MINISTRY OF EDUCATION AND TRAINING
UNIVERSITY OF TRANSPORT AND COMMUNICATIONS

NGUYỄN DUY TRUNG

RESEARCH ON CONSTRUCTING
THE HYDROELECTRIC TURBINE SPEED CONTROL
SYSTEM FOR INTERCONNECTED AREA
BASED ON FUZZY LOGIC
AND ARTIFICIAL NEURAL NETWORKS

Course: Control Engineering and Automation
Code: 9520216

SUMMARY OF
ENGINEERING DOCTORAL THESIS

HÀ NỘI – 2020


The Thesis was completed at:
UNIVERSITY OF TRANSPORT AND COMMUNICATIONS

Scientific Instructors::
1. Prof., Dr. Lê Hùng Lân
2. Associate Prof., Dr. Nguyễn Văn Tiềm
Reviewer 1:
Reviewer 2:
Reviewer 3:

The thesis will be defended to the University - Level Doctoral

turbine speed control system.
Research on building regional-linked hydroelectric turbine speed
control system on the basis of fuzzy logic and artificial neural networks
to improve control quality.
3. Research method
Study the actual technological process of the operating mode of the
hydroelectric automation system.
Research, construct and survey a simulation model of a hydraulic
generator turbine based on Matlab / Simulink simulation tool with actual
parameters of the unit, using new intelligent control algorithms.
4. Object and scope of the study
- Researching equipment and technology for turbines for
hydroelectric plants in single and two regions.
- Study the process of operating the plant and power system, study the
fault types of the unit and the influence of parameters such as: Unit
failure, generator capacity, frequency when load changes. , linking with
factories in the power supply area.
Designing PI type fuzzy logic controller is based on optimization
algorithms such as instrumentation optimization (PSO), genetic
algorithm (GA), differential evolution (DE).


2
Synthesis of the neuron controller combined with predictive control
algorithms (ANN - MPC), nonlinear regression (NARMA), adaptive
control with reference model (MRAC) applied in frequency control load
numbers of hydropower systems linking the two regions.
5. Scientific and practical significance
* Scientific significance:
Develop the intelligent control algorithms based on the application of

significant proportion. In 2014, hydroelectricity accounted for about
32% of total electricity production. According to forecasts of Power


3
Planning VII (PDP VII), by 2020 and 2030, the proportion of
hydropower is still quite high, corresponding to 23%.

Figure 1.1. Hydroelectric plant model
1.2. Automation systems in hydroelectric plants
In the hydroelectric plant, the automation system in the plant is very
important, because all operations and troubleshooting are done
automatically.
1.3. The problem of controlling frequency and active power in the
electrical system
1.4. The problem of frequency control of generation with regional
linkage
1.5. Review of studies
- Overseas research:
In the world, the research on integrated control systems for singlearea has been studied for a long time, now basically solved the small load
and independent power generation. Currently getting more attention in
applying intelligent control theory such as fuzzy system and artificial
neural network system.
The problem of automatic generator control (AGC) or LFC load
frequency control in electrical systems has a long history and is one of
the most important topics of interconnected electrical systems. In an
electrical system, the LFC controller as an auxiliary generator plays an
important and fundamental role in maintaining the electrical system
reliability at an adequate level. In LFC practice, components rapidly
change system signals that are virtually invisible due to the filters

(2 regions)

PS for DG and
RERs

The optimal
control method

Objective
functions

Trend research
direction

Three area
(3 regions)

Smart gird

Adaptive
controls

Computer –
based control

Four area
(4 regions)

Small gird


number
2
Thủy điện - Nhiệt điện
2
Thủy điện - Nhiệt điện, Ga
2 Hydropower- Thermal power, Gas
2
Hydropower - Wind - Diesel
2
Thermal power, Gas
2
Thermal power
2
Thermal power
3 Hydropower- Thermal power, Gas
2
Hydropower- Thermal power
2
Hydropower- Thermal power
3 Hydropower- Thermal power, Gas
2 Hydropower- Thermal power, Gas
2
Thermal power
2 Hydropower- Thermal power, Gas
2
Hydropower- Thermal power
2
Thermal power
2
Thermal power

IPSO
ICA
DMPC
QOHC
IPSO
DMPC
BFOA
ANFIN -PS
PSA
CSA


5
Document
[89]
[90]
[91]
[92]
[93]
[94]
[95]

Area
Source type
number
2
Thermal power
2 Hydropower- Thermal power, Gas
2 Hydropower- Thermal power, Gas
Hydropower- Thermal power 2


- Research in the country
In which [9] studied PID controller with fuzzy correction applied to
the problem of hydroelectric turbine operating load in independent mode.
In [8] "Application study of neural fuzzy network to build control
algorithm for hydroelectric turbine velocity control" applied fuzzy neural
network algorithm to adjust PID controller parameters. In [10], research
and application of modern measurement and control solutions to improve
the quality of frequency stability in small and medium hydropower
plants. The method of backstepping, optimal control and Kalman
filtration has been introduced to build adaptive controller to improve the
quality and stability of turbine rotation frequency in small and medium
hydro power plants.
1.6. Select a topic name and research direction
Through analysis, the author chooses the title of the topic: "Research
on building regional-linked hydroelectric turbine speed control system
on the basis of fuzzy logic and artificial neural networks"
1.7. Thesis objectives
- Researching and building models
of interconnected area hydropower
turbine speed control system.
Research
on
building
interconnected area hydroelectric
turbine speed control system on
the basis of fuzzy logic and
artificial neural network, using
optimization
algorithms

ut ( s)

(2.1)

where TW  Lur is constant water start time at rated load, (s),
ag hr

2.1.2. Model of electric - hydraulic servo system,
Wg ( s) 

 g e ( s)
1

 xe ( s) 1  s.Tg

2.1.3. Model of hydraulic turbine
1  Tw s
 P m ( s)
w t ( s) 

 g ( s) 1  0.5Tw s
2.1.4. Model generator

(2.3)

(2.4)


7
 ( s)


Remote
controll 1

Pm1

PL1

Tw1.s  1
0.5Tws
. 1

1
Tp1.s1

1
M 1s  D1

ACE1
Speed ​1

Wing direction 1

Tua bin 1

Generator 1

1
s


Tw2.s  1
0.5Tws
. 1

Tua bin 2

Pm2
1
M 2 s  D2

PL 2

Generator 2

1
R2

Figure 2.15. Control system mathematical model diagram
Hydropower links the two regions


8
In the thesis, the simulation examples are performed with values
of system parameters as follows [11,16,18]:
Tg1  Tg 2  48.7(s) ; Tw1  Tw 2  1(s)

Tr1  Tr 2  0.513 (s); M1  M 2  0.6 (s);
D1  D2  1 (pu); R1  R2  2.4 (Hz/p.u)
T12  0.0707 (pu)


Governor

Turbine

Generator

∆f1

Compute
∆Ptie12

ACE 2(t)

Governor
FLC 2

Turbine

Generator
∆PL2

∆f2

CONTROL-AREA 2

Figure 3.7. Hydropower network diagram linking two regions


9
3.2.1. Design FLC1 and FLC 2 type PI controller

Fuzzy logic
controller

un(t)



Gu

u(t)

U(t)

u(t)

Control
plant

y(t)

ym(t)
Sensor &
transmitter

Figure 3.8. Typical PI type fuzzy logic controller architecture for the controller

Table 3.1. Suggested fuzzy rule table for PI / PD type fuzzy controller
E(t)
NB
NM

NS
NM

DE(t)
ZE
PM
PM
PS
ZE
NS
NM
NM

PS
PM
PS
ZE
NS
NS
NM
NB

PM
PS
ZE
NS
NM
NM
NM
NB



11
T

T

0

0

J   | e(t ) | dt   | r (t )  y (t ) | dt  min

(3.14)

Figure 3.17. Simulation results for a single-area hydropower plant
(a) Load change; (b) Frequency deviation (speed) response

Figure 3.18. Compare the three FLC controllers for the case
single area hydroelectric plant
Hydropower system links the two regionsThe target function used in
optimization is given by the formula (3.15) below:
T

J   | f1 (t ) |  | f 2 (t ) |  | Ptie,12 (t ) | dt  min
0

(3.15)



NEURAL NETWORK TO STABILIZE THE LOAD FREQUENCY
4.1. Question
4.2. Applying artificial neural network to synthesize zone-linked
hydroelectric turbine speed controller
4.2.1. Basic concepts of neural networks
4.2.2. Methods of training artificial neural networks

4.2.2.1. Supervised Learning
4.2.2.2. Reinforcement learning
4.2.2.3. Unsupervised Learning (Unsupervised Learning)
Table 4.1 Comparing three learning methods of neural networks
Human brain
Learn with the guidance of a teacher
Learning with teacher evaluation
Self-study

Artificial neural network
Supervised Learning
reinforcement learning
Unattended study


14
4.2.2.4. Single layer transmission network
4.3. Strategies for controlling turbine speed in the problem of
hydropower system frequency control using artificial neural networks
4.3.1. The frequency-load control strategy uses a NARMA-L2 controller

Figure 4.9. Model of 2-zone linked hydroelectricity using
NARMA - L2 controller based on ANN

controller
4.4.2.1. Simulation results for NARMA and MRAC controller
(i) In the first scenario, the variable load occurs in each area at
different times and intensity (see Figure 4.16-4.18).

Figure 4.16. Simulation results for the first simulation scenario
(a) Load changes; (b) Dynamic response of frequency deviation in the first
region; (c) Dynamic response of frequency deviation in the second region


16

Figure 4.17. Deviation of exchanged power on the line
in the first simulation case

Figure 4.18. The target function for the first simulation scenario
Table 4.3. Comparison results are based on several control criteria
in the first simulation case
Standard
IAE
ISE
ITAE
ITSE*10-3

PID
ACE1
ACE2
8.6040
9.0902
0.4119

Figure 4.19. Simulation results for the second scenario
(a) Change the load in the first sector
(b) Variation of frequency deviation in the first region


17

Figure 4.20. Objective functions for the second simulation
Table 4.4. Quality comparison of controllers based on two control
standards IAE and ISE for the second simulation case
Standard
IAE
ISE

PID
39.1473
6.7665

NARMA
24.9040
2.8997

MRAC
24.6673
2.8608

The simulation results show superior efficiency of NARMA and MRAC
neuron controllers compared to PID controllers.
4.4.2.2. Simulation results for MPC controller


0.6432

4.5. Conclusion chapter 4
The simulation results using MATLAB / Simulink software have
demonstrated the superiority of the above three controllers over the
classic controller like PID.
When evaluating and comparing each LFC controller using artificial
neural network, some comments can be made as follows:


19
- NARMA-L2 controller has faster network training time because
there is no need to identify the control object, when the control object is
linearized.
- The MRAC controller requires two processes: controller object
identification and neural network training for the controller.
- The MPC controller only needs the process of identifying the control
object, but because MPC is the predictive controller, it takes a lot of time
to run the simulation.
The results are published in the work number [CT2], [CT3] in the
published list of scientific works of the thesis.
CHAPTER 5
ANALYZING AND EVALUATING THE EFFICIENCY OF
SOLUTIONS FOR INTELLIGENT CONTROL OF TURBINE SPEED
IN HYDROELECTRIC PLANTS

5.1. Question
5.2. Synthesize and analyze control solutions for single-area and
interconnected area hydropower plants
5.2.1. Schematic diagram of a single-area hydroelectric plant using



21

Figure 5.10. Mechanical power
deviation for zone 1

Figure 5.11. Mechanical power
deviation for zone 2

Figure 5.12. Speed deviation for zone 1

(a) Speed deviation

(b) Speed deviation

Figure 5.13. Speed deviation for zone 2

(a) Power deviation

(b) Power deviation

Figure 5.14. Power deviation on line by area 1.2
- In area 1 (zone 1).


22

(a)


MRAC
0.40091
0.40094

NARMA
0.26399
0.26401

Remote control
PID
FLC
MPC
0.43384 0.41620 0.342361
0.51285 0.41622 0.34238

Table 5.2. Comparison of the controllers based on ITAE quality
criteria for line interchange power deviation between two zones
Remote controll
Comparison
criteria P12_ MRAC P12_ NARMA P12_ PID P12_ FLC P12 _ MPC
ITAE
0.314738
0.32221
4.154547 0.32321 0.270638

5.3. Conclusion chapter 5
In this chapter, the thesis has synthesized, simulated, and compared
different solutions applying fuzzy logic controller and neural network
designed in the previous chapters for single-zone hydropower and 2-zone
linkage.