Chapter 144Harmonic Reduction Solutionby Applying On-Line Trained AdaptiveNeural Controller for Shunt Active Filter

Nguyen Thi-Hoai Nam, Chun-Tang Chao, Chi-Jo Wangand Cheng-Ting Hsu

Abstract This paper proposes an intelligent control method for shunt activepower filter to compensate the harmonic distortion in three phase power systems.For electric power systems, harmonics contamination generated by the nonlinearnature of the load is a serious and harmful problem. Shunt active filter (AF) hasbeen employed to mitigate line current harmonics. In the presented system, FuzzyLogic Controller (FLC) is first designed to implement the AF. Then the initialtraining data with two inputs, the error and derivate of the error, and one outputsignal from FLC can be obtained. Finally, a Neural Network (NN) with on-linetraining features is designed by using S-function in Simulink to achieve betterperformance. Simulation results show the effectiveness of the proposed activepower filter system which improves the power quality, reduces the current har-monics and obtains better transient and steady-state responses.

Keywords Active filter � Harmonics reduction � Neural network � On-linetraining

144.1 Introduction

In recent years, with the rapid economic development, all kinds of nonlinear loadsbased on power electronic devices (diode and thyristor rectifiers, electronic startersand arc furnaces, etc.) have been used in power systems and induced theappearance of dangerous phenomenon of harmonic currents flow in the electricalfeeder networks, producing distortions in the current and voltage waveforms.

N. T.-H. Nam � C.-T. Chao (&) � C.-J. Wang � C.-T. HsuDepartment of Electrical Engineering, Southern Taiwan University of Scienceand Technology, Tainan, Taiwan, Republic of Chinae-mail: tang@mail.stust.edu.tw

J. Juang et al. (eds.), Proceedings of the 2nd International Conference on IntelligentTechnologies and Engineering Systems (ICITES2013), Lecture Notes in Electrical Engineering 293,DOI: 10.1007/978-3-319-04573-3_144, � Springer International Publishing Switzerland 2014

1181

As a result, harmful consequences occur: equipment overheating, malfunction ofsolid state material, interferences with telecommunication systems, etc. The qualityof powersupply is seriously reduced, so power quality distortion has become aserious problem in electrical power systems due to the increase of nonlinear loadsdrawing non-sinusoidal currents. Nowadays, Shunt active filter (AF) have beenwidely used for harmonic mitigation as well as reactive power compensation, loadbalancing, voltage regulation, and voltage flicker compensation [1, 2]. NeuralNetwork (NN) based control methodologies have emerged in recent years aseffective means of solving nonlinear control problems [3–5]. The proposed designis based on the control model that uses on-line trained adaptive NN controller tocontrol the AF for reducing the harmonics of the electrical power systems. It doesnot exceed the thresholds fixed by the IEEE 519 standards [6]. Research resultshave been simulated and verified in the Matlab/Simulink software environment.

144.2 Mathematical Model of Shunt Active Filter

AF is a power electronics device based on the use of power electronics inverters.The Shunt Active Power Filter is connected in a common point connectionbetween the source of power system and the load system. The AF will prevent thesource of the polluting currents circulating in the power system lines, as shown inFig. 144.1 [7].

Suppose that iL, iF, iS are receiver absorbed current, desired power supplycurrent and the current AF must provide, respectively. Then we have the rela-tionship among them in formula given by:

iF ¼ iL � iS ð144:1Þ

If we let

iL ¼ if þ iH ð144:2Þ

where if is the fundamental component of load current, also the desired powersupply current iS, and iH is the harmonic current generated in load branch. FromEqs. (144.1) and (144.2), we obtain Eq. (144.3)

iF ¼ iH ð144:3Þ

Supply is iL

iF iF

Shunt Active Filter

iL iS

Fig. 144.1 Power systemwith nonlinear load and shuntAF

1182 N. T.-H. Nam et al.

Eq. (144.3) indicates that shunt active power filter is intended to generateexactly the same harmonics contained in the polluting current iL but with oppositephase.

144.3 Control System Design

144.3.1 Control Structure of Active Filter

To control AF to produce current tracking with the load current harmonic, thecontrol structure as shown in Fig. 144.2 is used [7, 8], where BPF is a band passfilter, LPF is a low pass filter, and PWM is pulse width modulation. In this paper,we apply the method of notch filter which consists of two identical band-passfilters in series. The fundamental frequency is set to 50 Hz. The transfer functionsof the BPF and LPF are given by Eq. (144.4).

HBPF sð Þ ¼ KBPF :B:S

s2 þ B:sþ x2c

; HLPF sð Þ ¼ KLPF

ssþ 1ð144:4Þ

144.3.2 On-Line Trained Adaptive Neural Controller Design

In the paper, the proposed on-line trained adaptive neural controller is designed tocontrol the AF for harmonic mitigation as well as reactive power compensation forthree phase power system, as shown in Fig. 144.3, where iref is the harmoniccurrent generated in power system and it is also the reference signal for thecontroller to track. The output current of the AF (iF) is expected to have the sameharmonic current value contained in the polluting current but with opposite phase.Both signals are compared to generate error (e) as the input of on-line trainedadaptive neural controller. The output of on-line trained adaptive neural controlleruN, which passes through LPF, is compared with carrier signal (vs), and the

RS LS iL RC LC

eS

iS vS

PWM Controller

Non-linerload

BPFAF

vF

iref

Lf

Inverter

iF

LPF

Fig. 144.2 The active filter control structure

144 Harmonic Reduction Solution 1183

resulting signal is used to switch 6 IGBTs of AF filter. The three-layer NNstructure shown in Fig. 144.4 is adopted to implement the proposed NN controller.

The net input to a node j in the hidden layer is calculated according to thefollowing equation.

netj ¼X

Wji � Oi

� �þ hj ð144:5Þ

The output of node j is

Oj ¼ f netj

� �¼ tanh b � netj

� �ð144:6Þ

where Oj represents the output of units in the hidden layers, netj is the summedinput to the units in the hidden layers, Wji are the connective weight between input

iFiref K

τS+1AF

vS

e

Referencemodel

uN

irefrm

Neural NetworkController

erm3 4

rm

k

EK e K e

O

∂= +

∂

2K e

1K

Fig. 144.3 On-line trained adaptive NN controller diagram

O1

net1

O2

net2

Oj

netj

OJ

netJ

Ok

netk

Plant

biasunit

biasunit

...

uN

Xp

Xrmerm

input layer i

hidden layer j

outputlayer k

Wji

Wkj

3 4rm

k

EK e K e

O

∂= +

∂

e

O2

K2

O1

K1

e

Fig. 144.4 Schematicdiagram of the proposed NNcontroller

1184 N. T.-H. Nam et al.

layers and hidden layers, b[ 0 is a constant, and f denotes the activation function,which is a hyperbolic tangent activation function

f netj

� �¼ 2

1þ e�b�netj� 1; �1\f netj

� �\1

� �ð144:7Þ

The net input to a node k in the output layer is

netk ¼X

Wkj � Oj

� �þ hk j ¼ 1; 2; . . .J; k ¼ 1; 2; . . .K ð144:8Þ

The corresponding output of neural network is

Ok ¼ f netkð Þ ¼ tanh b � netkð Þ ð144:9Þ

The energy function E is defined as

EN ¼12

XrmN � XpN

� �2¼ 12

e2N ð144:10Þ

where XrmN and XpN represents the outputs of the reference model and the plant atthe Nth iteration. The back-propagation algorithm is used to update the weights inthe NN [9].

It should be claimed that before implementing the on-line trained adaptive NNin Fig. 144.3, a PD-like Fuzzy Logic Controller (FLC) has been designed tocontrol the AF. The performance comparison can be found in the followingsection.

144.4 Simulation Results

Simulation results with methodologies of on-line trained adaptive NN controller isimplemented by Matlab/Simulink, as shown in Fig. 144.5, for electrical powersystem with AF. Table 144.1 summarizes the simulation parameters. The trainingresult of desired output and practical output is shown in Fig. 144.6 indicating thatthe practical output almost tracks the desired output with the error converging to 0(Figs. 144.7).

The system simulation results applying on-line trained adaptive NN controllerare shown in Figs. 144.8–144.10. The THD value drops to 1.03 %.

Figure 144.11 shows how AF current tracks the reference signal iref with on-line learning NN controller. In the first 0.01 s, the output of controller has nottracked its reference signal successfully, but after 0.01 s these two signals almostcoincide. Once again, it indicates the active operation of AF that helps to eliminatethe harmonics in power system. Moreover, consider that the motor runs in differentconditions: heavy load, medium load, and light load, the respective fundamentalcurrents are 700, 470, and 280 A. Table 144.2 reports the THD values in eachcase.

144 Harmonic Reduction Solution 1185

uC

uB

uA

Discrete,Ts = 5e-06 s.

powerguiUs

Vabc

Iabc

A

B

C

a

b

c

Three-PhaseV-I Measurement

+

Rs Ls

+

+

In1

In2

In3

G

Neural Network Controller

A

B

C

LoadIs

a

b

c

A

B

C

I_load measurement

ilA

ilB

ilC

I_load

I_compare

a

b

c

A

B

C

I_Filter Measurement

ifA

ifB

ifC

I_Filter

ifA

G

Out1

Out2

Out3

Active Filter Fo=50Hz

Fo=50Hz

Fo=50Hz

Fo=50Hz

Fo=50Hz Fo=50Hz

+

+

+

Rc Lc

Fig. 144.5 The simulation model of electrical power system with AF

Table 144.1 Simulation parameters of electrical power system

Parameter Value

Supply’s voltage esand frequency fs 220 V, 50 HzLine’s inductance Lsand resistance Rs 0.03 mH, 0.1 XImpedance upstream of the rectifier LcandRc 0.07 mH, 0.3 XInductance LDC, capacitor CDC, resistance RDC 0.3 mH, 470 lF, 0.45 XActive filter input DC supply: capacitor C, E,r 3.46 9 10-3 F, 600 V, 5x10-4 XActive filter output impedance Lf 1.2 mHifiref’s calculating, band-pass filter: damping factor n 0.707LPF corrector: gain K, time constant s 1, 5 e-6 sCarrier signal’s peak amplitude & frequency 10, 10 kHz

0 200 400 600 800 1000 1200 1400-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Desired outputPractical output

Fig. 144.6 Desired output and practical output

1186 N. T.-H. Nam et al.

0 200 400 600 800 1000 1200 1400-0.6

-0.4

-0.2

0

0.2

0.4Error

Fig. 144.7 Error after training

0 0.02 0.04 0.06 0.08 0.1 0.12

-400

-300

-200

-100

0

100

200

300

400

Time (s)

isa

[A]

Fig. 144.8 Supply current isa with NN

0 2 4 6 8 10 12 14 16 18 200

20

40

60

80

100

Harmonic order

THD= 1.03%

Mag

(%

of F

unda

men

tal)

Fig. 144.9 Harmonic spectrum of isa with NN

144 Harmonic Reduction Solution 1187

144.5 Conclusion

In this paper, the NN controller for AF is developed to reduce the harmonic currentfor nonlinear loads in electric power systems. It shows that with AF and on-linetrained adaptive NN controller design, the system achieves abetter response underdifferent running conditions of load. The simulations results also show that ShuntAF achieves good dynamic and steady-state responses. Furthermore, the THDvalue doesn’t exceed 5 %, conforming to the IEEE 519 standards.

0 0.02 0.04 0.06 0.08 0.1 0.12-500-400-300-200-100

0100200300400500

Time (sec)

Mag

nitu

de [A

]isaila

Fig. 144.10 Currents isa and ila applying AF with on-line trained adaptive NN controller

0 0.02 0.04 0.06 0.08 0.1 0.12-200-100

0100200300400500

Time (sec)

Mag

nitu

de [A

] IRE-AFIAF-USING NEURAL NETWORK

Fig. 144.11 AF current and its reference with on-line trained adaptive NN controller

Table 144.2 Total harmonic distortion (THD) (%) in different running conditions of load

Controller Load

Heavy load:Rdc = 0.07

Medium load:Rdc = 0.45

Light load:Rdc = 1.2

Without active filter THD = 3.46 THD = 12.54 THD = 20.33With AF using FLC THD = 0.49 THD = 1.04 THD = 3.23With AF using NN

controllerTHD = 0.45 THD = 1.03 THD = 2.44

1188 N. T.-H. Nam et al.

References

1. Elmitwally, A., Abdelkader, S., & Elkateb, M. (2000). Performance evaluation of fuzzycontrolled three and four wire shunt active power conditioners. IEEE Power EngineeringSociety Winter Meeting, 3, 1650–1655.

2. Yongjing, W., & Min, Z. (2005). Parallel type harmonic suppressor with an active filter and alc filter connected in series. Marine Electric and Electronic Technology, 4, 14–16.

3. Padma, S., Rajaram, M., Lakshmipathi, R., & Madhubalan, S. (2011). Neural networkcontroller for static synchronous series compensator (SSSC) in transient stability analysis. InPower Electronics (IICPE) (pp. 1–6).

4. Nitin, G., Singh, S. P., & Dubey, S. P. (2010). Neural network based shunt active filter forharmonic and reactive power compensation under non-ideal mains voltage. In IEEE (ICIEA)(pp. 370–375).

5. Chen, S. C., Jyh, L. Y., Sum, N. V., & Mao, S. M. (2013). A Novel Fuzzy Neural NetworkController for Maglev System with Controlled-PM Electromagnets. Intelligent Technologiesand Engineering Systems, 234, 551–561.

6. Khalid, S., & Dwivedi, B. (2011). Power quality issues, problems, standards and their effects inindustry with corrective means. International Journal of Advances in Engineering andTechnology, 1, 1–11.

7. Hocine, B., & Hind, D. (2005). Shunt active filter controlled by fuzzy logic. EngineeringSciences, 2005(18), 231–247.

8. Karuppanan, P., & Mahapatra, K. K. (2012). PI and fuzzy logic controllers for shunt activepower filter-a report. ISA Transactions, 51, 667–671.

9. Michael, N. (2005). Artificial intelligence: A guide to intelligent systems (2nd ed.). England:Pearson Education.

144 Harmonic Reduction Solution 1189