You are on page 1of 7

I

Abstract

This thesis deals with the control of level in the drum boilers which
are used in electrical power plants. New techniques are applied a kind of
artificial intelligence known as neural networks for identification and control
of a drum boiler.
Identification networks are trained offline using real process
data recorded at Al-Daura power plant / unit no.4. Drum level is selected as
an output of identification network while pressure, feed water flow, steam
flow are selected as an input.
The first step of work was to find network with the least number of
hidden neuron and the least value of mean square error. The network is
simulated and its output is plotted with the real data. Two types of networks
are tested feed forward neural networks and recurrent neural networks. The
predicted mean square error is more than 1.0 E-04 for feed forward networks
and less than 1.0 E-06 for recurrent networks.
The second stage is the prediction of the inverse model. A MIMO
system network was used for this purpose. Computer programs are
implemented using neural network tool box which is available in MATLAB
version 6.5.
II

CONTENTS

Section Subject Page

Abstract I
Contents II
Nomenclature IV
Abbreviations VI
Chapter One:Introduction
1.0 Introduction 2
1.1 Drum Boilers 3
1.2 Boiler Level Control Strategies 4
1.2.1 Single Element Control 5
1.2.2 Two Element Control 6
1.2.3 Three Element Control 7
1.3 Neural networks 9
1.4 Aim of The Present Work 10
Chapter Two: Literature Survey
2.0 Introduction 12
2. 1 Process Dynamics and Control 12
2.2 Neural Networks
2.2.1 Process Control Applications 15
2.2.2 Neuro- Control Techniques 19
Chapter Three:Neural Networks Identification& Control
3. 0 Introduction 24
3.1 The Mathematical Model of Neuron 25
III

Section Subject Page

3.2 Backpropagation Algorithm


3.2.1 Definition 26
3.2.2 Derivation of Backpropagation Algorithm 29
Chapter Foure:Case Study Results & Discussion
4.0 Introduction 38
4.1 Data Collection and Analysis 38
4.2 Identification 41
4.3 Recurrent Neural Network 49
4.4 Inverse Model 53
Chapter Five: Conclusion and Future Work
5.1 Conclusion 57
5.2 Recommendation for Future Work 57
References and Appendixs
References 59
Appendix (A):Recorder Charts
Drum Recorder Chart A1
Main Steam Flowrate Chart A2
Appendix (B): Computer Programs
Input Data File & Data Treatment B1
Feed- forward Simulation Program B4
Recurrent Neural Network Simulation B6
Program
IV

Nomenclature

Symbol Definition Units

E Error Function
EP Error for pattern p
Oi Output of Neuron i
Oj Output of Neuron j
P Pressure kg/ cm2
q FW Feed Water Flowrate Ton/hr
qS Main Steam Flowrate Ton/hr
net j Net Input to the Neuron j
Nh Number of Neurons in Hidden Layer
r Reference Signal
u Manipulated Input
Wj,i Weight Factor, multiplied by the signal going from
Neuron i in the input layer to neuron j in hidden layer.
Wk,j Weight Factor, multiplied by the signal going from neuron
j in the hidden layer to neuron k in the output layer.
ym Model Output
yp Plant Output
y Output Signal


V


Symbol Definition Units
mentum Factor

k Generalized Error Signal.

 Learning Rate.
 pk The desired output for the kth component
VI

Abbreviations

Symbol Definition

CSTR Continuous Stirred Tank Reactor


(CV's) Controlled Variables
(DV's) Disturbance Variables
FE Flow Element, Orifice Plate
FIC Flow Indicator Controller
FT Flow Transmitter
FTNN Feed Through Neural Network
FV Flow Control Valve
FY Flow Function, Square Root Extractor
HS Hand Switch
IA Instrument Air
LIC Level Indicator Controller
LT Level Transmitter
LV Level Control Valve
LY Level Function
MIMO Multi Input Multi Output System
MPC Model Predictive Control
MSE Mean Square Error
MV's Manipulated Variables
VII

Symbol Definition

NARMA Nonlinear Auto Regressive Moving Average


NARMA-,L1 Modern Algorithm of NARMA
NARMA-L2 Modern Algorithm of NARMA-L1
pH Acidity Function
PI Proportional plus Integral Control
PID Proportional, Integral & Derivative Control
SISO Single Input Single Output System
SP Set Point

You might also like