chapter 4 harmonic elimination of fully controlled ac–dc...

65

CHAPTER 4

HARMONIC ELIMINATION OF FULLY CONTROLLED

AC–DC CONVERTERS USING ARTIFICIAL NEURAL

NETWORKS

4.1 INTRODUCTION

In this chapter a shunt active power filter is developed to eliminate

the supply current harmonics using Artificial Neural Networks. For several

years, active power filters (APFs) have been recognized as advanced

techniques for harmonic compensation (Agaki 1996). The concept of the

active power filter is to detect or extract the unwanted harmonic components

of a line current, and then to generate and inject a signal into the line in such a

way to produce partial or total cancellation of the unwanted components.

APFs are very able to suppress the current harmonics and to correct power

factor, especially in fast-fluctuating loads, in comparison to other

compensation devices. A lot of recent research work investigates and tries to

improve APFs by developing new topologies with power electronics

technologies or new control laws. Active power filters could be connected

either in series or in parallel to power systems. Therefore, they can operate as

either voltage sources or current sources. The shunt active filter is controlled

to inject a compensating current into the utility system so that it cancels the

harmonic currents produced by the nonlinear load. Artificial intelligence (AI)

techniques, particularly the neural networks, are recently having significant

impact on power electronics and motor drives. Neural networks have created

a new and advancing frontier in power electronics, which is already a

66

complex and multidisciplinary technology that is going through dynamic

evolution in the recent years (Bimal K. Bose 2007). ANN techniques have

been applied with success in control of APF and are very promising in the

field. Indeed, the learning capacities of the ANNs allow an online adaptation

to every changing parameter of the electrical network, e.g., nonlinear and

time-varying loads. Most of these control constraints are quite still very

challenging with classic control methods (Djaffar Ould Abdeslam et al 2007).

Active Power Filters (APFs) have been used as effective way to

compensate harmonic currents in nonlinear loads. APFs basically work by

detecting the harmonic components from the distorted signals and injecting

these harmonic current components with a current of the same magnitude but

opposite phase into the power system to eliminate these harmonics (Hsiung

and Cheng Lin 2007). Active filter topologies have been recently discussed in

many papers (Marks and Green 2002). A comprehensive study of practical

topologies was given by Akagi (1997). Obviously, fast and precise harmonic

detection is one of the key factors to design APFs. The simplest way to

identify harmonics and generate the harmonic current is to use discrete

Fourier transformation (DFT) or Fast Fourier Transformation (FFT).

Although it has wide applications, the DFT or FFT has certain limitations in

harmonic analysis (Chan and Plant 1981). Presently, neural network has

received special attention from researchers because of its simplicity, learning

and generalization ability, and it has been applied in the field of engineering,

such as in harmonic detection since early 1900s (Pecharanin et al 1994).

In this thesis an improved scheme is proposed in which only the

actual load current is given to adaptive neural network and the reference

current is obtained. The load voltage is given to a DC regulator with a simple

PI controller, which gives the actual filter current, and which is compared

with reference current the gating pulses to Insulated Gate Bipolar Transistors

67

(IGBTs) are provided through a hysteresis band comparator. The program

developed can be used to eliminate on line harmonics. The same program is

used for eliminating the supply current harmonics in the AC–DC converters

and AC voltage controllers.

4.2 PRINCIPLE OF THE SHUNT ACTIVE POWER FILTERS

In a power system, the voltage (or current) source waveform

usually consists of a fundamental component, some harmonic components,

and random noise. Among these contents, the fundamental component

generally has a significantly dominant proportion.

The active filter (AF) technology is now mature for providing

compensation for harmonics, reactive power, and/or neutral current in ac

networks. It has evolved in the past quarter century of development with

varying configurations, control strategies, and solid-state devices. AF’s are

also used to eliminate voltage harmonics, to regulate terminal voltage, to

suppress voltage flicker, and to improve voltage balance in three-phase

systems. This wide range of objectives is achieved either individually or in

combination, depending upon the requirements and control strategy and

configuration, which have to be selected appropriately. The term active power

filter is a generic one and is applied to a group of power electronic circuits

incorporating power switching devices and passive energy-storage-circuit

elements, such as inductors and capacitors. The functions of these circuits

vary depending on the applications. They are generally used for controlling

current harmonics in supply networks at the low-to-medium voltage

distribution level or for reactive power and/or voltage control at high-voltage-

distribution level (Arrillaga et al 1997). These functions may be combined in

a single circuit or in separate active filters. Active power filtering based on the

injection method is basically performed by replacing the portion of the sine

68

wave that is missing in the current drawn by a nonlinear load. This can be

accomplished in two stages. The first stage consists of detecting the

amplitudes and phases of the AC harmonic currents (or any system quantity

associated with them), which are present in the AC line. The second stage is

the injection of the appropriate harmonic currents (or insertion of appropriate

harmonic voltages) at the appropriate frequency so as to supply the AC

harmonic currents required by the nonlinear load. Since their basic operating

principles were firmly established in the 1970s (Bird et al 1969, Sasaki and

Machida 1971), active power filters have attracted the attention of power

electronics researchers/engineers who have a concern about harmonic

pollution in power systems (Akagi et al 1986, Peng et al 1990, Akagi et al

1990). Moreover, deeper interest in active filters has been spurred by

1. the emergence of semiconductor switching devices such as

IGBTs and power MOSFETs, which are characterized by fast

switching capability and insulated-gate structure;

2. the availability of digital signal processors (DSPs), field-

programmable gate arrays (FPGAs), analog to digital (A/D)

converters, Hall-effect voltage/current sensors, and

operational and isolation amplifiers at reasonable cost

(Hirofumi Akagi 2005).

Active filters can be divided into single-phase active filters and

three-phase active filters. Research on single-phase active filters has been

carried out, and the resultant papers have appeared in technical literature.

However, single-phase active filters would attract much less attention than

three-phase active filters because single-phase versions are limited to low-

power applications except for electric traction or rolling stock. A generalized

block diagram of an active power filter is presented in Figure 4.1.

69

Figure 4.1 Generalized block diagram for active power filters

The information regarding the harmonic current, generated by a

nonlinear load, for example, is supplied to the reference-current voltage

estimator together with information about other system variables. The

reference signal from the current estimator, as well as other signals, drives the

overall system controller. This in turn provides the control for the PWM

switching pattern generator. The output of the PWM pattern generator

controls the power circuit via a suitable interface. The power circuit in the

generalized block diagram can be connected in parallel, series or

parallel/series configurations, depending on the connection transformer used.

The active harmonic source within the filtering network is basically

a static converter connected to a DC source. The converter must be controlled

to provide the proper filtering harmonic currents or voltages. This is

accomplished by shaping the DC input source into an output waveform of

appropriate magnitude and frequency through modulation of semiconductor

switches (Grady et al 1990).

70

The harmonic converter can use either a DC voltage source or a DC

current source. The DC source of a voltage converter consists of a capacitor

that resists voltage changes, while that of a current converter consists of an

inductor that resists current changes. In both cases, the DC source receives its

power from the AC power system. Converters are referred to as either

voltage-fed or current-fed according to the type of DC-side source. The basic

voltage and current source converter topologies are displayed in Figure 4.2(a)

and (b).

(a)

(b)

Figure 4.2 (a) Single-phase and three-phase current-source converter

(CSC), (b) Single-phase and three-phase voltage-source

converter (VSC)

71

In the current-source converter, a diode is placed in series with

every switch to avoid reverse breakdown of the switch when the voltage

across the switch during the OFF-period is negative. In voltage-source

converter, an inverted diode is placed across each switch to provide a path for

the current when the current cannot pass through the switch. The power

electronic circuits and devices used in both types of converter are quite

similar. Most of the existing active power filters utilize switching devices

such as Power MOSFETs and IGBTs for switching speeds up to 50 kHz.

However the most attractive device is the IGBT. It has the merit of fast

switching capability and requires very little drive power at the gate and

capable of handling high power. At present, the voltage-source converter is

more favorable than the current-source converter in terms of cost, physical

size, and efficiency.

4.3 HARMONIC ESTIMATION TECHNIQUES

One important issue that assesses and evaluates the quality of the

delivered power is the estimation or extraction of harmonic components from

distorted current or voltage waveforms. In order to provide high quality

electricity, it is essential to accurately estimate or extract time varying

harmonic components, both the magnitude and the phase angle, to mitigate

them using active power filters. There are several harmonic estimation

techniques are available among which the Fast Fourier Transform (FFT), the

Kalman filter (KF) and Artificial Neural Networks (ANN) are the most

popular. Figure 4.3 displays some of these estimation technique references.

Estimation Techniques

FFT Kalman Filter ANN

Figure 4.3 Harmonic estimation techniques

72

4.3.1 Discrete Fourier Transform (DFT) The DFT-based algorithm (fast Fourier transform (FFT)) for

harmonic measurement and analysis is a well-known technique and is widely used due to its low computational requirement. In this approach the

coefficients of individual harmonics are computed by implementing Fast

Fourier Transform on digitized samples of a measured waveform in a time window. The description of the algorithm is well documented in many

references (Cooly and Tukey 1965, Harris 1978, Oran Brigham 1988).

There are several performance limitations inherent in the FFT

application. These limitations are (Oran Brigham 1988):

1. the waveform is assumed to be of a constant magnitude during

the window size considered (stationary), 2. the sampling frequency must be greater than twice the highest

frequency of the signal to be analyzed , and

3. the window length of data must be an exact integer multiple of power-frequency cycles.

It has been reported in (Girgis 1991) that failing to satisfy these conditions will result in leakage and picket fence effects and hence inaccurate

waveform frequency analysis. Moreover, the DFT-based algorithm can cause

computational error and may lead to inaccurate results if the signal is contaminated by noise and/or the DC component is of a decaying nature

(Dash et al 1996).

As far as the active filters are concerned, and because the

transformation process takes time, the harmonic compensation will be delayed by two cycles if the FFT is used as an estimation tool (Osowski 1992). This

will influence the performance of active filtering in case that the load current

is in fluctuated state.

73

4.3.2 Harmonic Estimation Using Kalman Filter

In the Kalman filter approach (Haili Ma et al 1996) (Moreno Saiz

et al 1997), a state variable mathematical model of the signal, including all

possible harmonic components, is used. Dash and Sharaf (1989) were among

the first who utilized the Kalman filter technique to estimate the stationary

harmonic components of known frequency from unknown measurement

noise. Girgis et al (1991) generalized the work in reference to predict time-

varying harmonics too. However, it was pointed out in (Girgis et al 1991),

that the Kalman filter scheme requires more computational process to update

the state vector when estimating the time varying harmonics compared to the

stationary. Later, Haili Ma and Girgis (1991) utilized the Kalman filter

approach to identify and track the harmonic sources in power systems. A

hardware implementation of the Kalman filter to track power system

harmonics was presented by Moreno Saize et al (1997). In the Kalman filter

approach, a state variable mathematical model of the signal, including all

possible harmonic components is used. The Kalman filter technique estimates

the harmonic components by utilizing a smaller number of samples and in

relatively shorter time as compared to FFT (Girgis et al 1991). However,

Kalman filter technique suffers from being computationally demanding due to

transcendental function evaluations, which makes its unfit for on-line

applications.

4.4 HARMONIC ESTIMATION USING ARTIFICIAL NEURAL

NETWORKS

There are many available algorithms for estimation of power

system harmonic components based on learning principles. Some of ANN

algorithms are based on the back propagation learning rule (Hartana and

Richards 1990, Mori 1992, Osowski 1992), while others utilized the LMS

74

(Widrow-Hoff) learning rule (Pecharanin et al 1995, Dash 1996). Hartana and

Richards (1990) were among the first who used back propagation ANN to

track harmonics in large power systems, where it is difficult to locate the

magnitude of the unknown harmonic sources. In their method, an initial

estimation of the harmonic source in a power system was made using neural

networks. They used a multiple two-layer feed forward neural network to

estimate each harmonic amplitude and phase.

Mori et a1 (1992) have provided a basic ANN mode1 to estimate

the voltage harmonics from real measured data. In their paper, a comparison

between the conventional estimation methods for predicting the 5th harmonic

is given. Pecharanin et al (1995) presented an ANN topology, based on the

back propagation learning rule, for harmonic estimation to be used in active

power filters. They taught the neural network to map the amplitude of the 3rd

as well as the 5th harmonic from a half cycle of a distorted current waveform.

This method has a limited applicability in active filtering since it does not

consider the detection of the harmonic phase angles in which it may increase

the distortion and make the case worse if the injected signal is of the wrong

phase. The main drawback of the back propagation ANN is the requirement

of the huge data set required for training. Also, the back propagation ANN

may lead to inaccurate results because of the random-like behavior and the

large variations in the amplitude and the phase of the harmonic components

and/or in the presence of random noise (Dash et al 1997).

Osowski (1992) provided an ANN that is based on the least mean

square (LMS) learning principle to estimate the harmonic components in a

power system. Later, Dash et al (1996) utilized the ADALINE, a version of an

ANN, as a new harmonic estimation technique. The learning rule of the

method is based on the LMS introduced by Widrow-Hoff. ADALINE is an

adaptive technique. Its main advantages are speed and noise rejection. It

75

proves to be superior to the Kalman Filter technique in finding the magnitudes

and phases of the harmonics (Dash et al 1996).

4.4.1 ADAptive Llnear NEuron (ADALINE)

The ADALNE is a two layered feed-forward perceptron, having N

input units and a single output unit. This allows their outputs to take on any

value, whereas the perceptron output is limited to either 0 or 1. Both the

ADALINE and the perceptron can only solve linearly separable problems.

However, here it makes use of the LMS (Least Mean Square) learning rule,

which is much more powerful than the perceptron learning rule. The LMS or

Widrow-Hoff learning rule minimizes the mean square error and thus moves

the decision boundaries as far as it can from the training patterns. The

ADALINE is described as a combinational circuit that accepts several inputs

and produces one output. Its output is a linear combination of these inputs. An

ADALINE in block diagram form is depicted in Figure 4.4.

The input to the ADALINE is TnxxxxX ).,,.........,,( 210 , where 0x , is

called a bias term or bias input, is set to 1. The ADALINE has a weighted

vector TnwwwwW ).,.........,,( 210 and its output is simply written as

0 1 1 2 2 ..........Tn ny W X w w x w x w x (4.1)

In a digital implementation, this element receives at time k an input

signal vector or input pattern vector Tnkkkk xxxxkX ).,,.........,,()( 210 and a

desired response ),(kyd a special input used to affect learning. The

components of the input vector are weighted by a set of coefficients; the

weight vector is given by Tnkkkk wwwwkW )..........,,()( ,210 . The sum of the

weighted inputs, i.e., )()()( kXkWky T is then computed.

76

Figure 4.4 Structure of the Adaptive Linear Neuron (ADALINE)

During the training process, input patterns and corresponding

desired responses are presented to the ADALINE. An adaptation algorithm,

usually the Widrow-Hoff LMS algorithm, is used to adjust the weights so that

the output responses of the input patterns become as close as possible to their

respective desired responses. This algorithm minimizes the sum of squares of

the linear errors over the training set. The linear error e(k) is defined to be the

difference between the desired response )(kyd and the linear output )(ky , at

time or sample k.

The Widrow-Hoff learning delta rule calculates the changes to

weights of ADALINE to minimize the mean square error between the desired

signal output )(kyd and the actual ADALINE output )(ky over all k. The

weight adjustment, or adaptation, equation can be written as (Bernard and

Michael 1990).

77

( ) ( )( 1) ( )( ) ( )T

e k x kW k W kx k x k

(4.2)

where k = time index of iteration,

W (k) = weight vector at time k,

X (k) = input vector at time k,

)()()( kykyke d = error at time k

= learning parameter

Training Algorithm for an ADALINE is shown below:

Step: 0. Initialize weights (small random values are usually used).

Set learning rate α.

Step: 1. while stopping condition is false

Do Steps 2 to 6.

Step: 2. for each training pair t:S do Steps 3 to 5.

Step: 3. Set activations of input units, n........1i

ii SX

Step: 4. Compute net input to output unit:

iin W.XbYi

Step: 5. Update bias and weights, n.....1i

)( inoldnew ytbb

ioldinewi XyintWW )(

Step: 6. Test for stopping condition: If the largest weight change

that occurred in step 2 is smaller than a specified tolerance, then

stop; otherwise continue.

78

4.4.2 ADALINE as Harmonic Estimator

The ADALINE is used to estimate the time varying magnitudes of

selected harmonic in a distorted waveform. Consider a distorted signal f (t)

with the Fourier series expansion:

1

( ) sin( )N

n nn

f t A n t

(4.3)

where An and φn, are the amplitude and the phase angle of the harmonic,

respectively, N is the total number of harmonics. The discrete-time

representation of f (t) will be:

1 1

(2 ) (2 )( )N N

n n

n n

AnCos Sin nk AnCos Cos nkf kNs Ns

(4.4)

where Ns, = sampling rate = os FF , Fs= sampling Frequency, and Fo= nominal

system frequency. The proposed current ADALINE network is shown in

Figure 4.5. The input to the ADALINE input vector is chosen to be:

( ) ( )( )TSin n k Cos n kx k

Ns Ns

(4.5)

where n is the selected harmonic order, t (k) is the time and its desired output

Yd(k) is set to be equal to the actual signal, f (k). The weight vector is set to

be

1 1 1 1[ cos sin ........ cos sin ]n n n nWo A A A A (4.6)

79

The tap weight vector of the adaptive filter is denoted by

)()()( kykyke d and perfect tracking is attained when the tracking error e(k)

approaches zero. This optimum condition is realized when the estimation

variance and the lag variance contribute equally to the mean square deviation.

Therefore,

( ) ( ) ( ) ( )Td oy k y k f k W x k (4.7)

Figure 4.5 Proposed current ADALINE network

4.5 THE PROPOSED ADALINE BASED SHUNT ACTIVE

POWER FILTER

The block diagram of the proposed power line conditioner using

active power filter is shown in Figure 4.6. The proposed Modular Active

power filter connected to the electric Distribution system. The line current

signal is obtained and fed to an ADALINE which extracts the fundamental

components of the line current signal. In the controller block, fundamental

component is compared with distorted line current to generate modulating

signal. This modulating signal is used to generate Pulse-Width Modulated

(PWM) switching pattern for the switches of the active line conditioner

module (Vazquez and Salmeron 2003, Rukonuzzaman and Mutsuo Nakaoka

2002). The output current of the active filter is injected into the power line.

The injected current, equal-but opposite to the harmonic components to be

eliminated. Harmonics are suppressed by connecting the active filter modules

to the electric grid.

80

Figure 4.6 Block diagram of the proposed scheme

4.6 SIMULATION RESULTS AND DISCUSSION

The effectiveness of the proposed model is verified with a ANN

based shunt active power filter for single phase and three phase fully

controlled AC-DC converter is analyzed and results are discussed.

4.6.1 ANN Based Shunt Active Power Filter for Single Phase Fully

Controlled Converter

The simulink model for ANN based shunt active power filter for

single phase fully controlled converter and the control block is shown in

Figures 4.7 and 4.8 respectively.

81

Figure 4.7 Simulated diagram of proposed ANN based shunt active

filter connected for single phase fully controlled converter

Figure 4.8 Block diagram of the control unit for single phase fully

controlled converter

82

The converter is tested with different firing angles. First the

converter is operated with a firing angle of 18o. Figure 4.9 shows the

waveforms of the supply current, load current and filter current after the

application of the proposed active power filter. Figures 4.10 and 4.11

represent the harmonic spectrum of the supply current before and after the

application of the proposed shunt active power filter respectively. The supply

current waveform is improved and its THD is reduced to 2.14% from 16.32%.

Table 4.1 shows the comparative results of individual harmonics and %THD.

Figure 4.9 Supply current, Load current and filter current waveforms

of the converter (with filter for a firing angle of 18o)

Figure 4.10 Harmonic spectrum of the supply (Load) current (without

filter for the firing angle of 18o)

83

Figure 4.11 Harmonic spectrum of the supply current (with filter for the

firing angle of 18o)

Table 4.1 Comparative analysis of % harmonic distortion with the

proposed ANN based shunt active power filter (For firing

angle α =18o)

Sl. No Harmonic order

% Harmonic Distortion without Neural Network Based

Active Filter

with Neural Network Based

Active Filter 1. 1 100.00 100.00 2. 3 11.11 1.48 3. 5 7.18 0.80 4. 7 5.28 0.68 5. 9 4.19 0.51 6. 11 3.49 0.44 7. 13 3.00 0.37 8. 15 2.65 0.34 9. 17 2.37 0.28 10. 19 2.16 0.26 11. 21 1.98 0.20 12. 23 1.84 0.18

%THD 16.32 2.14

84

In the second case the firing angle of the converter is varied to 36o.

Figure 4.12 shows the waveforms of the supply current, load current and filter

current after the application of the proposed active power filter. Figures 4.13

and 4.14 represent the harmonic spectrum of the supply current before and

after the application of the proposed shunt active power filter respectively.

The supply current waveform is improved and its THD is reduced to 2.43%

from 22.43%. Table 4.2 shows the comparative results of individual

harmonics and %THD.

Figure 4.12 Supply current, load current and filter current waveforms




85

Figure 4.14 Harmonic spectrum of the supply current (with filter for the

firing angle of 36o)


proposed ANN based shunt active power filter (For firing

angle α =36o)



Active Filter


Active Filter 1. 1 100.00 100.00 2. 3 18.36 1.93 3. 5 9.40 0.84 4. 7 5.94 0.71 5. 9 4.27 0.53 6. 11 2.42 0.46 7. 13 2.12 0.38 8. 15 2.02 0.32 9. 17 1.84 0.28 10. 19 1.72 0.23 11. 21 1.51 0.21 12. 23 1.31 0.19

%THD 22.43 2.43

86

Finally the firing angle of the converter is varied to 54o. Figure 4.15

shows the waveforms of the supply current, load current and filter current

after the application of the proposed active power filter. Figures 4.16 and 4.17

represent the harmonic spectrum of the supply current before and after the

application of the proposed shunt active power filter respectively. The supply

current waveform is improved and its THD is reduced to 3.49% from 25.11%.

Table 4.3 shows the comparative results of individual harmonics and %THD.

Figure 4.15 Supply current, load current and filter current waveforms




87

Figure 4.17 Harmonic spectrum of the supply current (with filter for the firing angle of 54o)


proposed ANN based shunt active power filter (For firing angle α =54o)



Active Filter


Active Filter 1. 1 100.00 100.00 2. 3 19.93 1.93 3. 5 9.67 0.84 4. 7 6.27 0.71 5. 9 5.78 0.53 6. 11 4.68 0.46 7. 13 3.82 0.38 8. 15 3.42 0.32 9. 17 2.69 0.28 10. 19 2.26 0.23 11. 21 1.92 0.21 12. 23 1.62 0.19

%THD 25.11 3.49

The proposed shunt active power filter is also tested with different

loads i.e. by connecting a single phase half controlled converter parallel to the

single phase fully controlled converter. The fully controlled converter is

88

operated with the same load conditions with a firing angle of 18o. The half

controlled converter is operated with a firing angle of 36o to supply a RL load

(R=180Ω and L=1H). Figure 4.18 shows the waveforms of the total supply

current, load current of fully controlled converter, load current of half

controlled converter, total load current and filter current after the application

of the proposed active power filter. Figures 4.19 and 4.20 represent the

harmonic spectrum of the load (supply) current of the half controlled and fully

controlled converter, before application of the proposed shunt active power

filter respectively. The THD value for the supply current of half controlled

converter is 23.55% and for that of fully controlled converter is about

16.53%. Figures 4.21 and 4.22 represent the harmonic spectrum of the total

supply current before and after the application of the proposed shunt active

power filter respectively. So the supply current waveform is improved and its

THD is reduced to 3.70% from 16.30%. Table 4.4 shows the comparative

results of individual harmonics and %THD.

Figure 4.18 Supply current, Load current and filter current waveforms of the parallel combination of half controlled and fully controlled converter

89

Figure 4.19 Harmonic spectrum of the supply (Load) current of half

controlled converter (without filter for the firing angle of

36o)

Figure 4.20 Harmonic spectrum of the supply (Load) current of fully

controlled converter (without filter for the firing angle of

18o)

Figure 4.21 Harmonic spectrum of the total load current (without filter)

90

Figure 4.22 Harmonic spectrum of the total supply current (with filter)


proposed ANN based shunt active power filter for combined

half controlled and fully controlled converter



Active Filter


Active Filter 1. 1 100.00 100.00 2. 3 14.08 3.23 3. 5 5.61 1.24 4. 7 2.13 0.46 5. 9 2.05 0.49 6. 11 2.26 0.51 7. 13 1.75 0.39 8. 15 0.91 0.20 9. 17 1.09 0.23 10. 19 1.98 0.43 11. 21 2.54 0.55 12. 23 2.58 0.56

%THD 16.30 3.70

91

4.6.2 ANN Based Shunt Active Power Filter for three phase fully


The simulink model of ANN based shunt active filter for three

phase fully controlled converters is shown in Figure 4.23. The ANN based

control unit for shunt active power filter is shown in Figure 4.24. In the

control circuit the supply currents of all the phases are extracted separately

and processed to get the filter current. The specifications for the three phase

fully controlled converter is, Input Voltage Vs1, 2, 3= 400V, f=50Hz, Rs=0.1Ω,

Ls=0.05mH. Load connected to the converter is 1000 watts. The converter

output is simulated for the firing angles of 0o and 30o respectively for without

neural network based active power filter and for with neural network based

active power filter and the results are compared. The supply current, load

current and filter current wave forms for firing angle of 0o and 30o are shown

in Figures 4.25 and 4.26 respectively.

Figure 4.23 Simulated circuit diagram of proposed ANN based shunt

active filter connected for three phase fully controlled

converter

92

Figure 4.24 Block diagram of the control unit for three phase fully


Figure 4.25 Supply current, load current and filter current waveform of

ANN based shunt active power filter (for α=0o)

93

Figure 4.26 Supply current, load current and filter current waveform of

ANN based shunt active power filter (for α=300)

The harmonic spectrum for without and with ANN based shunt active power filter for the firing angle of 0o is shown in Figures 4.27 and 4.28 respectively. Similarly Figures 4.29 and 4.30 represent harmonic spectrum for without and with ANN based shunt active power filter for the firing angle of

30o. The comparative results are shown in the Tables 4.5, 4.6 and it shows the improved performance of the ANN based shunt active power filter. Table 4.7

shows the comparison total harmonic distortion for different firing angles.

Figure 4.27 Harmonic Spectrum for the supply current (without filter

for firing angle α = 0o)

94

Figure 4.28 Harmonic Spectrum for the supply current (with filter for firing angle α = 0o)

Figure 4.29 Harmonic Spectrum for the supply current (without filter for firing angle α = 30o)

Figure 4.30 Harmonic Spectrum for the supply current (with filter for firing angle α =30o)

95

Table 4.5 Comparative analysis of % harmonic distortion with the proposed ANN based shunt active power filter (For firing angle α =0o)



Active Filter


Active Filter 1 1 100.00 100.00 2 5 22.63 1.95 3 7 10.11 0.84 4 11 7.87 0.65 5 13 4.48 0.37 6 17 3.60 0.30 7 19 2.14 0.18 8 23 1.66 0.14 9 25 1.16 0.11

%THD 26.77 2.29

Table 4.6 Comparative analysis of % harmonic distortion with the proposed ANN based shunt active power filter (For firing angle α =30o)

Sl. No. Harmonic order


Active Filter


Active Filter 1 1 100.00 100.00 2 5 27.63 2.13 3 7 12.43 0.93 4 11 11.70 0.85 5 13 6.55 0.47 6 17 7.31 0.53 7 19 4.04 0.29 8 23 5.04 0.36 9 25 2.27 0.21

%THD 34.54 2.62

96

Table 4.7 Comparison of total harmonic distortion for different firing

angles

Sl. No.

Firing angle in (degrees)

% Total Harmonic Distortion(%THD) without Neural Network Based

Active Filter


Active Filter 1 0 26.77 2.29 2 15 30.08 2.49 3 30 34.54 2.62 4 45 42.67 2.71 5 60 62.81 2.78

4.7 CONCLUSION

The proposed neural network based shunt active power filter can

compensate a highly distorted line current by creating and injecting

appropriate compensation current. The simulation was carried out for single

phase and three phase fully controlled converters. In the test cases simulated

for different firing angles, the THD of the supply current is improved to less

than 5% with the proposed shunt active power filter. In addition to this the

simulation was carried out for the additional nonlinear loads (single phase

half controlled converter) and the THD in the supply current is also improved

to less than 5%. The performances reached by the proposed method are better

than those obtained by more traditional techniques as mentioned in the

introduction. It enhances the reliability of active filter. The overall switching

losses are minimized due to selected harmonic elimination. Speed and

accuracy of ADALINE results in improved performance of the Active Power

Filter.

chapter 4 harmonic elimination of fully controlled ac–dc...

Documents