Chapter-1

r
Introduction

pu
ar th
ag
Kh rig
y
IIT p

1.1 General
o

Owing to the wide spread use of adjustable speed drives (ASDs), switch mode power supplies
C

(SMPS), electric furnace and other pulse power (such as rail guns and aircraft launchers in all-
electric ships etc.) and non-linear loads, the power distribution networks (in a stand-alone or
large power systems) are facing huge problems to maintain stability and provide clean power to
the consumers [1-10]. The situation is further deteriorated due to connecting the renewable
energy sources into the systems [8]. As for example a wind turbine is subjected to different
mechanical modes related to the mechanical systems, such as the tower, the drive train, the
blades, the shaft and so on. The energy exchange and the interaction between the electrical and
mechanical system coupled through the generator are potential cause of resonance that generates
 2 Introduction

sustained low frequency oscillations in the system. In effect the quality of the electric power
deteriorates and several FACTS (Flexible AC Transmission System) devices are proposed to
overcome such problems.

Power quality problems include both voltage and current distortion in the power system. Voltage
distortion is taken care by the series type compensator, whereas the current distortion is

r
removed/reduced by the shunt type compensators. A combination of shunt and series (i.e. hybrid
is also available). A medium power active power filter mostly uses a voltage source inverter

pu
(VSI) to feed the right nature of compensating current. The VSI is operated in current controlled

ar t
mode to inject a right nature of compensating current to eliminate harmonics from the source
h
current. Note that the conventional passive filters are also used to mitigate unwanted harmonics

ag
and capacitor banks are employed to improve power factor of the system. Passive filters have the
Kh rig
advantages of low cost and losses, however they have the problems of harmonic resonance with
the source and/or the load. Moreover, they need to be tuned properly to take care of a wider
frequency range. Active filters may completely replace their passive counterpart. This requires
higher voltage/current switches for medium/high power applications. Fig.1.1 shows a typical
y

block diagram of a shunt active power filter. The filter injects harmonic currents into the load to
shape the source current to a sinusoid. The combined system of shunt passive and series active
IIT p

filters is proposed by Akagi and Nabae[11] and Peng[12]. It aims to act as a "harmonic isolator"
o

and present zero impedance to the external circuit at fundamental frequency and a high resistance
to source or load harmonics [13-18].
C

This thesis deals with the shunt compensation systems and in particular the shunt type active
power filters. Performance of the system (or the specific change that is necessary to also
compensate the reactive power demand) as a static VAR compensator (i.e. STATCOM) is also
discussed. Fig.1.2 shows one of the most common configurations of the active power filter.
 3 Chapter-1

r
pu
ar th Fig.1.1. Single line diagram of shunt active power filter.

ag
Kh rig
1.2 Aim and scope of the thesis

Power systems have to cope with severe load unbalance, reactive power demand and a variety of
non-linear loads which injects significant amount of harmonics. Passive, active and hybrid filters
y

are used to eliminate harmonics [11-104]. This thesis aims to investigate shunt active power
filters. The scope of thesis includes all the three most important areas of active power filters viz.
IIT p

(i) reference generation techniques,[11-65] (ii) tracking methods[66-86] and (iii) topological
developments[87-104].
o

The shunt type active power filters need to compensate the harmonic currents. Therefore, it is
C

important to generate the compensating-current-reference. Classical approaches include

synchronous reference theory and p-q-r theory based techniques. These methods work fine for
systems with no distortion in source voltage. However, artificial neural network (ANN) based
techniques work better for loads with severe unbalancing and with source voltage distortion. The
thesis presents an in-depth treatment on reference generation techniques.

To effectively eliminate the harmonics, it is important to track the reference current properly.
This thesis investigates both traditional PI controller and sliding mode controller based
 4 Introduction

techniques. Feedback linearization is applied for reduction of computational burden and practical
implementation of the controllers.

Depending on the type of non-linear loads, voltage and power level, availability of power
devices, performances indices such as maximum THD allowable, efficiency etc. and cost of the
equipment, the active power filters have several topological variations. This thesis investigates

r
some of them. Simple VSI based power filters as well as hybrid filters, where an active power
filter is connected in series with a passive filter (series L-C network to eliminate or reduce

pu
specific lower order harmonics) are studied. The passive filter in a hybrid filter is tuned at higher

ar t
harmonic (compared to the fundamental). Therefore, the passive filter also reduces the operating-
h
voltage-level for the active power filters. This is a huge advantage in case of high/medium

ag
voltage systems. However, such advantage is at the cost of dynamic performance of the active
Kh rig
power filter. A solution to this is to use a normal active power filter in parallel with a hybrid
filter. Such configuration is especially suited for STATCOM applications, where reactive power
is to be compensated in addition to the higher order harmonics.
y

Use of two hybrid filters in parallel has some unique advantages. Firstly two passive filters in
two hybrid filters eliminate most of the dominant lower order harmonics. Therefore, the load on
IIT p

the two active filters becomes minimal. This further allows the switching frequency of the two
inverters very different. With proper selection of switching devices, the inverter operating at low
o

frequency may compensate for the reactive power demand and compensate remaining lower
order harmonics, while, the other inverter operating at higher switching frequency may
C

compensate all the higher order harmonics to make the source current sinusoid. This thesis has
investigated in detail on the topological alternatives of shunt type hybrid (i.e. passive-active
combination) active power filters.
 5 Chapter-1

1.3 Literature survey

1.3.1 Survey of different reference generation techniques
Fast generation of the correct reference is one of the major areas to be explored in the field of
shunt active power filter. The reference generation technique may be broadly classified into two
categories. One works in time domain another deals with frequency domain techniques. Both of
these methods have their own advantages and limitations. Time domain methods are easy to

r
implement and are less sensitive to noise than their frequency domain counterparts. However

pu
major drawback of time domain schemes is delay in compensation. Traditional time domain

ar t
method employs low pass filter. Low pass filtering opts for a tradeoff between compensation in
h
time delay and accuracy. Implementation of this scheme also requires more number of

ag
transducers [16, 23-24]. On the other hand frequency domain schemes require less number of
Kh rig
transducers but involve delay in computation [25-28].
y
IIT po
C

Fig.1.2 circuit diagram of shunt active power filter

 6 Introduction

r
pu
ar th Fig.1.3. Few important reference generation techniques.

ag
Fig.1.3 shows a general classification of the important techniques available in reference
Kh rig
generation. Subsequent subsection discusses all the all major reference generation technique
briefly.

1.3.1.1 Time domain reference generation technique

y

A. P-Q Theory:
IIT p

So far most commercial APFs have been designed on the basis of reactive power theory to
calculate reference for harmonic mitigation. The instantaneous reactive power theory was
o

introduced by Akagi et al. in 1984 [11]. Fig.1.4 shows the block diagram. Advantages of this
method are:
C

• Easily implementable
• Excellent steady state performance with ideal supply voltage.
Instantaneous reactive power based approaches suffer from following drawbacks:

• Large number of voltage and current transducers are required,

 7 Chapter-1

r
pu
ar t
•
h Fig.1.4. Block diagram of instantaneous P-Q based reference technique.

This method is poor in compensation of harmonic current if source voltages are not

agsymmetrical.
Kh rig
• Performance is poor if APF is also required to compensate the negative and zero
sequence current in the load,
• Having delay in compensation.
y

The above limitations have been attended by reporting improved version of the reactive power
based approaches by several authors [12-15]. While most of the demerits are overcome, the
IIT p

challenge at present is to take care of unbalance and distortion in supply voltages [20-22].

B. Synchronous reference frame theory

o

This technique is popularly called as synchronous reference frame method and is reported in
C

[16, 22] and also in a different version (as stationary reference frame) in [24]. Here, the reference
frame d-q axes are determined by angle θ with respect to the α-β frame (as shown in Fig.1.5). All
positive sequence fundamental quantities will appear as DC in d-q frame and other harmonics
will appear as ripples. A low pass filter (LPF) is used to extract the fundamental component.
Fig.1.5 shows the scheme.
 8 Introduction

Advantages:

• This method is best suitable for harmonic compensation with sinusoidal source voltage.
Disadvantages:

• Large number of voltage and current transducers are required,

• Having delay in compensation.

r
pu
ar th
ag
Kh rig
y
IIT po

Fig.1.5 Block diagram of d-q based reference technique

C

Therefore, in general the Time Domain Method has the following advantages:

• Easily implementable and robust.

• Immune to wide-band noise and transducers distortion.
• Uses large number of voltage and current transducers.
• Compensation is poor in case of distorted supply voltage.
 9 Chapter-1

1.3.1.2 Frequency domain methods

Three widely used reference generation techniques are discussed here. These are Fast Fourier
Transform (FFT), Recursive Discrete Fourier Transform (RDFT) and Wavelet.

A. FFT based reference generation technique

r
This frequency domain method mainly utilizes the principle of Fourier analysis and can estimate
the harmonics very fast if a regular pattern for the current is available [25-28]. FFT is a quicker

pu
version of Discrete Fourier Transform (DFT). This estimation is carried out online after a delay

ar t
of a cycle of fundamental frequency. Fig.1.6 explains the scheme where i(0)….i(N) are the
h
discrete current samples and I(0),I(1)…..I(2),..I(N) are the corresponding Fourier coefficients and

ag
Φ1,. Φn are there respective phases.
Kh rig
Advantage:

• Easily implementable.
• Voltage transducers are not required.
y

Disadvantage:
IIT p

• Requires at least one cycle to estimate the reference current,

o

• Failed to estimate reference accurately if the current samples are corrupted with noise or
transient.
C
 10 Introduction

r
pu
ar th
ag
Kh rig
Fig.1.6. Block diagram of FFT based reference generation Technique
y

B. RDFT based reference generation technique

IIT p

RDFT uses the same principle as that of DFT, but introduces the concepts of sliding window as
shown in Fig.1.7 and also reported in [27]. In this reference generation technique, spectrum-
o

update is implemented in a step, larger than one sampling period to avoid over-sampling. For
active power filter, calculation of the harmonic components at each sampling period is not
C

necessarily required [104]. Estimation of the Fourier coefficient is carried out by the scheme by
an N-point wide sliding window which moves forward for p number of points in each step
(Where 1<p<N). Thus, estimation is carried out only with p number of samples those are new.
This is why the RDFT technique comes with the benefit of lower computational burden as
compared to FFT or DFT.
 11 Chapter-1

r
pu
ar th
ag
Kh rig
Fig.1.7. RDFT based reference generation technique.

Advantages:

• Voltage transducers are not required.

y

• Computational delay is less than FFT.

Disadvantages:
IIT p

• Failed to estimate reference accurately if the current samples are corrupted with noise or
transient
o

• Fidelity of the technique vastly depends on sliding window.

•
C

C. Wavelet Based Reference Generation Technique:

Wavelet may be used for extraction of fundamental component of load current [30, 36-37].
Different types of mother wavelets are used. The cut-off frequency of the MRA (multi resolution
analysis)-filters needs to be selected properly to reduce the sensitivity to power-frequency-
variation.
 12 Introduction

Advantages:

● Fast and accurate estimation of the fundamental component,

● The technique works satisfactorily even with distorted current waveforms,

● May be made insensitive to power frequency variation.

r
Disadvantage:

pu
● Performance of the system totally depends on the design of mother wavelet.
Drawbacks of Frequency Domain Schemes:

ar th
The limitations of the frequency domain approaches are that the designer has to consider the

ag
effect of aliasing. Anti-aliasing filter used for this purpose required to be very accurate otherwise
Kh rig
whole calculation will be erroneous. Sampling instant and zero crossing of fundamental are
required to be synchronized. Otherwise, phase estimated by this process will be erroneous. If
number of samples is not a power of 2, then zero padding is required. More over this method of
analysis is very susceptible to noise and transients.
y

1.3.1.3 Other methods

IIT p

Many other methods of reference generation technique have been reported. These include a
combination of passive filter and selective harmonic elimination for optimum performance.
o

Repetitive [69-71] control schemes have also been applied for achieving better performance.
C

These schemes are simple, easily realizable and hence are very attractive. However, their
performance is inadequate under severe unbalance in the load-side network. Recently there has
been increased effort to apply soft-computation based technique. Furthermore, combined GAs
and conventional analysis [61-65] and more advanced techniques such as Bacterial Foraging [60]
based compensation are reported. However, the major problem experienced is in the speed of
convergence. ANN based, adaptive filter based and fuzzy based reference generation technique
are also employed. These two methods are discussed briefly in the following subsections.
 13 Chapter-1

A. Evolutionary computing based reference generation technique

GA, ANN, bacteria foraging, particle swam optimization technique are the major constituent of
evolutionary computing based reference generation technique. ANN based reference generation
technique is most popular among the other evolutionary computing based reference generation
technique.

r
ANN based reference generation technique

pu
Artificial Neural Network (ANN) based reference generation techniques are successfully used

ar t
for shunt active power filter [39-51]. A typical adaptive neuron (Adaline) based reference
h
generation technique is shown in Fig.1.8. The ANN is so configured that the Fourier coefficients

ag
are mapped in terms of the weights. In most of the cases LMS algorithm is used for weight
Kh rig
training/adaptation.
y
IIT po
C

Fig.1.8. ANN based reference generation technique.

 14 Introduction

Advantage of ANN based system are:

• Its parallel computing power.
• Provides satisfactory results for the samples which are not much varied from training set.
Disadvantages of this method are listed below:

• Number of Adaline required to tune is equal to the number of harmonic considered in load

r
current, thus convergence is slow.
Generating input vector X=[coswt, sinwt……cosnwt, sinnwt]T is difficult and involves a

pu
•
tedious process.
•
ar t
Error being minimized by gradient based method has a chance to converge to local minima.
h
ag
B. Adaptive filtering method
Kh rig
y
IIT po
C

Fig.1.9. Adaptive filter based reference generation technique.

Adaptive filtering technique can be effectively used for reference generation of shunt active
power filter [29]. The adaptive filter based reference generation technique makes sine and cosine
functions from load current sample and no assumption is made on the harmonic content of the
 15 Chapter-1

voltage signal. Block diagram of this method is shown in Fig.1.9. Where η is learning rate of the
adaptive filter.

Advantage:
• This method of estimation also provides satisfactory result for distorted source voltages.
Disadvantages:

r
• This method is sensitive to variation in supply frequency.

pu
• This technique being gradient-based, may also converge to local minima.
As mentioned earlier this report consists of three soft-computing based technique. Three

ar t
methods are briefly described below.
h
ag
1.3.2 Survey on tracking methodology
Kh rig
Once the reference is generated, the VSI needs to track the reference current such that a correct
nature of compensating current is injected in to the PCC (point of common coupling). Many
controller strategies are employed to explore better performance of tracking [66-86]. A typical
block diagram for the controller of shunt active power filter is shown in Fig.1.10. Two different
y

controllers are necessary: one to maintain the DC link voltage of the VSI and another is to track
IIT p

the current reference.

o

Most popular techniques are PI controllers [66-68], standard hysteresis controller [104], one
cycle control [95-96], sliding mode control etc. [73-86]. Adaptive control [70] and negative
C

feedback based repetitive control [69-70] are also employed to improve the dynamic
performance and stability of the overall system. Sliding mode controller is more immune to
parameter uncertainties [73]. This feature encourages the researcher to apply sliding mode
controller (SMC) to mitigate tracking error in APF [74-86]. Singh et al. [85] applied sliding
mode on a three phase four wire system and SMC combined with fuzzy logic is reported in [84].
The SMC is used in [84] is to control the DC bus voltage. Furga et al. [83] reported an
improvement of the dynamic performance of the APF by applying SMC and a passive LC filter.
 16 Introduction

r
pu
ar th
ag
Kh rig
Fig.1.10. Control Block diagram of shunt active power filter.

Matas et al. [86] reported feedback a simple SMC, linearised by Tellegen’s theorem for single
phase APF. In [86] only one controller is reported.
y
IIT p

In this thesis an extension of the work reported by Matas et al. [86] is carried out applying SMC
to a three phase three wire system to study tracking and also to take up unbalance in the three
o

phase load. The feedback linearization based SMC is used for maintaining DC bus voltage.
Contrary to the available approach [86], this proposed controller utilizes two sliding mode
C

controllers, one for maintaining DC bus regulation and another for tracking the current reference.

1.3.3 Survey on converter topology

Industrial power converters and drives (IPCDs) are to be designed for minimum cost, losses, size
and with maximum reliability. IPCDs are basically nonlinear type loads those inject harmonics
into the system, while supplying the active power to the load. Introduction of strict legislation
such as IEEE519 limits the maximum amount of harmonics that a supply system can tolerate for
 17 Chapter-1

a particular type of load. Therefore, use of active or passive type filters is essential [87-104].
Passive filters have the advantages of low cost and losses, however they have the problems of
harmonic resonance with the source and/or the load. Moreover they need to be tuned properly to
take care of a wider frequency range. Active filter may completely replace the passive
counterpart. This requires higher voltage/current switches for medium/high power applications.
Use of hybrid filter, where a lower rating active filter is added in series with the passive filter,

r
has the merit of operating the active filter at a convenient voltage and current. Fujita and Akagi

pu
proposed such configuration in 1991 [87]. However, it required a transformer to couple the
passive filter with the active filter. Later the transformer is eliminated and Srianthumrong and

ar t
Akagi proposed a hybrid combination for application with diode rectifier [90]. Fig.1.11 shows
h
the configuration proposed. The passive filter connected in series is tuned at 7th harmonics and

ag
Kh rig
the active filter is operated at a much lower voltage (at 300V for a 3.3kV line). Bhattacharya et
al.[91] and Cheng et al. [92] proposed a dual hybrid configuration where the series filters are
tuned to eliminate 5th and 7th current harmonics. Fig.1.12 shows the proposed topologies.
y
IIT po
C
 18 Introduction

r
pu
ar th
ag
Kh rig
Fig.1.11. Hybrid (i.e. passive-active combination) type active power filter
reported by Akagi et al.
y
IIT po
C

Fig.1.12.a. Dual parallel topology reported by Bhattcharya et al. in 1997.

 19 Chapter-1

r
pu
ar th
ag
Kh rig
Fig.1.12.b. Dual parallel topology reported by Bhattacharya et al. in 1998.
y

Fig.1.12. Dual parallel topology

IIT p

Reduced switch topologies have the advantages of more reliability and less cost and complexity.
o

Reduced switch active power filter with one cycle control is proposed by Qiao and Smedley [95]
and Bala et al.[96]. The 3rd leg of the inverter is eliminated and the 3rd phase is connected to the
C

mid point. derived by voltage splitting capacitors) of the DC bus. This topology has the problem
of voltage balancing across the DC link capacitors. This is later improved by Wu et al where 3rd
phase is connected to the negative pole of the DC link [97]. A detail survey on topological
developments is available in [1-2].

Dual or multi converters [9-10, 98-102] are suitable for loads with higher degree of nonlinearity
or where a demand exists to supply the reactive power together with the harmonic compensation.
In this thesis an integration of passive and active reduced switch filters to optimally compensate
 20 Introduction

the load is proposed. Two inverters operate in parallel and at different switching frequency. Each
inverter is connected in series with a passive filter. The passive filter for the low frequency
inverter is designed to optimally eliminate 5th and 7th order harmonics whereas the passive filter
connected in series with the high frequency inverter is designed to optimally eliminate 11th and
13th order harmonics. The low frequency inverter operates typically around 550Hz and takes care
of all the lower order harmonics including reactive power demand of the system. The high

r
frequency inverter makes the source current to adhere to the IEEE519 standard.

pu
ar t
1.4 Organization of the Thesis
h
Chapter 1 presents an introduction to the work. It discusses the major developments those have

ag
Kh rig
taken place on Shunt Active Power Filters. The chapter also reports the motivation and scope of
the work. The thesis deals with all the three most important aspects, viz. (i) reference generation,
(ii) tracking and (iii) topological developments of shunt type APF. A brief literature survey on all
these three areas is available in this chapter.
y

Chapter 2 to 4 report reference generation techniques. Chapter 2 has exploited a hybrid genetic
IIT p

algorithm (hGA) for this purpose. The main motivation to go for the hGA is to improve speed of
convergence. The hGA based algorithm is simulated in MATLAB/SIMULINK and also
o

confirmed experimentally on a laboratory prototype.

C

It has been experienced that ANN is more convenient from the point of view of implementation.
Therefore, a simple ANN based reference generation technique is introduced in Chapter 3. The
proposed method is basically a replacement of a low pass filter by a properly trained ANN. The
technique although simple takes more time to converge. It is important to note that a predictive
algorithm is very useful to have a quick estimate of the reference current.
 21 Chapter-1

Chapter 4 first establishes the background to show how the change in DC link voltage can be
exploited for this purpose. An ANN based predictive and adaptive algorithm is reported that
shows excellent performance. Simulation and experimental results for different types of load
including severe unbalance such as single-phasing are reported in this chapter.

Once the reference is generated, it needs to be tracked with minimum error. Chapter 5 proposes

r
the use of sliding mode controller (SMC) for such purpose. A three-phase three-wire system is

pu
taken up for this study. To reduce the analytical burden and complexity in implementation,
feedback linearization is applied. Detail analysis of SMC based controller along with its

ar t
simulation and experimental verification are reported in this chapter.
h
ag
Kh rig
So far a simple VSI is used as a shunt type APF. This system although widely used, has many
limitations. Recent trend is to use low frequency and high frequency inverters in parallel for
VAR compensation as well as harmonic cancellation. Hybrid filters, where an inductor and a
capacitor are connected in series with the VSI are also reported. Such filters have the benefit to
y

optimize the rating of the APF and allow the VSI to operate at a very convenient voltage level.
Chapter 6 proposes parallel operation of hybrid filters at different switching frequencies. It
IIT p

starts with a simple hybrid dual parallel topology and goes further to explore some other
variations of the topology to overcome specific limitations. Detail simulation and experimental
o

results confirm the usefulness of the proposed configurations.

C

Chapter 7 concludes the work. It sums up the merits and limitations of the research work and
also suggests the direction for further investigation in this area.
 C 22

IIT po
y
Kh rig
ar th
ag
Introduction

pu
r