Your SlideShare is downloading. ×
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Artificial neural network based nitrogen oxides emission prediction and
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

Artificial neural network based nitrogen oxides emission prediction and

688

Published on

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
688
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
11
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 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.

×