Boiler performance optimization using process neural network

298 X INDIAN JOURNAL OF APPLIED RESEARCH
Volume : 3 | Issue : 7 | July 2013 | ISSN - 2249-555XResearch Paper Engineering
Boiler Performance Optimization Using Process Neural
Network
Subodh Panda Bikash Swain Sandeep Mishra
Associate professor & Head, Dept. of
AE&IE, GIET, Gunupur, Odisha,India
Asst. Professor Dept. of AE&IE,
GIET, Gunupur, Odisha,India
M.Tech 3rd semester student Dept.
of ECE, GIET, Gunupur, Odisha,
India
Keywords
Process neural network, blow down, Indirect losses , soft sensor, Heat exchanger,
Neurocontrol. Flash steam, feed water
ABSTRACT The important issue in the modern thermal power plant is to develop methodology concept algorithm tech-
nologies for designing a control system which must be able to evolve, self develop. self organized and self
evaluated and to self improve. Although linear model can provide acceptable performance for many systems they may be
unsuitable for non linearity. So it is highly required for a model that reflect the non liner relationship between cause and
effect variable. Implementation of soft sensor in neural network estimate process data using self organizing neural network.
Here basic requirement of design an neuron control with soft sensor are the knowledge of fundamental relationship of
process variable and the parameter in the question .It is called a process neural network which is an extension of traditional
neural network in which the input and output are time variant. So the data processing is better than traditional neural net-
work so it is highly suitable to minimize heat loss at blow down station and increase its ability by operating at peak.
1:-INTRODUCTION
The process of in corporating process neural network in the
power plant to ahead of the limits of conventional technique
and to improve the existing tools by optimizing the dynami-
cal performance of process such as blow down station in
power plant .The soft sensor, know also as virtual sensor are
soft ware tools capable of calculating quantities that can not
be measure or that are difficult to measure. They are based
on technology that provide an estimatimation of measure-
ments by creating a mathematical model for real data. The
increase the potential of neural network with this support
we can design a neural control where the actuating signal at
the input to the plant is generated by neural network based
control. The properly trained neural network act as control-
ler and appropriate actuating signal for the plant is directly
available at the networks output. . The main propose of this
research work is to use this technique in thermal power plant
for increase its efficiency through reducing various indirect
losses. When water is converted to steam, the dissolved sol-
ids are left behind in the boiler water. As steam is continu-
ally produced, evaporated, and replaced with new water, the
amount of solids in the boiler continues to increase indefi-
nitely. These will inevitably collect in the bottom of the boiler
in the form of sludge, and are removed by a process known
as bottom blow down. Insufficient blow down may lead to
carryover of boiler water into the steam, or the formation of
deposits. Excessive blow down will waste energy, water, and
chemicals. The optimum blow down rate is determined by
various factors including the boiler type, operating pressure,
water treatment, and quality of makeup water. Table1: Model
Report Format for Boiler Efficiency and Indirect Losses Sys-
tem
CODE: AL1 DATE: 19/12/12 12:00 AM FUEL: OIL

BOILER EFFICIENCY 69.35
INDIRECT LOSSES 30.65
(I) DRY FLUE GAS LOSS 7.42
(II) FUEL MOISTURE LOSS 6.31
(III) BLOW DOWN LOSSES 14.77
(IV) INCOMPLETE COMBUSTION
LOSS 0.00
(V) AIR MOISTURE LOSS 0.00
(VI) RADIATION & CONVECTION
LOSS 2.15
COURTESY:- THERMAX BOILER,J.K.PAPER,RAYAGADA
2: NEED OF WORK:- Although reducing blow down results
in substantial fuel savings, this function cannot be eliminated
entirely. A boiler operating on high quality feed water needs
very little blow down, while equipment using feed water con-
taining solids, alkalinity or silica requires a much higher rate,
may be even continuous discharge. A comprehensive energy
conservation program, apart from ensuring correct amount
of blow down must include the use of flash tanks and heat
exchangers to reclaim some of the energy wastage in boiler
blow down.
2.1 Flash Steam Recovery
Flash steam heat recovery is a method for recovering at least
85% of the blow down heating value. About half of the heat
contained in the blow down water is recovered in the form
of flash steam by discharging the flow to a flash tank, usually
operated at 5 psig. A portion of the blow down flashes to
steam at the lower pressure and is available for use in Flash
steam recovery is calculated using the formula:
A = (H – W) / L
Where:
• A = Flash steam %
• H = Boiler blow down water heat content, Btu/lb
• W = Water heat/content at flash pressure, Btu/lb
• L = Steam latent heat content at flash pressure, Btu/lb
Assuming, a flash tank is added to a boiler operating at 235
psig
2.2 Blow down heat recovery
Heat exchangers can reclaim the sensible heat from the blow
down that goes into sewerage for heating boiler makeup
water and the like. In most cases, the heat exchanger is de-
signed to reduce the temperature of the blow down water
to within 120 °F of the temperature of the makeup water.
Additional heat recovered is calculated from the following
formula:

INDIAN JOURNAL OF APPLIED RESEARCH X 299
Volume : 3 | Issue : 7 | July 2013 | ISSN - 2249-555XResearch Paper
M = J x (1- A) x (W – P)
Where:
• M = Additional daily heat recovery, Btu
• P = Water heat content at exchanger outlet, Btu/lb
The system shown below consists of a flash tank and a heat
exchanger. The temperature of the blow down leaving the
flash tank is usually still above 220°F. The heat of this flash
blow down can be used to heat makeup water by sending
it through the heat exchanger, while cooling the blow down
at the same time. Heating boiler makeup water saves on fuel
costs. .Fig 1
3: PROPOSE MODEL:-
3.1 Introduction
The generic non-linear solution to the system like Blow down
monitoring can be provided through process neural network
(PNN). It is so as the dynamic relationship between various
causes and effect variable is capable through artificial neu-
ral network (ANN). ANN has the benefit of being capable
to capture non-linear relationships. As we know the perfor-
mance of process manufacturing and control algorithm are
dependent upon the precession of the model associated
with it. The required potential to provide benefit to these
algorithms can be achieved through PNN when it is applied
to non-linear system.
3.2 Components And Working
In linear model the existing neural network provide a clear
view of relationship between effect variable and its causes.
The real benefit of using ANN in all linear modes is that they
can provide better modeling of non-linear relationship. For
this system there are various ANN structure but they process
some common features. They are generally composed up
numerous processing elements called as nodes which are
arranged together to form a network. The commonly used
processing element is one which weights the input signal
and than sum them with a bias assigned to it. The summed
weighted inputs are passing through a non-linear activation
function such as the hyperbolic tangent and this is the output
of neurons. The model approximation for blow down optima
ion is gradual learning based on the observation of the blow
down loop input/output data in real time.The functioning of
blow down station describe with the differential equation
Y(K+1)=a Y(K) +b X(k)
Y(k) and X(X) are output and input to BD respectively.
K is the time step number and a,b are constant coefficient
as per design of BD station with soft sensor in neural net-
work (PNN) as shown in fig.2Here the control signal X(K) for
actuating control value is determined by a pair of value y(k)
and y(k+1) which are the present and next blow down sta-
tion output(TDS, Drum level , BD temp.)The sequences value
of y(k) for k=1,2,3,………… is assumed to be known and is
provided by the reference blow down output (TDS, drum
level, Blow down temp.) generator. The conventional con-
troller initially produce the control signal x (k). The control-
ler hear in use is _Ve feedback configuration. The reference
blow down output generator inputs the present output value
of thee station to summing node S1. Due to the close loop
condition system between the reference plant output to the
summing node S1 and the ui nit plant output .To fulfill this
condition the close loop control l system must have classical
controller, plant & node S1 be in stable and it is definite high
gain conventional controller. At the initial training state the
neuro control provides little or no input to the summing node
S2.The neural controller associated with soft sensor hear is
memory based. It is not initially trained and its initial output
is zero to S2. There for the dominant components of the total
Actuating signal (k) is the output of traditional controller. The
control signal produced by PID controller in each control step
x(K) is generated in PNN of y(K), y(K+1), which is due to the
learning
rules of w(K+1)=WK + α [X(K)W(K)]
In this weight adaption 0>α is a small +ve learning con-
stant, K denots learning step number and also the number
of the control step, The initial weight value are set to zero.
Which corresponds to setting the entire characteristic sur-
face to zero at the beginning of training .The conventional
controller build the weight W (k) can now be interpreted as
updated control signal value (K) . At the learning value W, the
learning end up and x (k) the averaged value over number
neuro control learning steps.
Fig 2
4. APPLICATION OF PNN BLOW DOWN MANITORING.
The Process Neural Network which describe above can be
use successfully at BD STATION for its optimization. Here
the present measure valu of TDS of boiler drum water can
compare with set valu , at the same time to have a success-
ful drum level control to support the lode required by the
system we have to consider the steam flow as parameter
hence dischargeing BD and its monitoring is dependable to
steam flow also.To reduse the heat lossess in this system ,to
utilize the maximum amount of heat for heating feed water,
Blow down water temp. is a parameter for feed water control
which effect the boiler drum and at the end steam flow that
gives required TDS in discharge water.
The required algorithm regulate the error for minimization to
have a control signal to
Pneumatic control valve placed at three different point in
BD station..The valves regulate the excess water in the blow
down system and feed water line for monitoring TDS and
Feed Water temp.
The systematic use of PNN in BD station for its optimization

300 X INDIAN JOURNAL OF APPLIED RESEARCH
Volume : 3 | Issue : 7 | July 2013 | ISSN - 2249-555XResearch Paper
is shown in the fig(3 )
5. RESULT AND DISCUSSION
The blow down losses of the Boiler is one of the most sensi-
tive parameter involved in all Thermal Power Plants. This pa-
rameter is directly related to the control of the temp. of feed
water and the control of the TDS. A careful study was done
on the existing PI controller system in the Power plant and
compared with the designed PNN model.. The results ob-
tained with the proposed neural network of the boiler plant
shows that soft sensor can converge the neural network to
improve performance. fig. 4 shows comparison of neural net-
work results with desired blow down control and percentage
of deviation of BD monitoring without neural network. It has
been clearly proved that the PNN model is more accurate
and efficient making the system robust and reliable as com-
pared to the former..
The results of PNN are very sensitive to number of neurons.
Increasing the number of neurons in hidden layer will de-
crease the number of calculation steps with subsequent de-
crease in summed squared error. The proposed PNN control-
ler can replace a conventional controller, and is shown to
overcome most of the problems mentioned above. A training
algorithm is derived based on BP, enabling the neural net-
work to be trained with system output errors, rather than the
network-output errors. In the BP algorithm, weights need to
be modified by using the network-output error that is not
known when a multilayer perception is applied directly to the
controlled plant. Therefore, the proposed algorithm enhanc-
es the NN’s ability to handle control applications. The only a
priori knowledge about the controlled plant is the direction
of its response, which is usually easy to determine. The pro-
posed PNN controller has been applied to the BD control in
a thermal power plant and extensive simulations conducted
show promising results. As per Fig.5 we determine that the
maximum effective operating range is between 4500 μ mhos
and 5000 μ mhos, then we can set the controller to control
blow down at that range (as shown below). Below 4500
μmhos we are wasting water, chemical, and fuel. While above
5000 μ mhos we are risking the generation of wet steam.
This shows that inadequate treatment of boiler water can
waste as much as 15-18% of energy and can even result in
plant damage. Feed water if not properly treated shall re-
sult in scale buildup which can reduce boiler efficiency by
as much as 10-12% and can even result in plant damage .It
is important to maximize condensate return an additional is
to 18% of boiler energy from the steam is needed to re-heat
each pound of coal makeup water.
6. CONCLUSION
The proposed process neural network proves to be an ef-
ficient modeling system for calculation and optimization of
the Blow down. It significantly reduces the frequency of de-
viations and the degree of deviation of the TDS, F .W temp.
that can reduce indirect losses .the tripping of the boiler dur-
ing load fluctuations. Focusing on process control systems, a
new controller using soft sensor in neural networks has been
designed and tested for the Blow down control in a thermal
power plant. For such a control system, the negative effects
of a long system response delay and nonlinear elements are
the main obstacles in designing a high performance con-
troller and fine-tuning its parameters. Good performance,
a simple structure and algorithm, and the potential for fault
tolerance make the proposed PNN controller attractive for
process-control applications.By proper use of this PNN tech-
nique it is possible to increase boiler efficiency and also con-
sistency of boiler efficiency can also be maintained by reduc-
ing losses.. Hence an approach to design a PNN controller
used for the blow down control optimization in the process
of a power plant boiler has been presented. The optimization
of the BDC will reduce the in direct losses and improve the
efficiency in the boiler system.
REFERENCE • Dukelow, S. G. (2001), The Control of Boilers 2nd Edition, Instrument Society of America, USA | • Lab VIEW (1998), User Manual, National
Instrument Liptak, Béla G. (1999), optimization of Industrial Unit Processes, CRC Press LLC, USA | • [1] P. J. Werbos, "Backpropogation through
time: What it does and how to do it", proceedings of the IEEE, Vol. 78, No. 10, October 1990. | • [4] L. Fausett, Fundamentals of Neural Networks, New Jersey:
Prentice-Hall, 1994. | • [11] S. N. Sivnandan, S. Sumpathi, and S.N. Deepa, “Introduction to Neural Networks using MATLAB 6.0”, The Mc Graw Hill Publication, New
Delhi | • University. Boiler Efficiency Institute. Auburn, Ala. 1991. | • “Minimizing Boiler Blow down.” Steam Tip Sheet #9. Department of Energy. Office of Industrial
Technologies. Energy Efficiency and Renewable Energy. June 2001. | • Thermal power plants efficiency optimization using back propagation neural network ____
Subodh Panda and Bikash Swain, ATCIT, 40-43, ISBN NO:978-93-83060-00-9 | • Introduction to Artificial neural system jacek M .ZURADA ISBN, 81-7224-650-1 | •
Subodh Panda and Bikash swain. Enhancing thermal plant efficiency using soft computing. IJERT, Issn 2278-018, Vol2issue 3, March-2013 | • Preetie. Manke& sharad
tembhurne. Application of back propagation neural network to drum level control in thermal power plant IJCSI, vol9 issue2. no 1 march-2012. Issn -1694-0814 | •
Subodh Panda, Bikash Swain, Sandeep Mishra, Blow down Losses Control in thermal power plants using neural Network. International journal of Advancements in
Research & Technology, Volume2, Issue5, May-2013 ISSN 2278-7763. | • Improving energy efficiency of boiler system, A. Bhatia. CED Engineering.com. |

Boiler performance optimization using process neural network

More Related Content

What's hot (18)

Similar to Boiler performance optimization using process neural network (20)

Recently uploaded (20)

Boiler performance optimization using process neural network