Nâng cao chất lượng điện năng dùng mạch lọc tích cực dạng lai ghép sử dụng phương pháp điều khiển thông minh tt tiếng anh

MINISTRY OF EDUCATION – MINISTRY OF TRANSPORT
HO CHI MINH CITY UNIVERSITY OF TRANSPORT

CHAU VAN BAO

IMPROVING THE POWER QUALITY
USING THE HYBRID ACTIVE POWER FILTER
BY INTELLIGENT CONTROL TECHNIQUE

Major: Control Engineering and Automation
Code : 9520216

SUMMARY OF DOCTORAL DISSERTATION

Science supervisor:

1. Assoc., Dr Vo Cong Phuong
2. Dr. Chau Minh Thuyen

HCM CITY - 2019


The works have completed in: Ho Chi Minh City
Scientific supervisor:

1. Assoc., Dr Vo Cong Phuong
2. Dr. Chau Minh Thuyen

Reviewer 1: ……………………………………………….
Reviewer 2: ……………………………………………….
Reviewer 3: ……………………………………………….

The dissertation will be defended at the school-level thesis reviewing
council at:
Ho Chi Minh City University of Transport.
…………..at ………. hr, on…………………………………………..

The thesis can be found at the library:
- Library of Ho Chi Minh City University of Transport


1

INTRODUCTION
1. Reasons for choosing the topic
Along with the development of industry, the loads are increasing and
the majority of nonlinear loads are the cause of harmonics. Harmonics
cause a lot of harmful problems for electrical systems and electrical
devices, this is the cause of poor power quality.
Today, power quality issues are very much concerned by many
countries in the world. One of the methods to eliminate harmonics,
reactive power compensation Q in the electrical system is using an active
filter circuit (APF). APF has the advantage of working online with
electrical systems, no resonance occurs, regardless of the feature of the
load. However, its capacity is limited, its working efficiency is not high
and it is not used in medium and high voltage electrical grids.
Currently, in our country often use the static compensation capacitor
to improve power quality. However, the method of using capacitor is
ineffective, because only compensating Q without canceling harmonics is
the nonlinear load. In order to solve these problems, the hybrid active
power filter (HAPF) model is a necessity, it can compensate for the

integration of different harmonic sources and solve disadvantages of the
capacitor. Therefore, research on design, calculation and control for HAPF
has an important meaning contributing to improving the working
efficiency of filter circuit and improving power quality. Therefore, the
topic: "Improving the power quality using the hybrid active power filter
by intelligent control technique" is necessary.
2. Research purposes
− Theoretically: Find out the method of determining harmonic currents
more accurately; Determine HAPF parameters by multi-objective
optimization algorithms in considering the stability of the system; Find
out the new control method for HAPF so that it minimizes errors, reduces
transient time; Find out the new DC bus voltage stabilization method.
− Application: The results of the thesis can be applied to construction
of hybrid active power filter models to compensate reactive power Q and
eliminate the harmonics in the electrical system.
3. Object and scope of the research
− Research object: The study was conducted on HAPF model and
applied to low voltage grid.


2

− Scope of research: Only research to improve the power quality in
terms of total harmonic distortion (THD) and compensating reactive
power Q.
4. Research tasks
Using the methods, calculation, data and results of previous studies as
a basis for research and evaluation. Since then: Improve the p-q harmonic
detection method; Determine multi-objective optimization of HAPF
parameters; Control methods for HAPF; DC bus voltage stabilization

method. Application of Matlab software to simulate the above problems.
5. Research Method
− Analysis of harmonic detection methods, thereby improving its
shortcomings by improved harmonic detection method with more
accuracy and wider application scope. Analyse methods of determining
HAPF parameters. From there, propose a multi-objective optimization
method to determine HAPF parameters. Analysis of DC bus voltage
control methods, from which draw defects and give a method to stabilize
DC bus voltage in the direction of adaptive control. Provide control
strategies and control methods to solve problems such as wide application
range, flexibility, efficiency in filtering harmonic and reactive power
compensation.
− Use Matlab to simulate for methods.
6. The scientific and practical significance of the thesis
− Scientific significance: The thesis is a scientific work of theoretical
and practical significance, contributing to systematizing and clarifying
problems of harmonic filtering. From that, proposes the method of
determining harmonics, the method of determining parameters of HAPF,
DC bus voltage stabilization method and HAPF control methods to
improve power quality.
− Practical significance: The thesis has evaluated the situation,
demonstrated out the advantages and disadvantages of the harmonic
filters. The thesis is quite comprehensive and systematic, with practical
significance to the issue of improving power quality.
7. Structure of the thesis
The thesis consists of 143 pages and the order of parts is as follows:
Introduction; content (including 6 chapters); conclusions and suggestions;
list of published scientific works related to the thesis (including 10 papers
and 01 applied scientific research); there are 119 references and
appendices.

3

Chapter 1:
OVERVIEW OF FILTER
1.1. Issues of power quality
Non-linear loads are the cause of harmonics, which reduces power
quality. Harmonics cause many different problems in both the grid and the
load such as: overheating equipment, overheating transformers, deviation
control devices, power factor of the load decreases, causing losses in the
electrical system, increasing the cost of the customer and affecting the
stability of the grid. Therefore, power quality has become an increasingly
important issue for Electricity and electricity consumers.
1.2. Power quality
Power quality is a problem related to voltage, current, frequency,
causing electrical equipment to operate abnormally or damaged.
Total harmonic distortion (THD): 𝑇𝑇𝑇𝑇𝑇𝑇 =

2
�∑∞
𝐻𝐻≠1 𝐼𝐼ℎ

𝐼𝐼1

. 100%

(1.1)

1.3. Effect of harmonics on power quality

Although the sinusoidal source voltage is not distorted, but the
nonlinear load causes harmonics and undesirable effects on power quality
such as increased line losses, changing the voltage on the grid and grid
frequency.
1.4. Methods for harmonic filtering
1.4.1. Passive Power Filter (PPF)
This is a common solution to remove harmonics in electrical systems.
PPF is the simplest solution to minimize harmonics [27], [32], [36], [40].
PPF has a simple structure consisting of the three elements R, L, C. It is
low cost, easy to implement. However, it has disadvantages such as easy
resonance, instability, low reliability.
1.4.2. Active Power Filter (APF)
From the disadvantages of PPF, the APF was born to overcome the
disadvantages of PPF, it is very effective in improving power quality, it
has advantages such as flexible compensation, no dependent on property
of load, high efficiency, no occurs resonance with grid impedance. APF is
widely used to compensate Q and harmonic filtering [7], [25], [91], [96].
The basic principle of APF is based on harmonic currents of the load to
create a harmonic signal to compensate on the grid. However, the
disadvantage of APF is its high cost, low capacity, and difficult to apply
to high-voltage grids.
1.4.3. Hybrid Active Power Filter (HAPF)


4

To improve the efficiency of APF, the HAPF model was born and
developed [16], [26], [42], [62], [79]. HAPF's structure is a combination
of PPF and APF. Therefore, it has the advantages of both APF and PPF.
The most outstanding advantage of HAPF is its ability to work at high

voltage and high power grids with a relatively small capacity of APF.
Chapter 2: HARMONIC CURRENT DETECTION METHOD
2.1. Introduction
There are many methods of determining harmonic currents of
nonlinear loads such as: using low-pass, high-pass filter circuits [13], it
has the disadvantage of slow response and just a small change in frequency
will make these filters ineffective. The most common method is the p-q
harmonic detection method [17], [89], [104]. It has the advantage of being
simple and easy to implement. However, it also has the disadvantage of
slow response to fast changing loads and large amplitudes [29-30]. In this
chapter, we propose an improved method of the p-q harmonic detection
method using the fuzzy controller integrated into the pq method to
automatically adjust the DC components of P and Q to close to the desired
value, keeping the amplitude of the source current is not overshot when
the load changes large and the transient time is reduced.
2.2. p-q and i p -i q harmonic current detection method
2.2.1. The transformation from a-b-c coordinate system to α-β
coordinate system
The transformation from a-b-c coordinate system to α-β coordinate
system is implemented by Clarke [97].
2.2.2. p-q harmonic detection method
p-q harmonic detection method is proposed by Akagi [7], in Figure 2.2.
ua
ub
uc
iLa
iL b
iLc

C32

uα
uβ

iα
C32 iβ

p

C pq

q

LPF
LPF

p
q

−1
C pq

iαf
iβ f

iLaf
i +∑

C23 iLbf
Lcf

-

∑

+

- ∑
+

iLha
iLhb
iLhc

Figure 2.2. Principle diagram of p-q method
The harmonic components determinated are:
iaf 
 
1
−1  p 
C23C pq
ibf  =
q 
2
3
U
 
1
i 
 cf 

(2.13)

= i La − i Laf
i Lah

= i Lb − i Lbf
i
 Lbh
i = i − i
 Lch Lc Lcf

(2.14)


5

2.2.3. i p -i q harmonic detection method
e

ia
ib
ic

sinωt
− cosωt

PLL

iα

C 32

iβ

ip

LPF

ip

iαf

C
iq

LPF

C 23
iq

i βf

iaf
ibf
icf

-

+

+

iah
ibh
ich

+

Figure 2.3. i p -i q harmonic detection method
The fundamental components are:
iaf 
uα
 
1
C23 
ibf  = ∞
uβ
i  3∑ U n2
 cf 

uβ   p 

− uα   q 

(2.24)

n =1

Fuzzy

adjustor

Fuzzy
adjustor

2.3. Improved p-q harmonic detection method
To
improve
p
∆p
p +
+
∑
LPF
overshoot and reduce
+ ∑
d
the dynamic response
Kp
dt
time of p-q method.
−1
C pq
q +
q
∆q
The improved p-q
+
∑
LPF

+ ∑
d
harmonic detection
Kq
dt
method is proposed
in Figure 2.16.
Figure 2.16. Improved p-q harmonic
detection method
iLaf 
 p + Kp 

1
−1
C23C pq


iLbf  =
2
U
3
1
 q + K q 
 
iLcf 

The formula (2.13) is rewritten: 

iαf
iβ f

(2.27)

2.4. Simulation results
Table 2.4 and Table 2.5 compare the response of p and q in the p-q
method and the improved p-q method.
Table 2.4. Response of p
During the period (0÷0.2s)
During the period (0.2s÷0.4s)
Transient time Overshoot Transient time Overshoot
p-q method
0.05s
2.17%
0.05s
3.4%
Improved
0.016s
0.3%
0.025
0.5%
p-q method


6

Table 2.5. Response of q
During the period

During the period

(0÷0.2s)
(0.2s÷0.4s)
Transient Overshoot Transie Overshoot
time
nt time
p-q method
0.025s
20.2%
0.04s
21.13%
Improved p-q method
0.02s
0.42%
0.02s
2%
From the above results, we find that: The improved p-q harmonic
detection method has a shorter transient time, reducing the overshoot is
smaller than the p-q method. This has great implications for the stability
of the system.
Chapter 3: MULTI-OBJECTIVE OPTIMIZATION DESIGN FOR
HYBRID ACTIVE POWER FILTER
3.1. Introduction
Currently, the parameters of HAPF are mostly determined based on
basic formulas such as studies [24], [70], [98]. Therefore, the achieved
results may not satisfy the system stability condition. Multi-objective
studies such as Gen algorithm application for PPF design [20], [43]; using
the PSO algorithm [18], [95] for PPF design. In summary, previous multiobjective studies mainly computed for PPF, and APF parameters had little
research and multi-objective optimization studies without considering the
stability of the system.
To overcome this drawback, in this chapter, we perform a stable

analysis for HAPF to find the stability of the system. Then, use the SSA
multi-objective optimization algorithm to determine the best set of
parameters for HAPF.
3.2. Stable analysis for hybrid active power filter
Control block diagram of HAPF is shown as Figure 3.3.
I Lh

−1

+

X

Gc ( s)

Ginv (s )

Uinv

+

Gout (s )

−

+

X

I sh

I Fh

Figure 3.3. Control block diagram of HAPF
Transfer function of the load harmonic current I Lh according to the
supply harmonic current signal I sh :
I
1
G (s ) =
=
I
1 + G ( s ).G ( s ).G ( s )
(3.4)
sh

Lh

c

inv

out


7

From (3.4), the characteristic equation of the control transfer function:
D( s ) = a0 s 6 + a1s 5 + a2 s 4 + a3 s 3 + a4 s 2 + a5 s1 + a6 s 0 + a7

(3.5)

In order for the system to be stable, the formula (3.6) must be satisfied.
3.3. Multi-objective optimization design a1a2 − a0 a3 > 0
b a − a b > 0
for HAPF
 0 3 1 2
(3.6)
− System stability constraints:
b1b2 − b0b3 > 0
The HAPF system is stable when the 
c b −b c > 0
0 3 1 2
conditions in Equation (3.6) are satisfied.
c1c2 − c0 c3 > 0

− Constraints on resonance conditions in PPF: L and C values in a
branch must resonate at a certain frequency.
ωn L =

1
ωn C

− Constraints of R, L, C: Values of R,
L, C must be positive and satisfy the
condition (3.8) and resonance condition.
Begin
Enter upper and lower limits:
CF , C1 , L1 , R1 , L0 , C0 , Udc , Kp , Ki

No

Yes
Estimates fitness
Create vibrations

Move location

No

(3.7)

0 < Li ≤ Lmax
0 < Ci ≤ Cmax

(3.8)
The values of R max , L max and C max are
determined according to the formula (3.6).
− Maximum capacity compensated by
PPF but not over-maximized.
Qb min ≤ Qbi ≤ Qb max

Initialization
Spider size and position

Stability test

0 < Ri ≤ Rmax

THDis ≤ ε1

Q b min ≤ Q bi ≤ Q b max

Error ≤ ε2

Yes
End

Figure 3.4. SSA algorithm
flowchart
3.4. Simulation results

(3.9)
− Constraint on the value of DC bus
voltage:
U AC < U DC < U DC-max (3.10)
where: U AC is the AC voltage at the output
of the inverter.
− Constraint of controller parameters:
Parameters of controller must be positive
and satisfy the system stability condition
(3.6).
0 < K p < K pmax
0 < K i < K imax
(3.11)
Objective function:
min THDis
max Qbi
min Error

(3.14)

8

3.4.1. Traditional design
According to the article
[24], [46] we have the
parameters given in Table
3.2. Figure 3.6 shows the
waveforms in the traditional
design. The THD of i s
decreases from 27.65% to
1.897%, while Q decreases
from 4820VAr to 1490VAr,
which means Q compensated
is 3330VAr. Compensation
error
in
steady-state
decreases to ± 8A.

Figure 3.6. The waveforms in steady-state
of the traditional method

Table 3.2. HAPF parameters with traditional design methods.
CF
C1
L1
R1
L0
C 0 U DC THDi s Q bΣ Error

(µF) (µF) (mH) (Ω) (mH) (µF) (V)
(%)
(Var) (A)
116.8 349.2 29.77 0.01 0.2
80 535 1.963 3330
±8
3.4.2. Multi-objective optimization method using SSA
The multi-objective optimization method will find all HAPF parameters
including power circuit parameters and control circuit parameters.
Table 3.4. HAPF parameters with SSA method.
CF
(µF)
158,8

C1
(µF)
412,3

L1
(mH)
24,89

R1
(Ω)
0,017

From Figure 3.8,
THD of i s decreases
from 27.65% to
0.83%,

while
capacity Q decreases
from 4820VAr to
790VAr,
ie
the
compensation
capacity is 4030VAr,
the
compensation
error decreases from
± 100A to ± 3A.

L0
(mH)
1,2

C0
(µF)
61,6

U DC
(V)
785,3

Kp

Ki

30,6

0,15

THDi s
(%)
0,83

Error
(A)
±3

Figure 3.8. The waveforms in steady-state
of the SSA method


9

3.5. Conclusion
In this chapter, a new approach in multi-objective optimization design
for HAPF was provided. This approach allows us to calculate all the
parameters of both the power circuit part and the control circuit part of
HAPF. The achieved results are globally optimal and satisfying the system
stability condition. This study has practical implications in determining all
HAPF parameters that contribute to improving power quality in the
electrical system.
Chapter 4:
DC-BUS VOLTAGE STABILITY FOR HAPF
4.1. Introduction
This chapter presents an overview of the methods of stabilizing the DC
bus voltage used for HAPF. On that basis, a DC bus voltage stabilization

method is proposed. The simulation results demonstrate that the proposed
method has better results in reducing voltage ripple on the DC-bus,
compensation error and total harmonic distortion of the supply current in
steady-state. Especially, this method is able to stabilize the DC bus voltage
when the load changes.
4.2 Overview of DC bus voltage stability for HAPF
The paper [46] provides a method for stabilizing DC bus voltage using
fuzzy logic applied on two models: TLSC (Three-Leg Split-Capacitor)
and FLI (four-leg inverter). The paper [63] studied the DC bus voltage
stability for HAPF in series using the PZP (Pole-Zero Placement) method.
Research [108] provides a closed-loop numerical control algorithm to
control DC bus voltage for three levels APF using the space vector
modulation method. The paper [110] analyzed the effect of the DC bus
voltage controller to stabilize DC bus voltage and the compensation
efficiency of three-phase four-wire active filter circuit. The paper [69]
provides a DC bus voltage control method for APF using a PI controller
combining fuzzy controller.
In summary, all studies on DC bus voltage stability for HAPF only
focus on stabilizing the DC bus voltage for APF and in case the load does
not change. In this chapter, a new DC bus voltage stabilization method is
proposed, which can maintain the DC bus voltage even when the load
changes.
4.3. Analysis of DC bus voltage variation in HAPF system
According to [3], there are two main reasons that the DC bus voltage
change: the source voltage is not ideal and the load is nonlinear.


10

− Considering the case where the source voltage is not ideal: Suppose

the ideal source voltage is in the form:
=
u (t ) U m sin(ω0 t + ϕ )
(4.1)
When voltage sag happens, the supply voltage in the form:
=
u (t ) ηU m sin(ω0 t + ϕ + 1800 )
(4.2)
The hybrid active power filter topology is shown in Figure 4.1.
Source

Zs
Non-linear
load

Us
Rectifier

CF

C1

L0

iDC (t )

uDC (t )

L1

ia (t)

Inverter

N :1

C

C0

Figure 4.1 The hybrid active power filter topology
Single-phase equivalent circuit of the injection circuit in Figure 4.2.
i
We have u 1 is expressed as:


CF
C1




u
L1



ηU m  1  −δ t

 e sin(ωt + α )

−
u1 =
sin α  1 + C1 
u1

(4.3)

CF 

From the expression (4.3), we can see that:
when the voltage sag suddenly occurs, the
R
voltage maintained on the fundamental
Figure 4.2 Singlefrequency resonant circuit L 1 -C 1 -R will
phase equivalent
decrease exponentially, if this voltage is not fast
circuit of the
reduced, the capacitor will be charged and make
injection circuit
the DC capacitor voltage fast increase.
− Considering the case of load is nonlinear:


11

Suppose the voltage applied to the fundamental frequency resonant
circuit is u shn and the voltage at the inverter AC side output is u ahn . If
considering the ratio transformer is 1: 1, we can replace the output of the
fundamental frequency resonant circuit to the inverter output by an
inductance L. According to figure 4.1, we have: ushn = U shn sin(nωt ) and

=
uahn U ahn sin(nωt − θ n )
The average harmonic active power flowing into APF can be expressed
as:

U shn .U ahn .sin θ n
(4.6)
2nω L
From (4.6), we can see that: if 00<θ n <1800 then Phn > 0 , and the
active power of the nth harmonic transmits to the AC-side from DC-side
of the inverter, and as a result, the DC bus voltage will increase. If 1800<θ n <00 then Phn < 0 , and the active power of the nth harmonic
transmits to the DC-side from AC-side of the inverter and as a result, the
DC bus voltage will drop.
4.4. Proposed DC bus voltage stabilization method
The structure of the method is shown as in Figure 4.3.
Phn =

Lb

Rectifier
Us

Zs
C1

-

D

Ib

+

If

Boost
S

U C1

G1
+

C-

Inverter
G3

G5

G6

G2

L0

Rb
G4

C0
U DCact

+
-

P
I

K

Ifa,b,c

+

-

U DCref

Figure 4.3 Proposed DC bus voltage stabilization method
Three-phase balanced power supply through the three-phase
unbalanced bridge rectifier to generate the DC voltage U b , this voltage
oscillates from 1.5 to 1.73 times the amplitude of the source voltage. To
reduce the voltage and current variations, we add a capacitor C b and
inductance L b at the output of the rectifier. The voltage across the capacitor
C b is the input voltage of the Boost converter. After passing the Boost
Converter, a DC voltage will be generated (voltage on the C capacitor). It


12

is greater than the voltage on the C b capacitor and with an output voltage
ripple is very small. In order to stabilize the voltage on the C capacitor,
we must control the S switch to always keep a fixed voltage at the DCbus. According to the above analysis, if 0 < θ n < 1800 then the power is
transferred from the AC-side to the DC-side of the inverter, the voltage
across the DC-bus is greater than the reference value, then the S switch
will close, the energy on the capacitor will is discharged through R b and
the voltage on capacitor C is reduced. If the voltage on the capacitor C is
less than the reference value, then switch S will open. At this point, the
capacitor C will be charged from the Boost circuit.
4.5. Simulation results
Table 4.1 Parameters of the HAPF
Us
f
CF
C1
L1
C0
L0
C
(V)
(Hz) (µF)
(µF)
(mH) (µF) (mH) (µF)
220
50
120
349.2 29.77 690
0.2 10000
In research [3], using R b =1.5Ω. Parameters of the PI controller i s

K p =10 and K i =0.1.
iL
200
0
-200
is
200
0
-200
error
100
0
-100
800

UC

400
0

0

0.1

0.2
Time (s)

0.3

0.4

Figure 4.4 Waves with the methodology of the paper [3].
Table 4.2 Parameters of the HAPF with the proposed method.
CF
C1
L1
C0
L0
C
Cb
Lb
(µF)
(µF)
(mH)
(µF) (mH)
(µF)
(µF) (mH)
120
349.2 29.77
690
0.2
10000
10
4

Rb
(Ω)
1.5

13

The simulation result of the proposed DC bus voltage stabilization
method are shown in Figure 4.5.
Table 4.3 Comparison of the effectiveness of the proposed method and the
methodology of the paper [3].
Methods
Paper [3]
Proposed

ΔU C /U C
5%
1%

Before load is changed
THDi L
THDi s
28.06%
1.69%
28.06%
1.61%

Sai số
±10 A
±5 A

ΔU C /U C
5.49%
1.02%

After load is changed
THDi L
THDi s
30%
1.31%
30%
1.08%

Sai số
±10 A
±5 A

iL
200
0
-200
is
200
0
-200
error
100
0
-100
UC
800
400
0
0

0.1

0.2
Time (s)

0.3

0.4

Figure 4.5 Waves of the proposed method
4.6. Conclusion
In this chapter analysed the cause for changing DC bus voltage and an
overview of DC bus voltage stabilization methods. On the basis of that, a
new DC bus voltage stability method for the HAPF was proposed. The
results of the simulation proved that: The proposed method is capable of
stabilizing DC bus voltage with an output voltage ripply on the DC bus is
very small.
Chapter 5:
CONTROL METHOD FOR HAPF
5.1. Introduction
This chapter gives an overview of the control methods used for HAPF.
From there, the author proposes two control methods for HAPF. The first
is using a PI-fuzzy controller and the second is using adaptive HysteresisFuzzy Neural controller.
5.2. Mathematical model of HAPF


14
IS

From figure 5.1, we can

calculated I s as folows:
U − K U + (K + Z )I
(5.1)
I =

IF
ZF

ZS

IC

Z0

S

US

IL

1

I1

Z1

UC

C

2

F

L

K2 + ZF + ZS

S

with:
Z 0 Z1

Z1

;K =
K =
Figure 5.1. Single-phase
Z +Z
Z +Z
equivalent circuit of HAPF
5.3. Control strategies analysis for HAPF
Table 5.1 Parameters of the HAPF
CF
C1
L1
L0
C0
Ls
88 μF

349,2 μF 30mH 0.5mH 21μF 0.2mH
5.3.1. Control strategy based on I F : U C = KI F = K(I S - I L )
1

2

0

1

2

2

1.5

1.5

Rs
0.01Ω

0.5

0.5

0
400

0
400

300

1.5

200

1

100
0

1

100

0.5
0

K

2

300

2
1.5

200

pi

0.5
0

K

0

pi

(b) f =150Hz

(a) f =100Hz
2

2

1.5

1.5

1

P

1

P

1

1

P

P

1

0

0.5

0.5

0
400

0
400
300

2

300

1.5

200

K

2

0

1

100

0.5
0

1.5

200

1

100

0

K

pi

0.5
0

pi

(c) f =250Hz
(d) f =350Hz
Figure 5.2. Influence of ILon Is
5.3.2. Control strategy based on I S : U C = KI S
5

15

4

10

P

P

3
2

5
1
0
400

0
400
300

2
1.5

200

300

0

2

pi

(a) f =100Hz

1

100

0.5
0

1.5

200

1

100
K

K

0.5
0

0

pi

(b) f =150Hz


15

2

6
5

1.5

3

P

P

4
1

2
0.5

1
0
400

0
400
300

300

2
1.5

200

2

0

0

1

100

0.5

K

1.5

200

1

100

0

K

pi

0.5
0

pi

(d) f =350Hz
(c) f =250Hz
Figure 5.6. Influence of I L on I s
5.3.3. Control strategy based on I L : U C = KI L
30

20

25

15
20
P

P

10

15
10

5

5
0
400

0
400
300

300

2

2

1

100

0.5
0

K

0

1

100

0.5
0

K

0

pi

pi

(b) f =150Hz

(a) f =100Hz
40

80

30

60

20

40

P

P

1.5

200

1.5

200

10

20

0
400

0

400
300

2
1.5

200

1

100
K

300

2

0

0

1.5

200

0.5

1

100

pi

K

0.5
0

0

pi

(c) f =250Hz
(d) f =350Hz
Figure 5.10. Influence of IL on Is
From the above analysis, we can be seen that: control strategy U C =
KI L . In Figure 5.10 has a relationship between K and η is linear, when K
increases then η increases and vice versa, this is a best control strategy.
5.4. Proposed control method for HAPF
5.4.1. PI-fuzzy controller for HAPF


16

First, K P and K I of PI controller are calculated offline and do not
change during control. The fuzzy controller will produce ΔK P and ΔK I
values, according to which the K P and K I parameters of the PI controller
will be adjusted to match the load change:
 K P−new = K P−old + ∆K P

 K I −new = K I −old + ∆K I

(5.15)

The inputs of the fuzzy controller are e(t) and Δe(t): e(t)=i Lh -i Fh and
Δe(t)=e(t)-e(t-1). Membership functions of the inputs and outputs in
Figure 5.19.
nb

nm

ns

zo

ps

pm

- 0. 6

- 0.3

0

0. 3

0. 6

pb

nb

nm

ns

- 0.6

- 0.2

zo

ps

pm

0

0.2

0. 6

pb

∆ et()
-1

1

e (t )

∆ KI
-1

1

∆ Kp

Figure 5.19. Membership functions of the inputs and outputs
Table 5.2 Fuzzy Rules of ΔK P

Table 5.3 Fuzzy rules of ΔK I

Simulation results

Figure 5.21. Simulation results with the PI controller


17

Figure 5.24. Simulation results with the PI - mờ controller
Table 5.5 Result comparison table of the PI controller and PI-fuzzy
controller.
Before load
After load
changed
changed
THD i L
28,01%
39,24%

THD i s
5,71%
7,72%
PI controller
error
± 12A
± 20A
cosφ
0,94
0,94
THD i L
28,01%
39,24%
THD i s
1,56%
2,29%
PI-fuzzy controller
error
± 4A
± 5A
cosφ
0,98
0,98
5.4.2 Control method for HAPF using the Adaptive Hysteresis – Fuzzy
- Neural controller
−iLh ( s + 1)

−iLh ( s )

+

e ( s)
-

i Fh (s +1 ) Identification
-X +
and prediction
model
u(s)
Fuzzy - neural
Gout
inverter
controller
Cost
function

∆e

K
Hysteresis controller

Figure 5.4. Structure of the proposed control method

iFh (s)


18

The structure of the HAPF control method using adaptive Hysteresis –
Fuzzy-Neural controller is shown in Figure 5.4. There are two proposed

control modes for HAPF, one is the Hysteresis control mode and the other
is the adaptive Fuzzy-Neural control mode. The choice between the two
modes is determined by the switch K.
+ If |e(k)| > threshold value, select Hysteresis control mode.
+ If |e (k) | ≤ threshold value, select the adaptive Fuzzy-Neural control
mode.
−U khi e(k ) < V
− Hysteresis controller:

L

u  0 khi VL ≤ e(k ) ≤ VH
=
 U khi e(k ) > V
H


(5.25)

− Identification and prediction model
I Fh (s )

TDL

IW1,1

u (s )

TDL

IW1,2

LW2,1

1

b

I Fh ( s + 1)

1

1

b

2

Figure 5.5. Structure of the neural network
Control method using neural network model [93].
=
M

1
2

N2

∑ (− I

1
(k + j ) − I (k + j )) 2 +
2

Nu

∑ λ [ ∆u (k + p − 1)]

2

(5.26)

Lh
Fh
j N=
j 1
1

− Adaptive Fuzzy-Neural controller
+

The nodes at the input layer: O = e.ω , i = 1, 2

+

The nodes at the fuzzy layer

+

1

1

1i

1
O2 = ∆e.ω1i , i = 1, 2

 (O1 − c )2 
Oijfuzzy layer = µ Aij = exp − i 2 ij ; i = 1,2 ; j = 1,2,...7

2σ ij 


(5.27)

(5.28)

The nodes at rules layer
Okrules layer =

2

∏µ

j
Ai

, i = 1,2 ; j = 1,2,...7; k = 1,2,...,49.

(5.29)

i =1

+

The nodes at neural network

∑ O
=u =
∑ O
m

Oneural network

rules layer

k =1
m

k =1

k

ω4 k

rules layer

; m = 49

Cost fuction J is defined as follows:
2
1 n
=
J
( −iLh (k + 1) − iFh (k + 1) )
2 i=1
Reverse propagation method to update parameters:

∑

(5.30)

k

(5.31)


19

∂J
+ α [ω ( s ) − ω ( s − 1)]
ω 4 k ( s + 1) = ω ( s ) − η k
∂ω4 k


∂J
+ α [cij ( s ) − cij ( s − 1)]
cij ( s + 1) = cij ( s ) − η k
∂cij



∂J
+ α [σ ij ( s ) − σ ij ( s − 1)]
σ ij ( s + 1) = σ ij ( s ) − η k
∂σ ij


Differential coefficients ∂J , ∂J and
∂ω4 k ∂cij

∂J
∂σ ij

(5.32)

is determined as follows:

∂J
=− ( −iLh ( s + 1) − iFh ( s + 1) ) f ' (Okrulelayer ).Okrulelayer =δ k4 .Okrulelayer
∂ω4 k

(5.33)

rulelayer
4
'
)
With: δ k =−(−iLh − iFh ) f (Ok

Similarly, we have:
1
 ∂J
∂Oijfuzzylayer
∂J
fuzzylayer (Oi − cij ) fuzzylayer

=
=
δ
.Oij
k
∂cij
 ∂cij ∂Oijfuzzylayer
σij2

1
2
∂Oijfuzzylayer
 ∂J
∂J
fuzzylayer (Oi − cij )
=
=
δ
.Oijfuzzylayer

k
fuzzylayer
∂σij

σ3ij
 ∂σij ∂Oij

(5.34)

Simulation results

Figure 5.34 Waveform when using the Hysteresis – Fuzzy-Neural
controller when the system parameters do not change


20

− When the system parameters are changed (± 20% of initial value):

Figure 5.35. Waveform when using Hysteresis – Fuzzy-Neural controller
when system parameters change
− When the grid voltage is unbalanced, suppose the amplitude of
phase A voltage increases by 10% compared to phase B and phase C.

Figure 5.36. Simulation results with the proposed controller when the grid
voltage is unbalanced.


21

Figure 5.37. Simulation results with the proposed controller when the load
is changed.
Table 5.6. Compare the THD i s of the supply current in the case of using
different controllers.

Cases

Hysteresis
controller

THD i s %
Fuzzy neural
controller

Proposed
controller

System parameters do not
4.13
2.22
1.3
change
6.19
2.96
1.99
System parameters change
5.41
2.92
1.69
Grid voltage is unbalanced
5.04
2.34
1.73
Load is changed
From the simulation results in Table 5.5 and Table 5.6, we find that:

two proposed control methods have better response time, error in steadystate, smaller current total harmonic distortion and higher power factor
than PI controller.
5.5. Conclusion
In this chapter, an overview of control strategies for HAPF is provided.
The simulation results have clearly shown the effectiveness of each
control method. This analysis can be the basis for selecting a control
method for HAPF.


22

Chapter 6:
HAPF SYSTEM EXPERIMENT MODEL
6.1. Experimental model
To demonstrate the applicability of HAPF, an HAPF model was
experimented in the laboratory. Nonlinear load is modeled by a threephase uncontrolled bridge rectifier 30A/1000V with a load resistance of
12Ω/300W.
IGBT
SKM145GB123D
145A/1200V
module,
TMS320F28335 DSP Controller is the chip designed for control for power
converters, programmable controller with Code Composer Studio
software of Texas Instrument with clock frequency of 150MHz, There are
16 12-bit ADC channels with 80ns switching speed, 18 channels for 16 bit
pulse width modulation. Sensor circuit using LA 25NP current sensor and
voltage sensor LV 25P of LEM, optical isolation circuit using opto HCPL
– 3120.
The HAPF model is shown in Figure 6.8 and the waveform of the
supply current after compensation in Figure 6.12.

Figure 6.8. HAPF experimental Figure 6.12. Waveform of the
model
supply current after compensation
6.2. Conclusion
Some parts of HAPF have been implemented: nonlinear load model, 3phase inverter model using 6 IGBT, model of IGBT steering circuit,
central controller, harmonics and experimental model of HAPF system.
The experimental results on the model are not very good, because the
HAPF model is not complete, there is a lack of passive power filters so
the supply current has waveform near the sine (Figure 6.12). The control
algorithm used for the model is the PI algorithm, which does not apply
modern algorithms such as fuzzy or neural networks, since components of
the power block have not yet met the switching frequency of the control
block. The old measuring devices, not yet fully analyzed the parameters
of the power quality.


23

CONCLUSIONS AND RECOMMENDATIONS
1. Conclusions
The thesis presents the following issues:
- Overview of power quality, effects of harmonic components in the
electrical system.
- Overview of harmonic filtering methods and reactive power
compensation in the electrical system.
- Application scope of the method of determining harmonic p-q and
ip-iq. Since then, an innovative method to reduce overshoot of the p-q
method using the Fuzzy regulator has been proposed. This study also
contributes to improving the stability of HAPF system when the load

changes rapply and with a large amplitude. This is a good solution for
large capacity control applications.
- Multi-objective optimization study for HAPF using PSO and SSA
algorithms, these optimization algorithms have a new point that allows us
to determinate the parameters of both the power circuit part and the control
circuit part for HAPF in considering the stability of the system.
- Study of DC bus voltage stabilization methods for APF and HAPF. From
there, a new DC bus voltage stabilization method is proposed using the
Boost circuit and the energy discharge circuit through the resistor.
Compared to the DC-bus control method using a release circuit for the
Hybrid Active Power Filters, the simulation results demonstrate that the
proposed method has better results in reducing voltage ripple on the DCbus, compensation error and total harmonic distortion of the supply
current in steady-state. Especially, this method is able to stabilize the DC
bus voltage when the load changes.
- An overview of control strategy for HAPF. Accordingly, the control
strategy based on the harmonic current of the load is the most effective
control strategy.
- An overview of control methods for HAPF. Since then, the new two
control methods in considering the change of load: the first method uses
the PI-Fuzzy controller and the second method is the adaptive Hysteresis
- Fuzzy - Neural controller. Both of these methods have advantages such
as small compensation error, transient time is reduced and THD of supply
current is small.
In summary, the thesis has reviewed the issues related to power quality
and using the HAPF to improve the power quality in the system. Since
then, the author proposes five new scientific points to improve the power