2012 NIRMA UNIVERSITY INTERNATIONAL CONFERENCE ON ENGINEERING, NUiCONE-2012, 06-08DECEMBER, 2012

1

Thermal Power Plant Analysis Using Artificial
Neural Network
Purva Deshpande1, Nilima Warke2, Prakash Khandare3, Vijay Deshpande4
VESIT, Chembur, purva.deshpande@yahoo.co.in, nilimavwarke@gmail.com
3,4
Mahagenco, Mumbai. khandarepk@gmail.com, vijay.deshpande.gmit@gmail.com
1,2

Abstract--Coal-based thermal power stations are the leaders in electricity generation in India and are highly complex nonlinear systems. The thermal performance data obtained from
MAHAGENCO KORADI UNIT 5 thermal power plant shows that heat rate and boiler efficiency is changing constantly and the plant is probably losing some Megawatts of electric power, and more fuel usage thus resulting in much higher carbon footprints. It is very difficult to analyse the raw data recorded weekly during the full power operation of the plant because a thermal power plant is a very complex system with thousands of parameters. Thus there is a need for nonlinear modeling for the power plant performance analysis in order to meet the growing demands of economic and operational requirements. The intention of this paper is to give an overview of using artificial neural network (ANN) techniques in power systems. Here Back
Propagation Neural Network (BPNN) and Radial Basis Neural
Network (RBNN) are used for comparative purposes to model the thermodynamic process of a coal-fired power plant, based on actual plant data and this works as the internal model for prediction of the Heat Rate and Boiler Efficiency. This ANN model of the thermodynamics of a power plant is used to determine the influence of changes in different variables upon the heat rate and boiler efficiency through the use of sensitivity coefficients, which indicate the directions of change in the variable that will improve heat rate and boiler efficiency, and thus indicates the relative importance of these different variables. This information can be used to provide guidance to the plant operators and engineers as to where they should expend their efforts to improve the heat rate and boiler efficiency. Further variation in these key parameters predicted by sensitivity analysis helps in improvisation of Heat Rate and
Boiler Efficiency.
Index Terms-- Back Propagation Neural Network (BPNN),
Heat Rate and Boiler Efficiency performance, Radial Basis
Neural Network (RBNN), Sensitivity study.
I.

INTRODUCTION

The need for real-time diagnosis of steam plant has been recognised for many years. Current economic and social factors put stringent requirement on steam power plants to be operated at high efficiency. Although most power plants are equipped with extensive data acquisition and performance monitoring system, this large volume of data and alarms usually do not provide many means of intelligent interpretation and diagnosis of problems for predicting corrective measures. Power plant operators have to cope with the day today operational problems and are faced with several hundreds of measurement values. It is rather difficult for them to analyse all the data and evaluate accurately the performance of the plant operation. Thus, there is a strong
978-1-4673-1719-1/12/$31.00©2013IEEE

tendency towards the development of' expert system to monitor the real-time boiler efficiency and heat rate and to advise operation for appropriate actions. This work reports into the real-time complex domain of steam power plant operation by developing a prototype system which is interfaced to the power plant's data acquisition system and will provide useful operation guide and diagnostic aid to the plant operator to improve boiler efficiency and heat rate. The development was motivated by an immediate need for a powerful and flexible real time knowledge-based system for the power generation industries to improve boiler efficiency and heat rate [1]. To maximize the utility of coal use in power generation, heat rate and boiler efficiency improvement is necessary and thus these are key performance indicators
(KPIs) whose performance analysis is needed continuously.
Improvement of these KPIs has several benefits like prolonging the life of coal reserves and resources by reducing consumption, reducing emissions of carbon dioxide (CO2) and conventional pollutants, increasing the power output from a given size of unit and potentially reducing operating costs [2]. The Electric Power Research Institute (EPRI) initiated a systematic study of heat rate monitoring in 1982. A literature survey of this study shows that currently there exist different methods for performance-monitoring of coal-fired power plants. These approaches involve empirical relationships, approximations of the actual processes and, often, linearization of nonlinear phenomena [3]. A potential drawback of these approaches is that it is usually dependent upon system models based on ideal conditions. Hence ANN of the thermodynamics of a power plant can be used to determine the influence of changes in different variables upon the heat rate through the use of sensitivity coefficients. This information can be used to provide guidance to the plant operators and engineers as to where they should expand their efforts in order to improve the heat rate. In the existing plants, large number of operational data is captured on continuous basis by online monitoring system for the plants proper operation. The data is usually stored in the database format only. A highly developed instrumentation and control methodology provides an opportunity to improve the plant performance through the advance management of the above processes. An important issue in these highly computerized and automated systems is the quality of information provided by the sensors as well as quality of decisions passed to the actuators. Hence ANN is used as it is an intelligent system capable of self examination. The combined information of

2012 NIRMA UNIVERSITY INTERNATIONAL CONFERENCE ON ENGINEERING, NUiCONE-2012

2

sensors through automated control system along with the
ANN forms a decision support system in online thermal performance analysis for the plant personal [4]. In this present work performance analysis of heat rate and boiler efficiency are considered since they are KPIs in thermal power plant performance. Thus here application of ANN as analytical tool in predicting the relationship between various plant parameters and KPIs is done and to improve overall thermal plant efficiency is done with the help of ANN. The
ANN is a technique with a flexible mathematical structure that is capable of identifying complex nonlinear relationships between input and output data. Therefore, the present work aims to utilize an ANN to simulate the thermal power plant relationship using the applications of the RBNN and BPNN for simulation [5].

a non-linear mapping from the input space. The final layer performs a simple weighted sum with a linear output. The unique feature of the RBNN network is the process performed in the hidden layer. The idea is that the patterns in the input space form clusters. If the centres of these clusters are known, then the distance from the cluster centre can be measured. Furthermore, this distance measure is made nonlinear, so that if a pattern is in an area that is close to a cluster centre it gives a value close to 1. Beyond this area, the value drops dramatically. The most commonly used radial-basis function is a Gaussian function in a RBNN network; r is the distance from the cluster centre. Here the distance is calculated by Euclidean distance [6].
Fig.2 below represents the RBNN structure:

II. PRINCIPLE OF BPNN AND RBNN
1. BPNN
It is multilayer feed forward network and is the most widely used neural network, it has a strong nonlinear mapping ability, generalization ability, fault tolerance. The basic idea of BPNN is that the network training process is composed of the signal of the Forward Propagation and error
Back Propagation two Processes like Forward Propagation, the training sample input from the input layer through hidden layer to output layer to predict the output of training samples.
Determine whether the predicted output to meet the accuracy requirements, if the error of the predicted output and actual output does not meet the precision requirement and does not reach the maximum training time, then it will enter the error back-propagation phase, which is that error is transferred layer by layer in some form through the hidden layer to the input layer, and error is apportioned to each layer neural network, thus the error signal of each neuron as the basis of the right to amend the value of each neuron. Weight adjustment process is the network training process of learning, this process continuously loop until the network output error reduced to the required accuracy or to a pre-set maximum number of times [6].
Fig.1shown below represents the BPNN structure:

Fig.1: BPNN
2. RBNN
It is a 3-layer network, consisting of input layer, hidden layer and output layer. This method is an iterative process where the hidden layer adds the neuron to the network till the error goal is met. The input layer gets the input values with pre processing done and is given to the hidden layer performs

Fig.2: RBNN

III. METHODOLOGY PHASES
The integrated methodology consists of plant data acquisition phase, execution phase, sensitivity phase, improvisations of heat rate and boiler efficiency through
ANN and its impact on fuel consumption and fuel cost saving. 1. Plant Data Preparation Phase
The thermal power plant data of KORADI UNIT 5 are obtained from a 200 MW coal fired power plant, recorded every second continuously at generation control room (GCR) at MAHAGENCO Company, here nine input parameters are identified for causing deviation in heat rate and five input parameters are identified for constant change in boiler efficiency. These variables are identified by experienced plant people. Here the plant data is collected on an average 15 min interval data i.e. 96 blocks in 24 hours a day for over a period of one month from 13th April 2012 to 14th may 2012 and then is loaded into an database for preservation and further performance calculation. For execution of neural network 70% data is used for training while 30% data is used for testing the network [3].
2. Variables Selection
Variable selection is used to reduce the number of inputs, enhance the noise immunity, and to improve network generalization by removing spurious inputs known as
“outliers”. In this process ANN learns all relations between all such deviations and trained ANN will be able to find output i.e. Heat Rate and Boiler Efficiency when input data presented to it. Thus careful real plant variable selection is more important for ANN training [3]. Thus for heat rate nine variables and for boiler efficiency five variables are selected with 2246 rows for training and 750 rows for testing are

2012 NIRMA UNIVERSITY INTERNATIONAL CONFERENCE ON ENGINEERING, NUiCONE-2012, 06-08DECEMBER, 2012

selected. Here for BPNN training 2 hidden layer, [10, 10] neurons and RBNN training with spread as 1 and error goal as 0.0002 and 1 hidden layer the variables selected for heat rate and boiler efficiency is shown in Table I and II respectively. TABLE I
VARIABLES SELECTION FOR HEAT RATE

removal illustrates the relative importance of the predictor variables & thus we can conclude highest importance of parameter affecting heat rate & boiler efficiency. Here sensitivity analysis is performed on the better model between
BPNN and RBNN [3], [7].
5. KPI’s Prediction
Once critical parameters affecting the KPIs are predicted through sensitivity analysis and then on the trained ANN out of all the critical parameters varying the most critical parameter under operational constraints leads to improvisation of heat rate and boiler efficiency thus help in saving the fuel consumption and fuel cost.
The entire neural network training is accomplished in the
MATLAB2011A.
6. Results and Discussion:
As per above procedure BPNN and RBNN models are trained showing the following results.
TABLE III
TRAIN AND TEST RESULTS USING BPNN AND RBNN FOR HEAT RATE

TABLE II
VARIABLES SELECTION FOR BOILER EFFICIENCY

3. ANN Training and Validation:
Here BPNN and RBNN are used for comparative purposes.
Here training and validation of BPNN includes training algorithm to minimize the error between target output and network output. Bayesian Regularization algorithm is used for training and RBNN training includes the use of gradient descent algorithm for training. The objective of the BPNN and RBNN training includes comparison of better model for plant performance analysis which can be mathematically cast into the maximization of the correlation (R) and minimization of mean square error (M.S.E) between the target output and the network output [3].
4. Sensitivity Analysis
Sensitivity analysis serves to isolate the most critical parameters of heat rate and boiler efficiency deviation. Here backward stepwise elimination is used as the method for sensitivity analysis. This method assesses the change in
M.S.E of the network by sequentially removing input neurons from the neural network (rebuilding the neural network at each step). The resulting change in M.S.E for each variable
978-1-4673-1719-1/12/$31.00©2013IEEE

3

Fig 3: R graph of train data set for BPNN for Heat Rate

4

2012 NIRMA UNIVERSITY INTERNATIONAL CONFERENCE ON ENGINEERING, NUiCONE-2012

Fig 4: R graph of test data set for BPNN for Heat Rate

Fig 5: R graph of train data set for RBNN for Heat Rate

Fig 7: R graph of train data set for BPNN for Boiler Efficiency

Fig 6: R graph of test data set for RBNN for Heat Rate
TABLE IV
TRAIN AND TEST RESULTS USING BPNN AND RBNN FOR BOILER
EFFICIENCY

Fig 8: R graph of test data set for BPNN for Boiler Efficiency

2012 NIRMA UNIVERSITY INTERNATIONAL CONFERENCE ON ENGINEERING, NUiCONE-2012, 06-08DECEMBER, 2012

5

condenser vaccum and flue gas temperature at air preheater outlet-A. 7. BPNN Predictive Model in Improvisation on Heat
Rate and Boiler Efficiency
Here on the trained BPNN, Prediction of heat rate and boiler efficiency on new set of input data is done and improvisations in heat rate and boiler efficiency are obtained. For the improvisation the real data of 20 April from 7.15 pm to 7.30 pm is given to trained ANN and it predicts output i.e. heat rate as 2381.21kcal/kwh and boiler efficiency as 88.01%.
Fig 9: R graph of train data set for RBNN for Boiler Efficiency
Then by varying the most critical input parameter i.e. condenser vaccum for heat rate from the current value of
666.95mmHg to 663mmHg and for boiler efficiency flue gas temperature at Air Preheater outlet from 155.55 oC to 148 oC as these are the most critical parameters predicted through sensitivity analysis and the improvisations in heat rate of
2374.59kcal/kwh and boiler efficiency is 88.2154% is observed. Below the calculation of fuel saving and fuel cost saving is done using Performance Calculation Code (PTC) under
American Institute of Mechanical Engineers (ASME).
Calculation Method:
Fig 10: R graph of test data set for RBNN for Boiler Efficiency
1. Heat Rate:
Old heat rate is 2381.21kcal/kwh
From the results of BPNN and RBNN training, the BPNN is
New heat rate is 2374.59kcal/kwh found to be better model for plant performance analysis and hence
Heat rate deviation is 6.62kcal/kwh sensitivity analysis is performed on BPNN and sensitivity results
Net saving in Fuel Cost :( 20 April from 7.15 pm to 7.30 are shown in Table V and VI respectively. pm) =Heat rate deviation/Gross Calorific Value of fuel*Fuel
TABLE V
SENSITIVITY ANALYSIS FOR HEAT RATE
Cost* Gross Load
=6.62/3500*3000*141.27 i.e. 801.6063 Rupees
Net saving in Fuel Cost for whole day:
Net saving in Fuel Cost :( 20 April from 7.15 pm to 7.30 pm):*96 =801.6063 *96
=76954.2 Rupees
Net saving in Fuel :( 20 April from 7.15 pm to 7.30 pm):
=Net saving in Fuel Cost :( 20 April from 7.15 pm to 7.30 pm)/Gross Calorific Value of fuel
=801.6063/3000
=0.267202 Metric Ton (MT)
Net saving in Fuel for whole day:
TABLE VI
=Net saving in Fuel :( 20 April from 7.15 pm to 7.30 pm)*96
SENSITIVITY ANALYSIS FOR BOILER EFFICIENCY
=0.267*96
=25.6514 Metric Ton (MT)

Thus from the sensitivity results, it is seen that the most critical parameter affecting heat rate and boiler efficiency are

978-1-4673-1719-1/12/$31.00©2013IEEE

Fig 11: Fuel Saving/Fuel Cost Saving through Improvement in Heat Rate

Here Fig 11 shows Fuel Saving/Fuel Cost Saving through
Improvement in heat rate diagrammatically.

2012 NIRMA UNIVERSITY INTERNATIONAL CONFERENCE ON ENGINEERING, NUiCONE-2012

6

2. Boiler Efficiency:
Heat rate predicted by BP neural network
= 2406.83 kcal/kwh
Unit heat rate predicted due to old boiler efficiency
= (Heat rate/E1)*100
= (2406.83/88.01)*100
= 2734.723327 kcal/kwh
Unit heat rate predicted due to old boiler efficiency
= (Heat rate/E2)*100
= (2406.83/88.21)*100
= 2728.522843 kcal/kwh
Heat rate deviation is = 6.20 kcal/kwh
Net saving in Fuel Cost: (20 April from 7.15 pm to
7.30 pm)
= Heat rate deviation/Gross Calorific Value of fuel *Fuel
Cost* Gross Load
= 6.20/3500*3000*141.27 i.e. 750.7278857 Rupees
Net saving in Fuel Cost for whole day:
= Net saving in Fuel Cost: (20 April from 7.15 pm to
7.30 pm):*96
= 750.7278857 *96
= 72069.88 Rupees
Net saving in Fuel :( 20 April from 7.15 pm to 7.30 pm):
= Net saving in Fuel Cost :( 20 April from 7.15 pm to 7.30 pm)/Gross Calorific Value of fuel
= 750.7278857 /3000
= 0.250243 Metric Ton (MT)
Net saving in Fuel for whole day:
= Net saving in Fuel :( 20 April from 7.15 pm to 7.30 pm)*96
= 0.250243 *96
= 24.02333 Metric Ton (MT)

Fig 12: Fuel Saving/Fuel Cost Saving through Improvement in Boiler
Efficiency

Here Fig 12 shows Fuel Saving/Fuel Cost Saving through
Improvement in boiler efficiency diagrammatically.
Conclusion
In the present work on the thermal performance analysis of power plant the ANN methodology is being implemented.
BPNN with Bayesian Regularization and RBNN are used as training models of ANN. From the training results of both
BPNN and RBNN it is seen that BPNN with BR is more appropriate for this application. ANN capability of predicting the combined effect of input parameters on the output is of great help in this application and is achieved by prediction of critical parameters affecting heat rate and boiler efficiency through sensitivity analysis and improvisation in these outputs is obtained by varying the most critical parameter

predicted by sensitivity analysis on trained ANN. With the help of this improvisation in heat rate and boiler efficiency through ANN, the cost of fuel and fuel consumption is minimized. Thus development of this prototype system is useful and act as decision support system to today’s automated control system for power plant personal in maintaining the power plant functioning efficiently.
ACKNOWLEDGMENT
This work is supported by MAHAGENCO Company for sharing the valuable data of KORADI Unit 5 and VESIT
College for the valuable guidance .With the help of them, this work is accomplished successfully.
IV. REFERENCES
[1]

[2]

[3]

[4]

[5]

[6]

[7]

Alvin, C.R.Wong, C.Y.Teo, H.K. Ho, “Development of a knowledgebased system to improve power plant thermal efficiency”, IEEE 2nd
International Conference on Advances in Power System Control,
Operation and Management, December 1993, Hong Kong.
Robert Holzworth, “Integrated performance monitoring for asset optimization”, Proceedings of POWER 2009 ASME Power 2009 July
21-23, 2009, Albuquerque, New Mexico, USA
Vijaya V. Kantubhukta and Mohamed Abdelrahman, “A Feasibility
Study on using Neural Networks In Performance Analysis Of CoalFired Power Plants”, 0-7803-8281-1/042004 IEEE.
Firaz B.Ismail Alnaimi and Hussain H.AL-Kayiem, “Artificial
Intelligent System for Steam boiler diagnosis based on superheater monitoring”,Journal of Applied Sciences 11(9):1566-1572,2011.
Robert E. Uhrig , J. Wesley Hines and William R. Nelson , “Integration of Artificial Intelligence Systems For Nuclear Power Plant Surveillance and Diagnostics”.
Osman Ahmed Abdalla, Mohd Nordin Zakaria, Suziah Sulaiman and
Wan Fatimah Wan Ahmad, “A Comparison of Feed-forward Backpropagation and Radial Basis Artificial Neural Networks: A Monte
Carlo Study” 978-1-4244-6716-7/10/$26.00 ©201O IEEE.
Julian D. Olden, Michael K. Joy and , Russell G. Death , “An accurate comparison of methods for quantifying variable importance in artificial neural networks using simulated data”, Ecological Modelling 178
(2004) 389–397