acid rain and climate change are mainly due to pollutants like SO2, NOx and CO2 [2].
Climate change induced by these gases can cause damage to human health, agriculture,
natural ecosystems, coastal areas and other climate sensitive systems. Hence these industries
must continuously monitor their exhaust stacks for primary pollutants under regulations
promulgated under the New Source Performance Standards (U.S. Environmental Protection
Agency (EPA)) or the Clean Air Act Amendments [3].
Most of the fly-ash particulates can be removed by fitting electrostatic precipitators
and over 85 per cent of SO2 with the installation of a flue gas desulphurization plant. The best
way to reduce CO2 emissions is to improve power generation efficiency. However, no
practical methods exist for reducing NOx to such a degree, leading to increased research into
this area. During the combustion process in a coal-fired power plant, nitrogen from the coal
and air is converted into nitric oxide (NO) and nitrogen dioxide (NO2); together these oxides
of nitrogen are commonly referred to as NOx. The methods for reducing NOx emissions in
coal-fired power plants can be classified as either primary or combustion modification based
technologies [4], which achieve reduction of NOx formation by limiting the flame
temperature through control of oxygen in the flame. Secondary technologies are flue gas
treatment. NOx reduction by adding a reagent such as ammonia or urea into the flue gas is
classified as a secondary technique. Installation of low-NOx burners and implementation of
advanced boiler operation and control systems for NOx emission reduction would normally
be classified as combustion modification technologies. Although low-NOx burners are
usually sufficient to achieve the required target under current legislation, it is often at the
expense of other important operational parameters such as incomplete combustion, steam
temperature and boiler performance. This is one of the reasons why advanced operational and
control systems in coal-fired boilers is so important [5]. Apart from the physical modification
of boiler, researchers have tried to model and optimize boiler parameters for minimal
emissions [6].
Several works have been done to develop predictive systems for industrial emissions.
One of the earlier ideas was presented by S.S.S Chakravarthy, A.K Vohra and B.S Gill [8]
has developed a predictive emission monitors for industrial process heaters. The authors have
used heuristic optimizer genetic algorithm (GA) to tune the NOx kinetic parameters. L.
Zheng, S. Yu, M. Yu [9] used generalized regression neural network (GRNN) to establish a
non-linear model between the parameters of the boiler of 300MW steam capacity and the
NOx emissions. Researchers studied the non-linear problem for decades and many traditional
and meta-heuristic techniques including artificial intelligence methods have been developed
[11]. A machine-learning method for non-statistical model building, such as artificial neural
networks (ANN), can be improved to attain the desired accuracy level by training it on
experimental data [12]. In this work, the parametric field experiments to obtain the
relationship between the operating parameters and NOx emission concentration in flue gas are
introduced. The ability of ANN to model the NOx pulverized coal combustion characteristics
of a 500 MW thermal power plant under full load condition is demonstrated.
However, these models tend to be really comprehensive to build and they are time
consuming and also the appropriate parameters could not be determined immediately for
changed operating conditions [10]. Recently meta-heuristic approaches such as simulated
- 3. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
493
annealing (SA), tabu search, genetic algorithms (GA), and evolutionary programming have
been used to develop advanced on-line and real-time combustion optimization software
package in modern power plants. This article emphasizes the effectiveness of Simulating
Annealing (SA) combined with artificial neural networks (ANN) to determine the optimal
combustion parameters for minimum NOx emission. SA is employed to perform a search to
optimize the input space of the neural network model to determine the low NOx emissions
level to meet the requirement. Neural network modelling and Simulated Annealing described
in this study are implemented in Matlab 6.5.0 (MathWorks, Inc.) and run under the Microsoft
Windows 7 environment.
This paper is organized as follows: Section 2 presents a brief literature on artificial
neural network and Backpropagation neural network (BPNN). Section 3 describes the
modeling background in terms of the parameters used, case study of coal fired 500MW
power plant and the dataset used. Section 4 presents the results and analysis obtained from
the combined approach of ANN with SA for full load condition and Section 5 gives a brief
conclusion reached in this study.
II ARTIFICIAL NEURAL NETWORK (ANN)
An ANN is represented as a non-linear interconnected layer of processing nodes
which are normally referred to as neurons, a term borrowed from neurobiology. It is an
information processing paradigm made up of a set of algebraic equations. The most common
class of ANN is the multilayered feedforward network which primarily consists of input
layer, one or more hidden layer and an output layer [6]. Through the input layer, the
normalized raw data is fed into the network. The input layer holds the data and distributes
them into the network via interconnections to neurons in the hidden layer(s) where they are
processed by the activation function to obtain the output signal. Activity of each unit of the
hidden layer is determined by the activities of the input units and the weights on the
connections between the input and the hidden units as in Figure 1.
Fig. 1 ANN architecture for NOx prediction
Input
X1 X2 ..Xn
Hidden Layers
Output
Y
- 4. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
494
Likewise, the behaviour of the output units depends on the activity of the hidden units
and the weights between the hidden and output units. The inputs of each neuron in the hidden
and the output layers are summed and the result is processed by an input–output function
(transfer function). This type of network is also known as multilayer perceptrons (MLP) [14];
a typical MLP network is shown in
(1)
(2)
(3)
Where L, is the element activation function and is given below.
(4)
The variable wi is the weight term and xi is the network input. The network uses these
weights to identify the strength of the interconnections between neurons. These weights are
adjusted throughout the learning process. The behavior of an ANN depends on both the
weights and the transfer function specified for the units. For MLP network, learning
algorithms based on the gradient or Jacobian of the network error with respect to the weights
is preferred because of their superior performance. A common Jacobian-based algorithm is
the backpropagation algorithm. The performances of the models are gauged using standard
performance functions in the form of correlation factor (R) and root means square error
(RMSE).
(5)
Where Xi is the predicted value, Xiˆ is the true value and n is the number of testing samples.
2.1 Backpropagation Neural Network (BPNN)
Using the ANN as its fitness function, the optimization algorithm determines the
optimum levels of coal combustion parameters for minimum NOx emission. ANN, in
combination with optimization algorithm, has been used successfully in various studies to
solve a variety of optimization problems, including problems where the objective function is
discontinuous, non-differentiable, stochastic, or highly non-linear [15,16]. Here, the same
combination is applied for NOx emission optimization. The backpropagation neural network
however has been widely used to develop soft sensors for prediction of NOx [9]. The main
- 5. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
495
advantage of ANN is the ability to model a problem by the use of data associated with
process, rather than analysis of process by some standard numerical methods.
Emission Target
Emission Target
Measured Air flow Optimized Parameter
Feedback
Fig. 2 Block diagram for Boiler with BPNN
In this model the system's output error was propagated back through the plant using
its partial derivatives at an operating point as shown in Figure 2.
However, BPNN has some weaknesses, including the need for numerous controlling
parameters, difficulty in obtaining a stable solution and the danger of overfitting. The
solution shown by Zheng et. al. [9] points to the fact that BPNN is unreliable even if all of the
network objects are pre-determined. The model is developed in this paper using distinctive
algorithms as a benchmark using BPNN. The optimal network parameters are chosen by
varying the number of layers and number of hidden neurons per layer. The parameter that
gives the best performance is chosen and shown below:
• Number of layers = 2
• No of neurons (hidden layer) = 20
• Transfer functions (input layer) = tan-sigmoid
• Transfer function (hidden layer) = linear
• Training algorithm = Lavenberg-Marquat
III CASE STUDY
The experiments are carried out in a 500 MW tangentially fired dry bottom boiler
with a large furnace. The tilting fuel and combustion air nozzles including nine primary air
burners and ten secondary air burners are located in each corner of the furnace. All nozzles
can be tilted in vertical direction over about 30 degree from the horizontal axis, both upwards
and downwards. The burners on A, B, C, D, E, F, G, H, J levels were put into operation under
the rated load. The coal pulverisers are employed to supply the coal–air mixture to the
burners on the corresponding levels. The tangential firing system is employed to combust
bituminous coal. The arrangement of the burners is illustrated in Figure 3.
Boiler PlantController BPNN
Updated
- 6. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
496
Fig. 3 Arrangement of burners
Tangential firing helps in keeping the temperature of the furnace low so that NOx
emission is reduced considerably. In addition to the above the over fire air is provided which is
used as combustion process adjustment technically for keeping the furnace temperature low and
thereby low NOx formation. Each corner of the burner wind box is provided with two numbers of
separate over fire air compartments, kept one above the other and the over fire air is admitted
tangentially into the furnace. The over fire air nozzles has got tilting arrangement and
compartment flow control dampers for working in unison with the tilting tangential type burner
system for effective control of NOx formation. The primary air system delivers air to the mills for
coal drying and transportation of coal powder to furnace. The 500 MW units have two stage axial
flow primary fans. The typical variation in air flow rate and economizer flue gas outlet
temperature for 500MW boiler is shown in Figure 4 and Figure 5 respectively.
In total, 55 tests have been performed on this boiler, changing the boiler load, primary air,
secondary air distribution pattern, nozzles tilting angle, respectively, to analyze the characteristics
of the NOx emission of the tangentially fired system. Out of which, 15 test data pertaining to full
load condition, are used for this present study.
Fig. 4 Total Air Flow (TPH) Fig. 5 Variation Economizer Outlet Flue
Gas Temperature
1560
1760
0 20 40 60 80
Airflow(TPH)
Number of reading
370
375
380
385
390
395
400
0 20 40 60 80
Fluegastemperature(degree)
Number of reading
- 7. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
497
3.1 The NOx Emission Characteristics and Optimization
During all the experiments, the fineness of the coal is kept constant. NOx and O2
concentrations are monitored continuously in the boiler outlet prior to the air heater. Fly ash
samples are withdrawn from the flue gas by a constant rate sampling probe. The NOx
concentrations reported in this work are average values over several hours of stable operation,
and they are obtained under dry gas conditions. Selected 15 sets of test data under full
condition is provided in Table 1 and Table 2. The measured NOx emissions for full load
condition are summarized in Table 3.
Table 1 The important boiler operating parameters
S.
No
.
LOA
D
FG Temp at
ECO.
O/L(deg)
Total
Air
flow O2(%)
Windbox
Pr(mmWC
L)
Furnace
Windbox
DP(mmWC
L)
Coal
Flow
Feedrate of Mills in service(T/Hr)
(MW) L R (T/Hr) L R L R L R
(T/H
r) A B C D E F G H J
1 517 387 389 1660 2.5 3.37 50 45 50 50 342 63
60
58 56 57 56
53 53
44
2 518 382 370 1650 1.7 2.8 50 40 40 50 349 63 62
56
59 60 54 51
54
48
3 490 381 379 1680 2.61 3.5 40 30 35 40 356 56 56 49 54 47 47 50 51
46
4 510 374 364 1700 2.26 2.88 45 40 45 50 405 65 65 58
55
60 58 46 50
43
5 506 433 392 1640 2.36 1 50 45 50 50 344 54 53 57 52 53 51
51
42 46
6 515 383 389 1680 1.66 1 45 40 40 45 389 64 66 58 58 57 54 50 46 48
7 509 395 377 1650 2.07 1.68 50 40 45 50 357 64 61 56 55 45 46 43 45 42
8 503 378 377 1720 2.36 1.38 55 50 55 60 408 64 56 53 57 62 53 48 43 63
9 507 383 376 1640 1.97 2.04 50 45 50 50 332 61 59 55 60 55 47
49
41 54
10 512 382 362 1660 2.1 2.4 40 45 45 45 342 62 57
56
59
56
60 53 44 51
11 506 384 389 1620 1.44 2.29 75 70 70 75 317 55 53 57 54 52
54 52
46
55
12 510 379 366 1650 2.17 1.97 65 60 65 65 352 60 57 61 61 55 57 54 49
57
13 510 378 374 1640 2.3 2.1 60 50 60 65 336 56 59 61 52 56 53 55 53
14 506 379 364 1630 1.74 2.54 70 60 65 70 328 60 58 58 58 51
54
49 50
56
15 509 380 367 1670 3 2 75 65 70 75 347 63 56
60
61 53
56
55 51 58
- 8. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
498
Table 2 The important boiler operating parameters (Contd.)
S.
No.
Damper openings at Elevations(%)
Burner Tilt(degree)
AA A AB B BC C CD D DE E EF F FG G GH H HJ J JJ 1 2 3 4
1 30 20 40 19 40 20 40 20 40 18 40 20 40 23 40 19 40 20 30 -30 -27 -30 -27
2 30 20 25 20 25 19 25 25 25 20 25 20 25 25 25 19 25 19 29 -30 -27 -30 -28
3 30 20 20 20 20 20 20 23 20 20 20 20 20 20 20 20 20 19 30 -30 -30 -27 -27
4 30 20 24 20 24 25 24 23 24 20 24 20 24 24 24 20 29 19 30 -29 -27 -30 -27
5 30 20 20 20 25 24 20 20 20 20 20 20 20 22 20 20 20 20 30 -30 -30 -30 -28
6 30 20 40 20 40 20 40 25 40 30 40 23 40 25 40 19 40 20 32 -29 -27 -30 -27
7 30 25 30 20 30 25 30 24 30 20 30 25 30 25 30 20 30 20 30 -30 -27 -30 -27
8 30 20 20 18 20 18 20 18 20 18 20 18 20 24 20 20 20 20 30 -24 -24 -23 -23
9 30 20 30 19 30 20 30 30 30 43 30 20 30 24 30 19 30 20 28 -27 -30 -27 -27
10 30 20 18 20 18 20 18 45 18 19 18 45 18 20 18 20 18 10 30 -27 -24 -27 -27
11 30 22 18 20 18 20 18 20 18 22 18 20 18 21 18 20 18 18 30 -28 -30 -27 -26
12 30 25 30 20 30 20 30 20 30 19 30 30 30 20 30 19 30 18 30 -27 -27 -27 -27
13 30 24 25 24 25 24 25 30 25 40 25 23 25 30 25 25 25 19 31 -27 -30 -27 -27
14 30 25 20 20 20 25 20 30 20 42 20 20 20 20 20 20 20 19 30 -30 -30 -27 -30
15 30 18 18 20 18 24 18 20 18 22 18 23 18 25 18 22 18 20 30 -30 -30 -30 -30
The learning of the network is carried out through adjusting the weights by
continuous iterations and minimizing the error between experimentally measured response
and ANN model- predicted response [17]. Root mean square error (RMSE), Sum of square
error (SSE), and correlation factor (R) are used to evaluate ANN-GA performance. When the
RMSE are at the minimum, and ‘R’ value closer value to 1 represents high performance and
perfect accuracy. The predicted values and measured values for the full load condition are
given in Table 3. ANN model can be combined with the optimizing algorithms. Using the
ANN, the fitness function between the input operating parameters and the NOx emissions can
be obtained.
Table 3 The NOx emission under above operating condition and predicted NOx
Sr. No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Measured
NOx
464 325 454 388 514 469 545 510 485 353 525 385 460 415 470
Predicted
NOx
425 345 470 410 537 445 533 489 510 395 505 370 440 435 495
- 9. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
499
Because the ANN may be considered simply as a nonlinear input–output mapping,
such a mapping is so quick and it is suitable to be used as the objective function for the
optimization algorithms. The optimum operating parameters can be found employing the
searching ability of optimization algorithms.
In this study, simulated annealing is employed to optimize the NOx emissions.
Essentially, the SA method generates a sequence of solutions, which are successively
modified until a stopping criterion is satisfied. A temperature parameter is used to control the
acceptance of modifications. Initially, the temperature is set to a high value and is decreased
over iterations. If the modified solution has better fitness value than the current solution, it
replaces the current solution. If the modified solution is less fit, it is still retained as current
solution but with a probability condition. As the algorithm proceeds, the temperature
becomes cooler, and it is then less likely to accept deteriorated solutions. In each iteration, the
process of generating and testing a new trial solution is repeated for a specified number of
trials, to establish the 'thermal equilibrium'. The last of the accepted solution becomes the
initial solution for the next iteration, after the temperature is reduced, according to the
'annealing schedule'. Thus the main features of the SA process are: the transition mechanism;
and the cooling scheme.
IV RESULTS AND ANALYSIS
The main objectives for boiler combustion optimization are to help the operators to
perform clean and efficient utilization of coal. Thus the NOx optimization objective function
was derived from the weights and biases of the trained feed forward back propagation neural
network. Weights and biases of all layers of neurons were combined with transfer functions
of ANN model to achieve an equation pattern as the following steps. The 9 input layer nodes
with the 1st bias node connected to 10 nodes of hidden layer. Thus, there are 90 values of
weights and 10 values of biases on the layers between input and hidden layer. On the hidden
layer, the ‘tansig’ transfer function is used to calculate the sum of the 90 weighted inputs and
the 10 biases. The 10 nodes of hidden layer connected to one node of output layer. It means
the layers between hidden layer and output layer have 10 values of weights rows and one
value of bias. On the output layer, the ‘purelin’ transfer function is used to calculate the sum
of the 10 weighted inputs and one bias. Then sum of weights and bias in output layer is
displayed.
The ANN is used to train the operating parameters by considering the experimental
data from Table 1 and Table 2. It determines the weights between processing elements in the
input and hidden layer and between the hidden layer and output layers which minimize the
differences between the network output and the measured values. The experimental data
stated above are used to find the relation between the operational parameters and the NOx
emission concentration in flue gas under full load condition. The trained network achieved
highest R and lowest MSE (R=0.9627; RMSE=0.4908) using trial-and-error procedure.
Average performance of BPNN is given in Table 4.
- 10. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
500
Table 4 The performance of BPNN
Training
Data
RMSE R Approximate
Accuracy
Set I 0.4927 0.9627 93.88 %
Set II 0.4991 0.9501 93.69 %
Set III 0.4908 0.9484 92.03 %
The measured and predicted NOx emission concentration in flue gas shown in Figure
6 and Figure 7 indicates that the trained network is performing reasonably good in prediction.
Fig. 6 Measured and predicted NOx Fig. 7 Measured versus predicted NOx
The SA is employed to search the optimum solution to determine the optimum
operating parameters like the flue gas oxygen, nozzle tilt, flue gas outlet temperature and
secondary air burner damper opening position for minimum NOx emission. The other input
parameters, such as pulverizer feeder opening value, air flow rate through the pulverizer and
wind box differential pressure are usually not considered as adjustable factors for combustion
optimization purpose. These parameters can also be optimized with the same method if
necessary. The search process is progressive and the rate of convergence is very fast,
consuming a few minutes of CPU time on a modern desktop computer. The reduced NOx
emission concentration and optimum operating parameters such as the flue gas oxygen,
nozzle tilt, flue gas outlet temperature and secondary air burner damper opening position for
full load condition are listed in Table 5.
0
100
200
300
400
500
600
1 2 3 4 5 6 7 8 9 101112131415
NOxEmissions(ppm)
Number of sample
Measured
NOx
0
100
200
300
400
500
600
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
NOxEmissions(ppm)
Number of sample
Measured NOx
Predicted NOx
- 11. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
501
Table 5 Optimum operating parameters for 500MW
FG
Tem
p at
Eco.
Outl
et O2(%)
Damper openings at Elevations(%) Burner Tilt(degree)
(deg
C) L R AA A AB B BC C CD D DE E EF F FG G GH H HJ J JJ 1 2 3 4
373 1.4 1.5 30 20 31 20 29 19 27 19 26 18 26 17 22 20 21 20 19 17 20 -30 -29 -29 -30
The optimized results agree well with the experimental experience, leading to low NOx
emission in full load condition.
V CONCLUSIONS
In a nutshell, this paper has brought to focus the ability to model the NOx emission
from a 500 MW tangentially fired boiler under full load condition. It is developed and
verified with working parameters. The results show that the back propagation-feed forward
neural network method is accurate, and it can always give a general and suitable way to
predict NOx emission under various operating conditions and burning different coal.
Combined with simulated annealing, the optimum operating parameters can be achieved to
decrease the NOx emission. The results proved that the proposed approach could be used for
generating feasible operating conditions. It must be mentioned that the proposed algorithm is
robust in terms of the class of the problem that it can handle, be it classification or complex
regression with multiple dimensions. Overall, the primary objective of the paper has been
achieved, and in the future a hybrid model that involves both neural network and support
vector machines will be developed in order to further improve the predictive accuracy and
computational efficiencies.
VI ACKNOWLEDGEMENTS
Heartfelt gratitude to National Thermal Power Corporation Limited, India for
providing operational information of Power Plant to carry out this work successfully.
VII REFERENCES
[1] O. P. Rao, Coal gasification for sustainable development of energy sector in India,
Technical Report, Council of Scientific & Industrial Research, New Delhi, India, 2003.
[2] G. S. Springer and D. J. Patterson, Engine emissions: pollutant formation and
measurement (New York, Plenum Press, 1973).
[3] Chapter 85: Air Pollution Prevention and Control, U. S. Environmental Protection
Agency, Clean Air Act as of 2008.
[4] A. Copado, Boiler efficiency and NOx optimisation through advanced monitoring and
control of local combustion conditions, Sixth International Conference on Technologies
and Combustion for a Clean Environment, Porto, Portugal, 2001, 903-908.
- 12. International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-
6367(Print), ISSN 0976 – 6375(Online) Volume 4, Issue 3, May – June (2013), © IAEME
502
[5] A. Maciejewska, et al., Co-Firing of biomass with coal: constraints and role of biomass
pre-treatment, Technical Report, Joint Research Centre, Institute for Energy,
Netherland, 2006.
[6] S. Thompson and K. Li, Thermal power plant simulation and control (United Kingdom,
The Institution of Electrical Engineers, 2003).
[7] L. Fausett, Fundamentals of Neural Networks (New Jersey, Prentice-Hall, 1994).
[8] S. S. S. Chakravarthy, A. K. Vohra and B. S. Gill., Predictive emission monitors for
NOx generation in process heaters, Elsevier Journal of Computers and Chemical
Engineering, 2000, 1649–1659.
[9] Ligang Zheng, Shuijun Yu and Minggao Yu. Jiaozuo Henan, Monitoring NOx
emissions from coal-fired boilers using generalized regression neural network, IEEE,
China, 2008, 1916-1919.
[10] S. V. Patankar, Numerical heat transfer and Fluid Flow (New York, Hemisphere
Publishing Corporation, 1980).
[11] Y. Yu, et al., Neural-network based analysis and prediction of a compressor’s
characteristic performance map, Journal of Applied Energy, 84 (1), 2007, 48–55.
[12] S. A. Kalogirou, Applications of artificial neural networks in energy systems - A
review, Energy Conversion & Management (40), 1999, 1073 – 1087.
[13] MATLAB, Version 6.5.0, MathWorks Inc., Help Files, 2002.
[14] S. Haykins, Neural networks: a comprehensive foundation (2nd ed. Englewood Cliffs,
NJ : Prentice-Hall, 1999).
[15] R. Backreedy, Prediction of unburned carbon and NOx in a tangentially fired power
station using single coal blends, Journal of Fuel (84), 2005, 2196-2203.
[16] Z. Hao et al., Optimizing pulverized coal combustion performance based on ANN and
GA, Journal of Technology (85), 2003, 113– 124.
[17] A. T. C. Goh, Back propagation neural networks for modelling complex systems,
Journal of Engineering (9), 1995, 143-151.
[18] H. M. Yao, H.B. Vuthaluru, M.O. Tade and D. Djukanovic, Artificial neural network-
based prediction of hydrogen content of coal in power station boilers, Elsevier Journal,
2005.
[19] Manoj B. Karathiya, Deven J. Patel and Parag C. Shukla, “Application of Artificial
Neural Network in Metropolitan Landscape Water Eminence Assessment”,
International Journal of Computer Engineering & Technology (IJCET), Volume 4,
Issue 2, 2013, pp. 623 - 631, ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375.