This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication.

1

Back-Propagation Control Algorithm for Power Quality Improvement using DSTATCOM

Bhim Singh, Fellow, IEEE and Sabha Raj Arya, Member, IEEE

AbstractThis paper presents an implementation of threephase Distribution Static Compensator (DSTATCOM) using back-propagation (BP) control algorithm for its functions such as harmonics elimination, load balancing and reactive power compensation for power factor correction (PFC) and zero voltage regulation (ZVR) under nonlinear loads. A BP based control algorithm is used for extraction of fundamental weighted value of active and reactive power components of load currents which are required for estimation of reference source currents. A prototype of DSTATCOM is developed using a Digital Signal Processor (DSP) and its performance is studied under various operating conditions. The performance of DSTATCOM is found satisfactory with proposed control algorithm for various types of loads. Index Terms BP control algorithm, Harmonics, Load balancing, Weights, Power quality.

HE quality of available supply power has a direct economic impact on industrial and domestic sectors which affects the growth of any nation [1]. This issue is more serious in electronic based systems. Level of harmonics and reactive power demand are popular parameters that specify the degree of distortion and reactive power demand at a particular bus of the utility [2]. The harmonic resonance is one of most common problem reported in low and medium level distribution system. It is due to capacitors which are used for power factor correction and source impedance [3]. Power converter based custom power devices (CPDs) are useful for reduction of power quality problems such as power factor correction, harmonics compensation, voltage sag/swell compensation, resonance due to distortion, voltage flicker reduction within specified international standards[4-6]. These CPDs include Distribution Static Compensator (DSTATCOM), Dynamic Voltage Restorer (DVR) and
Manuscript received September 19, 2012. Accepted for publication March 20, 2013. Copyright (c) 2009 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending a request to pubs-permissions@ieee.org. Bhim Singh and Sabha Raj Arya are with the Department of Electrical Engineering, Indian Institute of Technology Delhi, New Delhi-110016, INDIA (e-mail: bhimsingh1956@gmail.com, sabharaj1@gmail.com).

I. INTRODUCTION

Unified Power Quality Conditioner (UPQC) in different configurations [7-9]. Some new topologies of them are also reported in the literature such as indirect matrix converter (IMC) based active compensator where dc link capacitor can be removed [10]. Other new configurations are based on stacked multicell converters where main features are on the increase in the number of output voltage levels, without transformer operation and natural self-balancing of flying capacitors voltage etc [11]. Performance of any custom power device depends very much upon control algorithm used for reference current estimation and gating pulse generation scheme. Some of the classical control algorithms are Fryze power theory, Budeanu theory, p-q theory and SRF theory [12-14], Lyapunov-function-based control [15] and nonlinear control technique [16] etc. Many non model and training based alternative control algorithms are reported in the literature with application of soft computing technique such as neural network, fuzzy logic and adaptive neuro-fuzzy etc [17-20]. Adaptive learning, self organization, real time operation, fault tolerance through redundant information are major advantages of these algorithms. Neural network based control algorithm such as Hopfield type neural network is also used for estimation of amplitude and phase angles of the fundamental component both with highly distorted voltage by assumption of known power frequency [21]. An improved adaptive detecting approach for extraction of error signal with variable learning parameters can be chosen for fast response to improve tracking speed, and low value in a stable period to improve accuracy [22]. Wu et. al [23] have proposed new control algorithm based on inverse control with neural network interface which applied for instantaneous calculation of switching on-off time in digital environment. A survey on iterative learning control (ILC) is presented by Ahn et al. [24] and it is classified in different subsection within the wide range of application. The main idea of ILC is to find an input sequence such that the output of the system is as close as possible to a desired output. Control algorithms reported in available texts such as quantized Kernel least mean square algorithm [25], radial basis function networks [26], feedforward training [27] can also be used for control of CPDs. An immune RBF (Radial Basis Function) neural network integrates the immune algorithm with the RBF neural network. This algorithm has the advantages in learning speed

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
2

and accuracy of the astringent signal. So, it can detect the harmonics of the current timely and precisely in the power network [28]. A multilayer perceptron neural network is useful for identification of nonlinear characteristics of the load. The main advantage of this method is that it requires only waveforms of voltages and currents. A neural network with memory is used to identify the nonlinear load admittance. Once training is achieved, the neural network predicts the true harmonics current of the load when supplied with a clean sine wave. Its application with SRF theory is described by Mazumdar el al. [29, 30]. Feed forward back propagation ANN consists of various layers as input layer, hidden layer and output layer. It is based on feed forward backpropagation with high ability to deal with complex non linear problems [31]. Back-propagation (BP) control algorithm is also used to design the pattern classification model based on decision support system. Standard BP model has been used with full connection of each node in the layers from input to output layers. Some applications of this algorithm are as to identification of user faces, industrial processes, data analysis, mapping data, control of power quality improvement devices etc [32]. Control of power quality devices by neural network is a latest research area in field of power engineering. Extraction of harmonic components decides the performance of compensating devices. Back-propagation algorithm which is trained the sample can detect the signal of power quality problem in real-time. Its simulation study for harmonic detection is presented in [33]. Many neural network based algorithms are reported with theoretical analysis in single phase system but their implementation to DSTATCOM is hardly reported in the available literature. In this paper, a back-propagation (BP) algorithm is implemented in three phase shunt connected custom power device known as DSTATCOM for extraction of weighted value of load active power and reactive power current components in nonlinear loads. Proposed control algorithm is used for harmonics suppression and load balancing in power factor correction (PFC) and zero voltage regulation (ZVR) modes with DC voltage regulation of DSTATCOM. In this BP algorithm, training of weights has three stages. It includes feed forward of the input signal training, calculation and back propagation of the error signals and upgrading of training weights. It may have one or more than one layer. Continuity, differentiability, non-decreasing monotony are the main characteristics of this algorithm. It is based on mathematical formula and does not need special features of function in learning process. It also has smooth variation on weight correction due to batch updating features on weights. In training process, it is slow due to more number of learning step but after training of weights, this algorithm produces very fast trained output response. In this application, proposed control algorithm on a DSTATCOM is implemented for compensation of nonlinear loads.

II. SYSTEM CONFIGURATION AND CONTROL ALGORITHM A VSC (Voltage Source Converter) based DSTATCOM is connected to a three phase ac mains feeding three phase linear/nonlinear loads with internal grid impedance which is shown in Fig. 1. The performance of DSTATCOM depends upon the accuracy of harmonics currents detection. For reducing ripple in compensating currents, tuned valued of interfacing inductors (Lf) are connected at ac output of the VSC. A three phase series combination of capacitor (Cf) and a resistor (Rf) represents the shunt passive ripple filter which is connected at point of common coupling (PCC) for reducing the high frequency switching noise of the VSC. The DSTATCOM currents (iCabc) are injected as required compensating currents to cancel the reactive power components and harmonics of the load currents so that loading due to reactive power component/harmonics is reduced on the distribution system. For a considered three phase nonlinear loads with approximate 24 kW, the compensator data are given in Appendix A.

AC Mains Source Ls Impedance Rs VSC Lf Cdc Gating Pulses Control Algorithm vsa vsb vsc iLa iLb iLc iCc vdc iCb iCa

a
isa

isb isc vsc Cf Rf

vsb vsa iLa iLb iLc

Linear, Nonlinear vdc loads

Ripple Filters

isa isb isc

Fig. 1. Schematic diagram of VSC based DSTATCOM

Fig. 2 shows the block diagram of back propagation training algorithm for estimation of reference source currents through the weighted value of load active power and reactive power current components. In this algorithm, phase PCC voltages (vsa, vsb and vsc), source currents (isa, isb and isc), load currents (iLa, iLb and iLc) and dc bus voltage (vdc) are required for extraction of reference source currents (i*sa, i*sb, i*sc). There are two primary modes for operation of this algorithm, first one is a feed forward and second is back propagation of error or supervised learning. Detail application of this algorithm for estimation of various control parameters are given below. A. Estimation of Weighted Value of Average Fundamental Load Active and Reactive Power Components A back propagation training [32, 33] algorithm is used to estimate the three phase weighted value of load active power current components (wap, wbp, wcp) and reactive power current

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
3

components (waq, wbq, wcq) from polluted load currents using feed forward and supervised principle. In this estimation, input layer for three phase (a, b, c) is expressed as, (1) ILap=wo+iLauap + iLbubp +iLcucp ILbp=wo+iLbubp +iLcucp+ iLauap (2) ILcp=wo+iLcucp +iLauap +iLbubp (3) where wo is the selected value of initial weight and uap, ubp, ucp are in-phase unit-templates. In-phase unit-templates are estimated using sensed PCC phase voltages (vsa, vsb, vsc). It is the relation of phase voltage and amplitude of PCC voltage (vt). Amplitude of sensed PCC voltages is estimated as, (4) 2(vsa 2 + vsb 2 + vsc 2 ) vt = 3 In-phase unit templates of PCC voltages (uap, ubp, ucp) are estimated as [13],

u ap =

v sb v sc v sa , u bp = , u cp = vt vt vt

(5)

Extracted values of ILap, ILbp and ILcp are passed through sigmoid function as a activation function and output signals (Zap, Zbp and Zcp) of feed forward section are expressed as, Zap=f(ILap)= 1/(1+e- ILap) (6) Zbp=f(ILbp)= 1/(1+e- ILbp) (7) Zcp=f(ILcp)= 1/(1+e- ILcp) (8)

Iap1=wo1+wapZap+wbpZbp+wcpZcp (9) Ibp1=wo1+wbpZbp+wcpZcp+wapZap (10) (11) Icp1=wo1+wcpZcp+wapZap+wbpZbp where wo1 and wap, wbp, wcp are the selected values of initial weight in hidden layer and updated value of three phase weights using average weighted value (wp) of active power current component as a feedback signals. Updated weight of phase a active power current components of load current wap at nth sampling instant is expressed as, (12) wap(n)=wp(n)+{wp(n)-wap1(n)} f(Iap1) zap(n) wp(n) and wap(n) are the average weighted value of active power component of load currents, updated weighted value of phase a at nth sampling instant respectively and wap1(n), zap(n) are phase a fundamental weighted amplitude of active power component of load current, output of feed forward section of algorithm at nth instant respectively. f(Iap1) and are represented as derivative of Iap1 components and learning rate. Similarly, for phase b and phase c, updated weighted value of active power current components of load current are expressed as, wbp(n)=wp(n)+{wp(n)-wbp1(n)} f(Ibp1) zbp(n) (13) wcp(n)=wp(n)+{wp(n)-wcp1(n)} f(Icp1) zcp(n) (14) Extracted values of Iap1, Ibp1 and Icp1 are passed through sigmoid function as a activation function to estimation of fundamental active components in term of three phase weights wap1, wbp1 and wcp1 as,

Fig. 2. Estimation of reference currents using BP control algorithm

Estimated values of Zap, Zbp and Zcp are fed to hidden layer as input signals. Three phase outputs of this layer (Iap1, Ibp1, Icp1) before activation function are expressed as,

wap1= f(Iap1)= 1/(1+e- Iap1) wbp1= f(Ibp1)= 1/(1+e- Ibp1) wcp1= f(Icp1)= 1/(1+e- Icp1)

(15) (16) (17)

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
4

Average weighted amplitude of fundamental active power components (wp) is estimated using amplitude sum of three phase load active power components (wap1, wbp1, wcp1) divided by three. It is required to realize load balancing features of DSTATCOM. Mathematically, it is expressed as, wp = (wap1+ wbp1+ wcp1) /3 (18)

First order low pass filters are used to separate the low frequency components. k is denoted as scaled factor of extracted active power components of current in the algorithm which shown in Fig. 2. After separating low frequency components and scaled to actual value because output of activation function is between 0 to1, it is represented as wLpA. Similarly weighted amplitude of reactive power components of load currents (waq, wbq, wcq) of fundamental load current are extracted as, ILaq=wo+iLauaq +iLbubq +iLcucq ILbq=wo +iLbubq+iLcucq +iLauaq (19) (20)

instant respectively. f(Iaq1) and are presented as derivative of Iaq1 components and learning rate. Similarly, for phase b and phase c, updated weighted value of reactive power current components of load current are expressed as, wbq(n)=wq(n)+{wq(n)-wbq1(n)} f(Ibq1) zbq(n) (30) wcq (n)=wq(n)+{wq(n)-wcq1(n)} f(Icq1) zcq(n) (31) Extracted values of Iaq1, Ibq1 and Icq1 are passed through activation function to estimation of fundamental reactive component in terms of three phase weights waq1, wbq1 and wcq1 as, waq1= f(Iaq1)= 1/(1+e- Iaq1) (32) wbq1= f(Ibq1)= 1/(1+e- Ibq1) (33) wcq1= f(Icq1)= 1/(1+e- Icq1) (34) Average weight of amplitudes of fundamental reactive power current components (wq) is estimated using amplitude sum of three phase load reactive power components of load current (waq1, wbq1, wcq1) divided by three. Mathematically, it is expressed as, wq = (waq1+ wbq1+ wcq1) /3 (35) First order low pass filters are used to separate the low frequency component. r is denoted as scaled factor of extracted reactive power components in the algorithm which shown in Fig. 2. After separating low frequency components and scaled to actual value because output of activation function is between 0 to1. It is represented as wLqA. B. Amplitude of Active Power Current Components of the Reference Source Currents An error in dc bus voltage is obtained after comparing reference dc bus voltage vdc* and sensed dc bus voltage vdc of a VSC and this error at nth sampling instant is expressed as, vde(n) = vdc*(n) vdc(n) (36) This voltage error is fed to a proportional-integral (PI) controller which output is required for maintaining dc bus voltage of the DSTATCOM. At nth sampling instant, the output of PI controller is as, wdp(n) = wdp(n-1) + kpd {vde(n) vde(n-1)} + kid vde(n) (37) where kpd and kid are the proportional and integral gain constants of the dc bus PI controller. vde(n) and vde(n-1) are the dc bus voltage errors in nth and (n-1)th instant and wdp(n) and wdp(n-1) are the amplitude of active power component of the fundamental reference current at nth and (n-1)th instant. The amplitude of active power current components of the reference source current (wspt) is estimated by addition of output of dc bus PI controller (wdp) and average magnitude of load active currents (wLpA) as, wspt=wdp+wLpA (38) C. Amplitude of Reactive Power Components of the Reference Source Currents An error in ac bus voltage is achieved after comparing amplitude of reference ac bus voltage vt* and sensed ac bus voltage vt of a VSC. Extracted ac bus voltage error vte at the nth sampling instant is expressed as,

(21) ILcq=wo +iLcucq +iLauaq +iLbubq where wo is the selected value of initial weight and uaq, ubq, ucq are the quadrature components of unit template. The quadrature unit template (uaq, ubq, ucq) of phase PCC voltage are estimated using eqn. (5) as,
u aq = ucq = ( ubp + ucp ) 3 ( 3u ap + ubp u cp ) 2 3

, ubq =

(3u ap + ubp ucp ) 2 3

, (22)

Extracted value of ILaq, ILbq and ILcq are passed through sigmoid function as a activation function to estimation of Zaq, Zbq and Zcq. Zaq=f(ILaq)= 1/(1+e- ILaq) (23) Zbq=f(ILbq)= 1/(1+e- ILbq) (24) - ILcq Zcq=f(ILcq)= 1/(1+e ) (25) Estimated values of Zaq, Zbq and Zcq are fed to hidden layer as a input signals. Three phase output of this layer (Iaq1, Ibq1, Icq1) before activation function can be represented as, Iaq1=wo1+waqZaq+wbqZbq+wcqZcq (26) Ibq1=wo1+wbqZbq+wcqZcq+waqZaq (27) Icq1=wo1+wcqZcq+waqZaq+wbqZbq (28) where wo1and waq, wbq, wcq are the selected value of initial weight in hidden layer and updated three weights using average weighted value of reactive power components of currents (wq) as a feedback signals. Updated weights of phase a reactive power components of load currents waq at nth sampling instant is expressed as, waq (n)=wq(n)+{wq(n)-waq1(n)} f(Iaq1) zaq(n) (29) wq(n) and waq(n) are the average weighted value of reactive power component of load currents and updated weights in nth sampling instant and waq1(n), zaq(n) are phase a weighted amplitude of reactive power current component of load currents, output of feed forward section of algorithm at nth

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
5

vte (n) = vt*(n) vt(n) (39) The weighted output of the ac bus PI controller wqq for regulating the ac bus terminal voltage at nth sampling instant is expressed as, wqq(n) = wqq(n-1) + kpt { vte(n) vte(n-1)} + kit vte(n) (40) where wqq(n) is part of reactive power component of source current and it is renamed as wqq. kpt and kit are the proportional and integral gain constants of the ac bus voltage PI controller. The amplitude of reactive power current components of the reference source current (wsqt) is calculated by subtracting the output of voltage PI controller (wqq) and average load reactive currents (wLqA) as, wsqt=wqq-wLqA (41)

is observed under nonlinear loads. B. Performance of DSTATCOM in PFC Mode The dynamic performance of a VSC based DSTATCOM is studied for PFC mode at nonlinear loads. The performance indices are as phase voltages at PCC (vs), balanced source currents (is), load currents (iLa, iLb, iLc), compensator currents (iCa, iCb, iCc), and dc bus voltage (vdc) which are shown in Fig. 3 under varying loads (at t=3.7 s to 3.8 s) conditions. The waveforms of phase a voltage at PCC (vsa), source current (isa) and load current (iLa) are shown in Figs.4 (a-c) respectively. Total harmonic distortion (THD) of the phase a at PCC voltage, source current, load current are found 2.86%, 2.94% and 24.82% respectively. It is observed that the DSTATCOM is able to perform the functions of load balancing and harmonic elimination with high precision.

D. Estimation of Reference Source Currents and Generation of IGBTs Gating Pulses Three phase reference source active and reactive current components are estimated using an amplitude of three phase (a, b and c) load active power current components, PCC voltage inphase unit templates and reactive power current components, PCC quadrature voltage unit templates as, isap=wspt uap, isbp=wspt ubp, iscp=wspt ucp (42) isaq =wsqt uaq, isbq= wsqt ubq, iscq=wsqt ucq (43) Addition of reference active and reactive current components is known as reference source currents and these are given as, isa*= isap+ isaq, isb*= isbp+ isbq, isc*= iscp+ iscq (44)

The sensed source currents (isa, isb, isc) and the reference source currents (isa*, isb*, isc*) are compared and current error signals are amplified through PI current regulators and their outputs are fed to PWM controller to generate the gating signals for IGBTs (Insulate Gate Bipolar Transistors) S1 to S6 of VSC used as a DSTATCOM. III. SIMULATION RESULTS AND DISCUSSION MATLAB with SIMULINK and Sim Power System (SPS) toolboxes is used for the development of simulation model of a DSTATCOM and its control algorithm. The performance of BPT algorithm in time domain for the threephase DSTATCOM is simulated for power factor correction (PFC) and zero voltage regulation (ZVR) modes of operation under nonlinear loads. The performance of control algorithm

Fig.3. Dynamic performance of DSTATCOM under varying nonlinear loads in PFC mode

C. Performance of DSTATCOM in ZVR Mode In ZVR mode, the amplitude of PCC voltage is regulated to the reference amplitude by injecting extra leading reactive

(a)

(b)

(c)

Fig. 4. Waveforms andPersonal harmonic spectra of (a) PCC of purposes, phase a (b) source current ofobtained phase afrom (c) load current of phase a in PFC mode Copyright (c) 2013 IEEE. use is permitted. Forvoltage any other permission must be the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
6

power components. The dynamic performance of DSTATCOM in terms of PCC phase voltages (vs), balanced source currents (is), load currents (iLa,iLb,iLc), compensator currents (iCa,iCb,iCc), amplitude of voltages at PCC (vt) and dc bus voltage (vdc) waveforms are shown in Fig.5 under unbalanced load at time(t) =3.7s to 3.8s duration.

shows the summarized simulation results demonstrating the performance of DSTATCOM. These results show satisfactory performance of DSTATCOM for harmonics elimination and load balancing of nonlinear loads.
TABLE I PERFORMANCE OF DSTATCOM Operating mode Performance parameters PCC voltage (V), %THD PFC mode Supply current (A) , %THD Load current (A) , %THD PCC voltage (V), %THD ZVR mode Source current (A), % THD Load current (A), % THD dc bus voltage (V) Nonlinear load (3-phase uncontrolled rectifier withR and L load ) 237.02V, 2.86% 32.17A, 2.94% 32.58A, 24.82% 239.63V, 3.09% 33.82A, 2.99% 32.94A, 24.94% 700V

IV. EXPERIMENTAL RESULTS A prototype of VSC based DSTATCOM is developed to validate the proposed control algorithm. ABB make current sensors (EL50P1 BB) and voltage sensors (EM010 BB) are used for sensing PCC voltages, dc bus voltage and currents signals. Back propagation training based control algorithm is used for the control of DSTATCOM using a TMS320F240 DSP processor. A Fluke (43B) power analyzer and Agilent make digital oscilloscope (DSO-6014A) are used for recording of steady state and dynamic state tests results respectively on a developed DSTATCOM at nonlinear loads. Hardware implementation data are given in Appendix B.

Fig. 5. Dynamic performance of DSTATCOM under varying nonlinear loads in ZVR mode

(a)

(b)

(c)

Fig. 6. Waveforms and harmonic spectra of (a) PCC voltage of phase a (b) source current of phase a (c) load current of phase a in ZVR mode

Harmonic spectra of phase a voltage at PCC (vsa), source current (isa) and load current (iLa) are shown in Figs. 6(a-c). THDs of the phase a at PCC voltage, source current, load current are observed 3.09%, 2.99% and 24.94% respectively. Three phase PCC voltages are regulated up to rated value. Amplitude of three phase voltages is regulated from 335.2 V to 338.9 V under nonlinear loads. It may be seen that the harmonics distortion of the source current and PCC voltage are within IEEE-519 standard limit of 5%. The PCC voltage is also regulated at different operating condition of loads. Table I

A. Performance of DSTATCOM at Nonlinear Loads Figs. 7 (a-i) show the waveform of phase a PCC voltage (vab) with source currents (isa, isb, isc), load currents (iLa,, iLb, iLc) and compensating current( iCa, iCb, iCc) under nonlinear loads. In Figs. 7(j-l), harmonic distortions of a phase source current, load current and PCC voltage are observed 4.3%, 27.0% and 2.9% respectively. These results show satisfactory performance of the BP control algorithm for harmonics elimination according to IEEE-519 guidelines in order of less than 5%

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
7

B . Dynamic Performance of DSTATCOM Figs. 8 (a)-(c) show the waveform of source currents (isa, isb, isc) load currents (iLa, iLb, iLc) and compensating currents (iCa, iCb, iCc) with PCC line voltage (vab) under unbalanced nonlinear loads. Unbalanced loads can be observed before load injection in phase a. Variation of dc bus voltage (vdc) and PCC voltage with source current (isa), load current (iLa) and compensating current (iCa) are shown Fig. 8 (d-e). Variation of dc bus voltage (vdc) and PCC voltage with source current (isa), load current (iLa) and compensating current (iCa) during load removal are shown Fig. 9 (a, b). These results show balanced source currents when load currents are not balanced and it proves the fast action of control algorithm after load injection. These results show the acceptable performance of control algorithm of DSTATCOM under unbalanced nonlinear loads.

vab

vab
Load injection

isa isb isc

iLa Loadinjection iLb iLc

(a) Ch.1-500V/div, Ch. 2, 3 and 4-(b) Ch.1-500V/div, Ch. 2, 3 and 420A/div, Time axis-10ms/div 20A/div, Time axis-10ms/div

vab

vdc isa
Load injection

iCa iCb iCc

iLa iCa

Load injection

(a)

(b)

(c)

(c) Ch.1-500V/div, Ch. 2, 3 and 4-(d) Ch.1-500V/div, Ch. 2, 3 and 420A/div, Time axis-10ms/div 20A/div, Time axis-10ms/div

vab isa iLa

(d) (e) (f)

Load injection

iCa

(g)

(h)

(i)

(e) Ch.1-500V/div, Ch. 2, 3 and 4- 20A/div, Time axis-10ms/div Fig. 8. Dynamic performance of DSTATCOM during injection of phase a in nonlinear loads (a) vab, isa, isb and isc (b) vab iLa, iLb and iLc (c) vab iCa, iCb and iCc (d) vdc isa, iLa and iCa (e) vab isa, iLa and iCa.

vdc isa iLa iCa

(j) (k) (l) Fig. 7. Performance of DSTATCOM under nonlinear loads (a-c) isa, isb and isc with vab (d-f) iLa, iLb and iLc (g-i) iCa, iCb and iCc (j-l) Harmonic spectra of isa, iLa and vab

vab isa
Load removal

iLa iCa

Load removal

(a) Ch.1-500V/div, Ch. 2, 3 and 4- (b) Ch.1-500V/div, Ch. 2, 3 and 420A/div, Time axis-10ms/div 20A/div, Time axis-10ms/div Fig. 9. Dynamic performance of DSTATCOM during removal of phase a in nonlinear loads (a) (a) vdc isa, iLa and iCa (b) vab isa, iLa and iCa

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
8

V. CONCLUSION A VSC based DSTATCOM has been accepted as the most preferred solution for power quality improvement as power factor correction and to maintain rated PCC voltage. A three phase DSTATCOM has been implemented for compensation of nonlinear loads using BPT control algorithm to verify its effectiveness. The proposed BPT control algorithm has been used for extraction of reference source currents to generate the switching pulses for IGBTs of VSC of DSTATCOM. Various functions of DSTATCOM such as, harmonic elimination and load balancing have been demonstrated in PFC and ZVR modes with DC voltage regulation of DSTATCOM. From simulation and implementation results, it is concluded that DSTATCOM and its control algorithm have been found suitable for compensation of nonlinear loads. Its performance has been found satisfactory for this application because extracted reference source currents exactly tracing the sensed source currents during steady state as well as dynamic conditions. The DC bus voltage of the DSTATCOM has also been regulated to rated value without any overshoot or undershoot during load variation. Large training time in the application of complex system, selection of number of hidden layer in system are the disadvantage of this algorithm APPENDIX A AC supply source: 3-Phase, 415 V (L-L), 50Hz; Source Impedance: Rs=0.04 , Ls=2 mH; Non-linear: Three phase full bridge uncontrolled rectifier with R= 13 and L= 200mH; Rating of VSC=10kVA (Approx. 30% higher from rated), Ripple filter: Rf =5 , Cf = 10F; Switching frequency of inverter=10kHz, Reference dc bus voltage: 700 V; Interfacing inductor(Lf) =2.75mH; Gains of PI controller for dc bus voltage: kpd =3.1, kid=0.9; Gains of voltage PI controller: kpt=2.95, kit =4; Selected initial weights: wo=0.4, wo1=0.2; Learning rate ()=0.6; Cut off frequency of low pass filter used in dc bus voltage = 15 Hz; Cut off frequency of low pass filter used in ac bus voltage = 10Hz.
[3]

[4] [5] [6]

[7] [8]

[9] [10] [11]

[12] [13] [14]

[15]

[16]

[17]

APPENDIX B AC supply source: 3Phase, 225 V (L-L), 50 Hz; Non-linear loads: Three phase full bridge uncontrolled rectifier with R= 43 and L=200mH; dc bus capacitance: 1650F; Reference dc bus voltage: 400 V; Interfacing inductor (Lf) =3 mH; Ripple filter: Rf =5 , Cf = 10F, Cut off frequency of low pass filter used in dc bus = 15 Hz; Selected initial weights: wo=0.4, wo1=0.2; Learning rate ()=0.6; Sampling time with used digital plate from (ts) =50s. REFERENCES
[1] [2] R. C. Dugan, M. F. McGranaghan and H. W. Beaty, Electric Power Systems Quality, 2ed Ed., McGraw Hill, New York, 2006. Alfredo Ortiz, Cristina Gherasim, Mario Manana, Carlos J. Renedo, L. Ignacio Eguiluz and Ronnie J. M. Belmans,Total harmonic distortion

[18] [19] [20]

[21] [22] [23] [24]

decomposition depending on distortion origin,IEEE Transactions on Power Delivery, vol. 20, no. 4, pp. 2651-2656, October 2005. Tzung Lin Lee and Shang Hung Hu,Discrete frequency-tuning active filter to suppress harmonic resonances of closed-loop distribution power systems,IEEE Transactions on Power Electronics, vol. 26, no. 1, pp. 137-148, January 2011. K. R. Padiyar, FACTS Controllers in Power Transmission and Distribution, New Age International, New Delhi, 2008. IEEE Recommended Practices and requirement for Harmonic Control on electric power System, IEEE Std.519, 1992. Tzung-Lin Lee, Shang-Hung Hu and Yu-Hung Chan,DSTATCOM with positive-sequence admittance and negative-sequence conductance to mitigate voltage fluctuations in high-level penetration of distributed generation systems,IEEE Transactions on Industrial Electronics, vol. 60, no. 4, pp. 1417-1428, April 2013. B. Singh, P. Jayaprakash and D.P. Kothari,Power factor correction and power quality improvement in the distribution system,Journal of Electrical India, pp. 40-48, April, 2008. Jinn-Chang Wu, Hurng Liahng Jou, Ya Tsung Feng, Wen Pin Hsu, Min Sheng Huang, and WenJet Hou,Novel circuit topology for three-phase active power filter,IEEE Transactions on Power Delivery, vol. 22, no. 1, pp. 444-449, January 2007. Z Yao and L. Xiao,Control of single-phase grid-connected inverters with nonlinear loads,IEEE Transactions on Industrial Electronics, vol. 60, no. 4, pp. 1384-1389, April 2013. Ali A. Heris, E. Babaei and Seyed H. Hosseini,A new shunt active power filter based on indirect matrix converter,in Proc. of 20th Iranian Conference Electrical Engineering, 2012, pp.581-586. M. Sadeghi, A. Nazarloo, S.H. Hosseini and E. Babaei,A new DSTATCOM topology based on Stacked Multicell converter,in Proc. of 2nd Power Electronics, Drive Systems and Technologies Conference, 2011, pp.205-210. Grzegorz Benysek and Marian Pasko, Power Theories for Improved Power Quality, Springer-Verlag, London, 2012. B. Singh and J. Solanki,A comparison of control algorithms for DSTATCOM,IEEE Transactions on Industrial Electronics, vol. 56, no. 7, pp. 2738-2745, July 2009. C.H. da Silva, R.R. Pereira, L.E.B. da Silva, G. Lambert-Torres, B.K. Bose and S.U. Ahn,A Digital PLL scheme for three-phase system using modified synchronous reference frame, IEEE Transactions on Industrial Electronics, vol.57, no.11, pp.3814-3821, Nov. 2010. Salem Rahmani, Abdelhamid Hamadi and Kamal Al-Haddad,A Lyapunov-function-based control for a three-phase shunt hybrid active filter, IEEE Transactions on Industrial Electronics, vol. 59, no. 3, pp. 1418-1429, March 2012. S. Rahmani, N. Mendalek and K. Al-Haddad,Experimental design of a nonlinear control technique for three-phase shunt active power filter,IEEE Transactions on Industrial Electronics, vol.57, no.10, pp.3364-3375, Oct. 2010. S.N. Sivanandam and S.N. Deepa,Principle of Soft Computing,Wiley India Ltd, New Delhi, 2010. Jyh Shing Rogor Jang, Chuen Tsai Sun and Eiji Mizutani,in Proc. of Neuro fuzzy and soft computing:A Computational Approach to Learning and Machine Intelligence, Person Education Asia, Delhi, 2008. Parmod Kumar and Alka Mahajan,Soft computing techniques for the control of an active power filter,IEEE Transactions on Power Delivery, vol. 24, no. 1, pp.452-461, January 2009. Avik Bhattacharya and Chandan Chakraborty,A shunt active power filter with enhanced performance using ANN-based predictive and adaptive controllers,IEEE Transactions on Industrial Electronics, vol. 58, no. 2, pp. 421-428, February 2011. L. L. Lai, W.L. Chan and A.T.P. So,A two-ANN approach to frequency and harmonic evaluation, in Proc. of Fifth International Conference on Artificial Neural Networks, 1997, pp.245-250. Xinrong Mao,The harmonic currents detecting algorithm based on adaptive neural network,in Proc. of Third International Symposium on Intelligent Information Technology Application, 2009, vol.3, pp.51-53. Jianhua Wu, Hali Pang and Xinhe Xu,Neural-network-based inverse control method for active power filter system,in Proc. of IEEE International Symposium on Intelligent Control, 2006, pp.3094-3097. Hyo-Sung Ahn, YangQuan Chen, and Kevin L. Moore,Iterative learning control: brief survey and categorization, IEEE Transactions on

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
9 Systems, Man and CyberneticsPart C: Applications And Reviews, vol. 37, no. 6, pp.1099-1121, November 2007. Badong Chen, Songlin Zhao, Pingping Zhu and Jose C. Prncipe,Quantized kernel least mean square algorithm,IEEE Transactions on Neural Networks and Learning Systems, vol. 23, no. 1, January 2012. Y. Hao, X. Tiantian, S. Paszczynski and B.M. Wilamowski, Advantages of radial basis function networks for dynamic system design,IEEE Transactions on Industrial Electronics, vol. 58, no.12, pp. 5438-5450, Dec. 2011. X. Jing and L. Cheng,An Optimal-PID control algorithm for training feed-forward neural networks, IEEE Transactions on Industrial Electronics, vol. 60, no.6, pp.2273-2283, Jun. 2013. Fu Guangjie and Zhao Hailong,The study of the electric power harmonics detecting method based on the immune RBF Neural Network,in Proc. of Second International Conference on Intelligent Computation Technology and Automation, 2009, vol. 1, pp.121-124. J. Mazumdar, R.G. Harley and G.K. Venayagamoorthy,Synchronous reference frame based active filter current reference generation using neural networks,in Proc. of 32nd IEEE Annual Conference on Industrial Electronics, 2006, pp.4404-4409. J. Mazumdar, R.G. Harley, F. Lambert and G.K. Venayagamoorthy,A novel method based on neural networks to distinguish between load harmonics and source harmonics in a power system, in Proc. of IEEE Power Engineering Society Inaugural Conference and Exposition in Africa, 2005 pp.477-484. A. Zouidi, F. Fnaiech, K. Al-Haddad and S. Rahmani,Artificial neural networks as harmonic detectors,in Proc. of 32nd Annual Conference on IEEE Industrial Electronics,2006, pp.2889-2892. Insung Jung, and Gi Nam Wang,Pattern classification of backpropagation algorithm using exclusive connecting network,Journal of World Academy of Science, Engineering and Technology, vol. 36, pp. 189-193, 2007. Chen Ying and Lin Qingsheng,New research on harmonic detection based on neural network for power system,in Proc. of Third International Symposium on Intelligent Information Technology Application, 2009, vol. 2, pp.113-116. Technology (IIT), New Delhi, India, in 1979 and 1983, respectively. In 1983, he joined Department of Electrical Engineering, University of Roorkee, as a Lecturer. He became a Reader there in 1988. In December 1990, he joined Department of Electrical Engineering, IIT Delhi, New Delhi, India, as an Assistant Professor, where he has became an Associate Professor in 1994 and a Professor in 1997.He has guided 41 Ph.D. dissertations 130 ME/M.Tech theses and 60 BE/B.Tech projects. He has been granted one US patent and filed nine Indian patents. He has executed more than seventy sponsored and consultancy projects. His fields of interest include power electronics, electrical machines, electric drives, power quality, FACTS (Flexible AC Transmission Systems), HVDC (High Voltage Direct Current) transmission systems and renewable energy generation. Prof. Singh is a Fellow of the Indian National Academy of Engineering (INAE), the National Science Academy (NSc), the Indian Academy of science (IASc), the Institute of Engineering and Technology (IET), the Institution of Engineers (India) (IE(I)), and the Institution of Electronics and Telecommunication Engineers (IETE). Sabha Raj Arya (M12) received his B.E. in Electrical Engineering from Government Engineering College Jabalpur, India, in 2002 and M.Tech. in Electrical Engineering with specialization in Power Electronics and ASIC Design from the Motilal National Institute of Technology, Allahabad, India in 2004. In July 2004, he joined Department of Electrical Engineering, Kalinga Institute of Industrial Technology (Deemed university) Bhubaneswar as a lecturer. In December 2006, he joined Department of Electrical Engineering, Sardar Vallabhbhai National Institute of Technology, Surat, as an Assistant Professor. He is currently at Department of Electrical Engineering, Indian Institute of Technology, Delhi, India, where he is pursuing towards Ph.D. degree under Quality Improvement Programme. His fields of interest include power electronics, power quality and design of Custom power devices.

[25]

[26]

[27] [28]

[29]

[30]

[31] [32]

[33]

Bhim Singh (SM99, F10) received Bachelor of Engineering (Electrical) degree from University of Roorkee, Roorkee, India, in 1977, and M.Tech. (Power Apparatus and Systems) and Ph.D. degrees from Indian Institute of

Copyright (c) 2013 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.