an add-on self-tuning control system

2378 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 61, NO. 5, MAY 2014

An Add-On Self-Tuning Control Systemfor a UPFC Application

Urvi Malhotra, Student Member, IEEE, and Ramakrishna Gokaraju, Member, IEEE

AbstractThis paper presents an add-on self-tuning (ST) con-trol scheme for a Unified Power Flow Controller (UPFC) to assistits conventional PI control system in damping power oscillations.The ST control algorithm is based on a Pole Shift (PS) techniquethat has previously been successfully applied in Adaptive PowerSystem Stabilizers (APSSs). For a wide range of operating con-ditions, the conventional PI-UPFC unless retuned suffers frominsufficient/nonoptimal damping performance. To overcome thisproblem, the authors propose supplementing the PI controllerswith an ST feedback loop comprised of an identifier and a PScontrol algorithm. With a CRLS identifier tracking the systemconditions online, this scheme eliminates the need for PI param-eter retuning. Further, to correctly identify system parametersduring large disturbances, a constrained recursive least squares(CRLS) identification procedure is adopted here. An improveddamping performance with the proposed add-on scheme assistingthe nonoptimal PI-UPFC is demonstrated for a two-area powersystem.

Index TermsAdaptive control, flexible AC transmission sys-tems (FACTS), power systems control, self-tuning (ST) control.

I. INTRODUCTION

AN ever-increasing demand in electricity accompaniedwith restricted building of new transmission facilities hascaused todays interconnected power systems to be overloaded.As a consequence, low frequency (0.1 to 0.8 Hz) inter-areaoscillations are often observed when power systems intercon-nected through weak tie lines are subjected to disturbances suchas faults, line outages and sudden load changes. Previously,power system stabilizers (PSS) installed at generator locationswere used to damp such system oscillations. However, PSSsare localized stabilizers and hence best suited for damping localoscillatory modes in the power system. Since inter-area modesare related to the dynamics of an inter-connected power system,damping control methodologies meant for transmission pathsevolved. The advancements in high-power semiconductors ledto the advent of flexible AC transmission systems (FACTS)technology. One such FACTS devicethe Unified Power FlowController (UPFC), proposed by Gyugyi in 1991 [1]is a ver-satile device capable of simultaneously controlling line power

Manuscript received July 9, 2012; revised February 16, 2013; acceptedMay 7, 2013. Date of publication July 9, 2013; date of current versionOctober 18, 2013. This work was completed at the University of Saskatchewanas part of the M.Sc. thesis research work of U. Malhotra. The work ofR. Gokaraju was supported by an NSERC Discovery Grant.

U. Malhotra is with the Saskatchewan Power Corporation in Regina, SK S4P0S1, Canada.

R. Gokaraju is with the University of Saskatchewan, Saskatoon, SK S7N5A5, Canada (e-mail: [email protected]).

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TIE.2013.2272272

flows as well as providing voltage support to the connected bus.In addition to being capable of controlling line power flows,the UPFC can also damp out power oscillations. This has beensuccessfully demonstrated in the past with a suitable UPFCdamping control system [2], [3]. A majority of these controlsystems has employed conventional-PI (or lead-lag) controllersas feedback controllers, providing satisfactory damping perfor-mance around a specific operating point. In [4], the FACTScontroller design was based on fixed-gain PI controller whereinthe controller gains remain fixed during the FACTS devicesoperation. The tuning of PI controllers is a complex task in abulk power system involving number of switching devices [5].

PI controllers are fixed-parameter based controllers. Thedrawback of such PI controllers is that their performance de-grades as the system operating conditions change [6]. In [7], asliding-mode control approach is proposed as a robust con-troller. However, the controllers performance is dependent onthe transfer function, which in the case of a power systemmodeled with a FACTS device, is difficult to find (to accuratelyrepresent the complexity of the system). Furthermore, given thedynamic and nonstationary characteristic of a power system,PI controllers unless retuned cannot ensure consistent dampingperformance over a wide range of operating conditions. Incontrast, artificial neural networks (ANNs) and fuzzy logic havebeen proposed to adapt the controller gains [8][10]. However,such approaches either require offline training or depend onthe inference rules. Some of the difficulties associated withthe tuning of membership functions in the fuzzy PID typecontrollers are very well explained in [10].

An artificial immune system (AIS) based modern controlapproach is proposed in [11] for generator excitation systems tosolve the power quality problems, but the design and analysisprocedures are complex and difficult. Recently [12] describedan adaptive controller for synchronous reluctance motor drivesto control the current, speed and position. The adaptive con-troller in [12] provided good tracking during steady-state con-ditions and small load disturbance conditions.

The proposed self-tuning (ST) controller in this paper differsfrom the ANN and Fuzzy Logic control approaches. UnlikeANNs, it does not require offline training nor is dependent oninference rules, which are essential for fuzzy approaches. Itinstead addresses these deficiencies by exploiting the principlesof indirect adaptive control theory. Indirect adaptive controlprinciple consists of an identifier and controller. The identifierworks on an estimated plant model of a power system; anauto-regressive moving average (ARMA) model. It identifiesthe ARMA parameters online and uses parameter estimates toadaptively tune the corresponding control parameters thereby

0278-0046 2013 IEEE

MALHOTRA AND GOKARAJU: ADD-ON SELF-TUNING CONTROL SYSTEM FOR UPFC APPLICATION 2379

yielding a satisfactory control performance. The adaptive iden-tifier and controller described works well for large disturbancesas well.

The proposed ST controller consists of an adaptive con-strained recursive least squares (CRLS) identifier and a poleshift (PS) controller. The PS controller is self-tuning type,which means that it adapts to the power system conditionsonline. The self-tuning control technique based on PS con-trol logic was previously implemented on APSSs and certainFACTS devices [13][15], however this is the first time it isbeing developed for a UPFC.

The performance of the proposed ST control scheme dependson accurate determination of the ARMA parameters. It waspreviously reported in [16] that an RLS identifier at the startof the estimation process gives poor estimation results dueto large initial transients present in the system. Such rapidparameter fluctuations result in the ringing of the controlleroutput. If the control is generated on the basis of these highlyvarying estimated coefficients of the ARMA model, it can leadto unwanted initial power swings in the power system.

To achieve parameter tracking, [16] proposes the use of reg-ularized constant trace algorithm on a Kalman filter identifier.The authors also use a larger 12th order autoregressive (AR)model to capture the system dynamics.

Reference [15] proposes a random walk (RW) method for theRLS identifier for better parameter tracking. The RW algorithmis enabled for a specified duration following a large disturbanceto obtain smoother variations in the identified coefficients. Adead-zone is also created for the controller to give time forthe RW identifier to estimate the parameters. Furthermore, theauthors use a large 10th order ARMA model for the plant and afixed pole-placement method to generate the control signal.

For self-tuning control, it is important to keep the compu-tation time low. A higher order model (10th or 12th model)means more calculations for the identifier and the controller.Therefore, using a reduced order model, which can capture theessential dynamics, was one of the objectives of this paper.

In this paper, a simple third-order ARMA model is proposedto represent the dynamic oscillations in the power system and ithas been found to be sufficient to represent the low frequencypower oscillations in the power system. This is because athird-order model has three poles: a pair of complex polesand a real pole. The complex poles represent the oscillatorybehavior of the power oscillations and could be used to dampthe overshoot using the PS controller; whereas the real rootrepresents the decaying part of the oscillatory response andcould be controlled with the PS controller to achieve the desiredsteady-state response.

This paper also proposes a computationally efficient modi-fication to the standard RLS identifier known as constrained-RLS (CRLS identifier) discussed in Section III-A. The methodis simple and involves only a minor calculation step. The CRLSidentifier smooths out parameter variations efficiently duringlarge disturbances (i.e., large power swings) following a faultcondition. Using the proposed approach, there is no need to cre-ate a control dead-zone but instead the controller and identifierare allowed to function all throughout the disturbance conditionin unison. It also implies that with the CRLS identifier, the

Fig. 1. Single line diagram of a UPFC installed in a two-area power system.

controller is able to adapt to drastic or sudden disturbances inthe power system configuration and still ensure a stabilizingcontrol performance.

The main contributions of this work are: 1) An adaptive add-on control scheme is proposed on top of a PI-control loop. Thescheme retains the functionality of the existing PI controllerswhile improving the overall damping performance with an add-on supplementary self-tuning control loop. It is worth notinghere that the previous literature cited the inability of the self-tuning controllers to handle large disturbance conditions. 2) Useof a constrained-RLS identifier instead of the classical RLSidentifier to smoothly track system parameters during largedisturbances. The CRLS identifier prevents ringing of thecontrol output during initial post-fault power swings therebyeliminating the need to disable the control action right after afault occurs.

The paper is organized as follows: Section II discusses theUPFC operation and control system model. A brief discussionof the proposed ST control scheme is presented in Section III.The proposed overall UPFC control system is applied to a multi-area test system and the results are presented in Section IV;Section V reports the conclusions of this work.

II. UPFC: OPERATION AND CONTROL SYSTEM MODEL

Fig. 1 shows a schematic representation of a UPFC installedat the sending-end bus, s that serves as an inter-tie of a two-area transmission network. In the past, a majority of literaturesuch as the ones described in [1], [2] have modeled the UPFCas a series reactance with a set of active and reactive nodalpower injections or by two ideal voltage sources. Such modelswere primarily chosen for their mathematical simplicity wherethe converter switching characteristics were ignored. However,it is important to include the switching characteristics espe-cially when studying the interaction of the converters and thepower system in response to system faults. Thus, in this workthe UPFC converter is modeled in PSCAD-EMTDC using atwo-level, six-pulse voltage sourced converter (VSC) topology


with the thyristor firing pulses generated through a pulse widthmodulation (PWM) switching technique.

A. Operation

A UPFC consists of two identical converters, the shunt VSCand the series VSC connected in shunt and series with the actransmission network, respectively. Each converter is connectedto the ac system through corresponding coupling transformers,Tsh and Tse and are operated from a common dc link supportedby a dc capacitor, C.

In general, the operation of a UPFC is governed by its fourcontrol inputs: msh, sh,mse, se that are governed by firingcommands given to its thyristors switches where m and de-note converter modulation index and phase angle, respectively.The primary function of the UPFC is provided by the seriesVSC that generates a voltage Vinj(t) = |Vinj | sin(tinj)at fundamental frequency () with variable amplitude (0 |Vinj | |Vinj |max) and phase angle (0 inj 2) that isadded to the ac system by the series coupling transformer,Tse. On the other hand, the shunt VSC serves two purposes.It primarily provides the real power demanded by the seriesVSC from the ac system via the common dc link. This is madepossible by keeping the dc link capacitor, C charged at alltimes, replicating an energy source for real power exchange. Italso acts as an independent reactive power compensator therebymaintaining the system bus voltage at a specified value. Withits four control inputs, the UPFC can be commanded to forcean appropriately varying line power to effectively damp poweroscillations.

B. Damping Control

The control of a UPFC can be divided into two partsthecontrol of the shunt VSC and the series VSC where each VSCis controlled by two control inputsmodulation index, m andphase angle, . The shunt and series VSCs can be operatedin different control modes [1] where a particular mode ischosen depending upon the application objective. For dampingof inter-tie oscillations, the series VSC is operated in AutomaticPower Flow Control mode as shown in Fig. 2(a). The d-qcontrol method suggested in [17] is adopted wherein the desiredPdesired and Qdesired are compared with the measured lineflows, Pmeas and Qmeas and the corresponding deviations inreal and reactive power (P and Q) are used to drive thed-q components (Vd and Vq) of series injected voltage, Vinj .Note here that mse, msh and se, sh correspond to steady-state converter settings that can be easily obtained dependingupon the desired inter-tie real and reactive power flows.

For the purpose of tightly regulating the connected busvoltage (< 5%), the shunt VSC is operated in Automatic VoltageControl mode as shown in Fig. 2(b). The bus voltage regulationis accomplished by varying the shunt modulation index, mshwhile the dc link voltage regulation is achieved by varying thephase angle order, sh.

The UPFC control of Fig. 2 is based on PI control lawthat is inadequate in tracking the system changes and becomeincreasingly less effective in damping power oscillations with a

Fig. 2. Overall UPFC control system for a damping application. (a) SeriesVSC control system; (b) Shunt VSC control system.

set of non-optimal PI parameters. On the contrary, the proposedST controller copes to system changes and offers an excellenttracking capability even for large disturbances such as three-phase faults. Further, in a UPFC the series VSC can directlyinfluence the power flow and hence the oscillatory modes.Keeping the aforementioned points in mind, the proposed STcontroller is added as a supplementary control loop to the seriesVSC to modulate mse as shown in Fig. 2(a) with the real powerdeviations, P as the control input.

III. SELF-TUNING (ST) CONTROLLERThe major drawback of the UPFC control system discussed

in Section II is that it is solely based on PI controllers. As men-tioned before, PI controllers create the need for frequent tuningand hence are inefficient in damping power oscillations underdifferent operating conditions. On the contrary, the proposed STcontroller being adaptive is independent of tuning. With the STcontroller in the feedback loop, it is expected to significantlyimprove the damping capability offered by the overall UPFCcontrol system.

Fig. 3 represents the schematic overview of the ST controller.Note that here the measured output y(t) from the power systemis the oscillations in real power P while the controller outputu(t) is the damping input mse supplied to the series VSC ofthe UPFC.

A. System Estimation: Adaptive CRLS IdentifierAmongst several identification algorithms used for online

parameter estimation, the classical recursive least squares(RLS) method has the advantage of simple calculation andgood convergence properties [18]. The algorithm assumes theprocess (here, power system including the UPFC) to be de-scribed by a discrete ARMA model of the form

A(z1)y(t) = B(z1)u(t) + e(t) (1)


Fig. 3. Self-tuning (ST) control.

where A(z1) and B(z1) are polynomials in backward shiftoperator z1 and are defined as

A(z1) = 1 + a1z1 + a2z2 + . . . anazna

B(z1) = b1z1 + b2z2 + . . . bnbznb

na and nb are the orders of the polynomials A(z1) and B(z1)and variables y(t), u(t) and e(t) are the system output, systeminput and white noise, respectively. Equation (1) can be re-written in a form suitable for identification as

y(t) = T (t)(t) + e(t) (2)y(t) = y(t|, ) + e(t) (3)

where y(t|, ) is the prediction of y(t) based on estimated sys-tem parameter vector, T (t) and measurement variable vector,(t) defined as

(t) = [a1 a2 a3 . . . ana b1 b2 b3 . . . bnb ]T (4)

T (t) = [y(t 1) y(t 2) . . . y(t na) u(t 1) . . . u(t nb)] (5)

where T (t) denotes the data vector storing the last na samplesof the system output and the last nb samples of the system input.The prediction error, (t) is given by

(t) = y(t) y(t|, ) = e(t). (6)The RLS criterion determines the most likely value of (t) thatminimizes the sum of the squares of the prediction error

J(N) =1

N

k=Nk=1

[e(t)]2 . (7)

The system parameter vector (t) can be calculated by thefollowing set of recursive equations:

(t) = (t 1) +K(t)[y(t) T (t 1)(t)

](8a)

where the weighing factor for the last identified parameters isequal to one and that for the prediction error is given by themodification coefficients (or gain matrix) K(t)

K(t) =P (t 1)(t)

(t) + T (t)P (t 1)(t) (8b)

P (t) =1

(t)

[1KT (t)(t)]P (t 1) (8c)

Fig. 4. Excursions in system response following a three-phase fault.

where (t) is the time varying forgetting factor, P (t) is thecovariance (of error in estimate) matrix, and K(t) is the gainvector. The forgetting factor (t) is calculated as

(t) = 0(t 1) + (1 0); 0 < 0 < 1. (8d)In every iteration, the RLS algorithm uses a set of old and

new system measurements to predict the system output y(t)and calculate the prediction error (t). Based on (t) and thepast history of measurements stored in error covariance matrixP (t 1), it then calculates a set of modification coefficientsK(t). Using K(t) it arrives at a new set of identified systemparameters (t) as in (8a). At the end of the current iterationthe information of the new measurements are stored for theiruse in the next sample. However, if a classical RLS identifieris extended to a power system imposed by large disturbances(faults), it experiences undesired rapid parameter fluctuations.Fig. 4(a) illustrates as an example, a noisy type first-swinginter-area power flow and corresponding ringing of controloutput [Fig. 4(b)] recorded when a three-phase-ground faultwas applied to the neighboring area.

The skewed response is due to the imperfections of theRLS algorithm as it is meant for a system with constant orslow varying parameters [18]. To overcome this drawback, aconstrained-RLS (CRLS) identification technique is used thatmodifies (8a) with the inclusion of a constraint term (t)

(t) = (t 1) +K(t)[y(t) T (t 1)(t)

](t) (9)

where (t) is calculated at each sampling instant as follows:

(t) =

{1.0 if N2N1 00N2

if N2N1 > 0(10)

where

N1 =(t)

2

N2 =K(t) [y(t) T (t 1)(t)]

2

.2 is the norm of the corresponding vector and 0 is a positiveconstant which determines the rate of parameter update. At thetime of fault inception (and the duration shortly after) on apower system, the ratio N2/N1 progresses quickly to be greaterthan 0. During this time period (t) takes the value 0/N2reducing the estimation rate and therefore contributing to asmooth controller action. After fault removal when the powersystem is moving to a stable operating point, the ratio N2/N1


Fig. 5. Excursions in a-parameters following a three-phase fault. (a) a-parameters with RLS identifier; (b) a-parameters with CRLS identifier.

gradually progresses to be less than 0 making (t) = 1. At thistime, the constrained-RLS algorithm automatically switchesback to the standard-RLS algorithm. It is worth noting that alarger 0(i.e., 0 > 0.75) slows down the estimation rate whichis undesirable; while choosing a smaller 0(i.e., 0 < 0.5) cancause the identifier to respond to even minor disturbances.Therefore, (t) is chosen as 0.5 to achieve optimal trackingcapability. The (t) in (9) initiates a smooth controller responsea scan be seen in Fig. 5. Fig. 5 gives a plot of varyinga-parameters versus time for the example illustrated in Fig. 4.The minimization of drastic parameter variation by the CRLSidentifier and its successful application to a faulted power sys-tem is evident from the power flow response shown in Fig. 5(c)as opposed to the power flow response of Fig. 4(a).

It should be pointed out that for all the studies using the STcontroller, a third-order ARMA system model is used in theidentification process. When a two-area system is subject toa disturbance it exhibits inter-area mode of oscillations in therange of 0.1 to 0.8 Hz, and a third-order model has been foundto be sufficient to represent this single mode of oscillation asdiscussed in the introduction.

A second-order model is not useful because it can only rep-resent either the oscillatory (pair of complex conjugate poles)or the non-oscillatory part (with real poles). With a 4th ordermodel, the system could be identified as two pairs of complexconjugate poles, or one pair of complex poles and two realpoles, or four real poles. Two pairs of complex poles can onlyrepresent the oscillatory response, whereas two or four real-poles are not needed to represent the non-oscillatory part of asingle mode inter-area oscillation.

A 5th, 7th, 9th or higher odd numbered models could be used.The 5th order, 7th and 9th order models were tried but therewas no noticeable damping improvement compared to a 3rdorder model. Therefore, 3rd order model was chosen to keepthe number of computations less for hardware implementation.

B. Control Algorithm: Pole Shift ControlOnce the system parameters are identified by the CRLS

identifier, the control signal is calculated based on the discretesystem model of (1) by assuming the transfer function of thefeedback loop is of the form

u(t)

y(t)= G(z

1)F (z1)

(11)

where

F (z1) = 1 + f1z1 + f2z2 + . . . fnf znf

G(z1) = g0 + g1z1 + g2z2 + . . . gngzng

and nf = nb 1, ng = na 1. The concept of the PS controlis discussed in detail in [13], [14]. It is based on the minimumvariance (MV) control and the pole assignment (PA) control.However, unlike the PA method where the closed-loop poles areassigned to specific locations, in the PS method the closed-looppoles are shifted radially toward the center of the unit circle inz-domain by a factor, (less than one)

T (z1) = A(z1)F (z1) +B(z1)G(z1). (12)

The characteristic equation of (12) of the closed-loop systemtakes the form of characteristic equation A(z1) of the open-loop system with the pole shifted by a scalar factor . Thus(10) takes the form

T (z1) = A(z1)F (z1) +B(z1)G(z1) = A(z1).(13)

Rearranging (13) in matrix form in terms of control parameters{f} and {g} gives

1 0 0 b1 0 0a1 1 0 b2 b1 0 a1 b2

ana ana1 1 bnb bnb1 b10 ana a1 0 bnb b20 0 a2 0 0 b3 0 0 ana 0 0 bnb

f1

fnfg0

gng

=

( 1)a1(2 1)a2

(na 1)ana00

or

M () = L() (14)


where the elements of matrix M are the parameters {a}, {b}that are identified by the identifier every sampling period. If thevalue of pole shift factor, , is known, (14) can be solved forthe control parameters {f} and {g} following which controloutput, u(t) can be obtained using (11). It can be observed thatthe control signal is a function of at any time t denoted as t.The control signal, u(t, t) can be expressed in a Taylor seriesof factor t at a particular point 0 as

u(t, t) = u(t, 0) +i=1

1

i!

[(i)u(t, t)

(i)t

](t 0)i.

(15)From (11) and (14), the control signal can be expressed as

u(t, t) = XT (t)(t) = X

T (t)M1L(t) (16)

where X(t) = [u(t 1).. u(t nf ) y(t) y(t 1)..y(t ng]T is the measurement variable vecor. The ith orderdifferential of u(t, t) with respect to t becomes[

(i)u(t, t)

(i)t

]t=0

= XT (t)M1L(i)(0) (17)

where i represents the order of the differentiation. Now (15) canbe expressed as

u(t, t) = u(t, 0) +

nai=1

siit (18)

where si = (1/i!)XT (t)M1L(i)(0).Once the input is computed at time t, the system predicted

output y(t+ 1) at time t+ 1 can be calculated as follows:

y(t+ 1) = XT (t)+ b1 [u(t, t)] (19)

where = [b2 b3 bnb a1 a2 ana ] is the identi-fied parameter vector.

The essence of PS algorithm is based on . The PS factor is calculated iteratively so as to minimize the one time-stepahead prediction error in system output y(t) given as

mint

J(t+ 1, t) = E [y(t+ 1) yr(t+ 1)]2 (20)

where J(t+ 1, t) defines the performance index, E is theexpectation operator, y(t+ 1) is next time-step predicted out-put and yr(t+ 1) is the next time-step reference output. Theminimization function in (20) is a constrained optimizationroutine subjected to a stability and control constraint. Thestability constraint restricts t to be within ((1/), (1/)) toensure closed-loop stability at all times, where represents theabsolute value of the largest root of T (z1). Further, the controlconstraint emphasizes control signal u(t, t) to satisfy umin u(t, t) umax, where umin and umax are the minimum andmaximum control signal boundaries, respectively. In a UPFCapplication where the PS control signal modulates mse bya supplementary control loop, mseis 1.0 to prevent over-modulation of the series-VSC.

Fig. 6. SISO discrete system with PS controller in feedback loop.

C. ST Controller on Discrete Test System

The performance of the ST control scheme with the proposedCRLS identifier is first investigated on a single input singleoutput (SISO) discrete system as shown in Fig. 6. The plantis an unstable third-order model. During the test, the referenceinput, uref is changed every 1 s to assess the performance of theST controller. That is, to excite the plant for identification andcontrol process, a step-input limited to 0.1 units is simulated.The term e(t) in Fig. 6 represents addition of white noise ofmagnitude 1% to the system.

The discrete system is of the form

y(t) 2.54y(t 1) + 2.69y(t 2) 1.05y(t 3)= 0.5u(t 1) + 0.1u(t 2) + 0.25u(t 3) + e(t) (21)

that corresponds to a plant model of

B(z1)A(z1)

=0.5z1 + 0.1z2 + 0.25z3

1 2.54z1 + 2.69z2 1.05z3 . (22)

The open loop plant poles are at 0.82 and 0.86 j0.74.Thus, the complex conjugate poles of the open loop systemare located outside the unit circle and poses the need for astabilizing feedback signal, u(t). With the PS controller in thefeedback loop (Fig. 6) the control signal is calculated using(18) based on the identified plant coefficients using the CRLSidentifier of (9). It is worth noting here that although the plantcoefficients are defined in (22) they appear unknown to theidentifier. For this test, convergence of the identified parametersto the actual parameters of (22) will essentially demonstratethe success of the identifier. The time-domain system responsereference input uref is plotted in Fig. 7. As can be inferred, thecontrolled plant is able to track the reference signal effectively.Further, the corresponding identified plant coefficients by theCRLS identifier are plotted in Fig. 7(c) and (d). The CRLSidentification not only converges quickly but also provides asmooth parameter tracking during input switching instances.The controller was able to perform satisfactorily even duringswitching instants due to the CRLS identifier.

Furthermore, the closed-loop stability as ensured by the PScontrol algorithm can be confirmed with the location of theopen and closed loop poles as illustrated in Fig. 8. Open looppoles are shown by while closed loop poles are shown byx. The identified open loop poles are at {0.7437, 0.7988j0.7203} and the closed loop poles are at {0.4821, 0.5083j0.501} indicating the evidence of radial pole-shifting processby the pole-shift factor to achieve closed loop stability alongwith a desired system response.


Fig. 7. System time response to step changes in reference input. (a) Plantoutput; (b) pole shift factor ; (c) plant parameter a; (d) plant parameter b.

Fig. 8. Dynamic pole-shifting by pole-shift factor .

IV. SIMULATION RESULTS

The proposed ST controller is introduced as an add-onscheme to the existing PI UPFC control system. The objectiveof this paper is to test the damping enhancement capability ofthe ST controller operating in unison with the non-optimal PIcontrollers. Kundurs two-area four-machine test system [1] ofFig. 9 investigated by researchers for inter-area mode analysishas been chosen for the study. The system loading is such thatnet power is exported from Area 1 to Area 2. Three oscilla-tory modes are present in this system; two intra-area modes(1.04 and 1.13 Hz), one in each area and one inter-area mode(0.52 Hz) where damping of the inter-area mode is of primeconcern. To improve the inter-area mode damping, a UPFCis installed on one of the tie-line connecting buses 7 and8 (sending-end of Area 1). The UPFC is equipped with theproposed supplementary ST controller (Fig. 9) to damp poweroscillations with real power deviation from bus 7 as the feed-back signal.

A. Time-Domain Analysis

The electromagnetic transient simulation software PSCAD/EMTDC is used for simulations. The synchronous generatorsof Fig. 9 are represented in the d q 0 reference frame bydetailed 7th order differential equations, transmission lines aremodeled as lumped impedances and system loads as constant

impedances. The dynamics of the synchronous machine ex-citers and governors are also included in the simulations withtheir details along with the system, UPFC and PI parametersgiven in Appendix.

Several fault types imposed at four different operating condi-tions as listed in Table I were studied. Note here that P78 refersto the inter-area line power flow with UPFC at the sending end.The UPFC PI parameters were optimized using multiple time-domain simulation based simplex algorithm [19] at operatingcondition II of Table I. A sampling time of 20 ms (50 Hz)chosen for the adaptive controller was found to be adequatefor damping low-frequency (0.10.8 Hz) oscillations. Fig. 10compares the inter-area power flow response between the PIcontrol and the PI control assisted with ST control for a 6cycle, 3-phase fault (Type A) applied at bus 8. The effective-ness of ST control is evident with the effective damping ofpower oscillations by the add-on controller as opposed to thestandalone-PI controllers. The corresponding series modulationindices (msePI and msePS) responsible for providingdamped power oscillations on tie-line 78 are shown in Fig. 11where the steady-state value (mse) is 0.475. It can be easilyseen that the proposed ST scheme greatly impacts power flowand hence power oscillation damping through larger variationsin mse. The ST control signal u(t) under the control limits of0.55 and the varying pole-shift factor are plotted in Fig. 12.

The behavior of at post fault occurrence is such that itprogresses to a lower value of = 0.6025 indicating maximumcontrol effort and slowly settles at = 1.0 once the subsequentoscillatory swings are adequately damped out. Also, duringfault the shunt VSC maintains the shunt bus voltage fluctuation|V7| within 0.05 p.u. along with regulating the dc link voltage(VDC(ref) = 42.4 kV for all test cases) that facilitates realpower exchange between the two VSCs as shown in Fig. 13(a)and (b). Further, the RLS identifier equipped with the trackingconstraint (t) of (10) offers a smooth parameter trackingability even during a large disturbance as shown in Fig. 14.

On the other hand, the improved power oscillation dampingof Fig. 10 due to the supplementary pole-shift algorithm can beverified from the corresponding pole-zero plot (in z-plane) ofFig. 15. The open-loop and closed loop poles are plotted as afunction of and their movement is captured from 2 s to 5 s.This snapshot is taken at t = 2.56 s when a pole-shift factorof = 0.485 shifted the open-loop poles indicated by a radially inwards to the closed-loop pole locations shown by ax. Thus the post-fault condition imposed a significant pole-shifting process. For the same power condition, a disturbanceof Type B is simulated with the system response (tie-line powerP78) and the corresponding pole-shift factor shown in Fig. 16.

The robustness of the proposed add-on ST controller indamping power system oscillations is validated (Figs. 1720)for different inter-area power flows and system configurationshighlighted in Table I. Fig. 17 depicts a comparison of powerflow responses with and without the ST controller in the sup-plementary loop when the transmission network is subjected toa Type C disturbance. From the responses, it can be ascertainedthat the oscillations are damped very quickly by the supplemen-tary controller proving its superiority over the conventional PI-UPFC control system.


Fig. 9. Two-area power system with UPFC at the sending-end of Area 1.

TABLE IINTER-AREA OSCILLATION DAMPING CASE STUDIES

Fig. 10. Tie-line power flow response to a Type A disturbance.

Fig. 11. Overall modulation of mse for Type A disturbance.

A similar observation holds true when the test system is sub-jected to a Type C disturbance at the center of tie-line 78 withthe damped inter-area power flow response shown in Fig. 18.However, the damping improvement by the ST controller isnot significant when compared with other operating scenarios

Fig. 12. ST controller output for a Type A disturbance.

Fig. 13. Shunt bus voltage and dc link voltage regulation by shunt VSC.

Fig. 14. Identified system parameters for a Type A disturbance.

since the PI controllers were optimally tuned for this particularoperating point. In another test, the tie-length is doubled from250 km to 400 km and the performance of the overall UPFCdamping model is tested for a disturbance of Type E and Type F.The corresponding damping responses are shown in Fig. 19.


Fig. 15. Dynamic pole movement as a function of for Type A disturbance.

Fig. 16. Identified system parameters for a Type A disturbance.

Fig. 17. System response to Type C disturbance.

Fig. 18. Damped inter-area power oscillations for a Type D disturbance.

Fig. 19. Damped inter-area power oscillations. (a) System response toType E; (b) system response to Type F.

Fig. 20. Power damping comparison for Type G disturbance.

Furthermore, the positive impact of the add-on controlscheme in damping inter-area oscillations was analyzed foroperating condition IV, Type G fault. The corresponding power-flow oscillation after the application of fault is plotted inFig. 20. It is clearly evident that the ST controller damps theoscillation within 10 s owing to its ability to adapt to differentoperating conditions without retuning.

B. Damping Improvement Assessment Using Prony Tool

The damping improvement shown in Table I is further in-vestigated using Prony analysis tool. Table II shows the studyresults for the base case (without any damping controller)and the four test cases with the standalone-PI controllers andsubsequently assisted with the proposed ST controller.

The Prony analysis tool is applied on the short circuitresponse of P78 to find the frequency and damping of thesystems inter-area mode. It is clear from Table II that theinter-area oscillatory mode frequency lies between 0.43 and0.58 Hz. Also, the corresponding damping ratio was found to be2.4%5.3% and being 5%, is significantly under-damped.The overall damping improvement seen with PI controllers istypically between 6% and 10% whereas with the add-on STcontroller is between 17% and 22%. However, the dampingimprovement by the ST controller observed for IInd operating


TABLE IIINTER-AREA OSCILLATION DAMPING CASE STUDIES

TABLE IIIMACHINE AND SYSTEM DATA

condition is the lowest since the PI controllers were optimallytuned under this test. Overall, the proposed scheme providedsignificantly improved damping performance.

V. CONCLUSION

The add-on ST controller provided significant damping im-provement for the inter-area mode as compared to the conven-tional PI-UPFC controllers. In addition, the proposed controlleris capable of adapting to different system configurations andcould be extended for any other type of FACTS device.

The use of constrained-RLS over the standard RLS identifierfor online parameter identification ensured favorable controlperformance even during large disturbances.

The proposed ST control methodology is currently being ap-plied for tuning of several FACTS (UPFC, STATCOM, SSSC)controllers in large power system configurations for achievingoptimum power oscillation damping without affecting indi-vidual device performance. Wide area control signals takinginto account issues such as communication delays [20] arebeing investigated. Experimental investigations using realtimehardware-in-the-loop simulations using a power system simu-lator would be reported in a future publication.

APPENDIX

See Tables III and IV.

TABLE IVUPFC DESIGN PARAMETERS

REFERENCES[1] N. G. Hingorani and L. Gyugyi, Understanding FACTS: Concepts and

Technology of Flexible AC Transmission Systems. New York, NY, USA:IEEE Press, 2000, pp. 315319.

[2] K. M. Son and R. H. Lasseter, A Newton-type current injection model ofUPFC for studying low-frequency oscillations, IEEE Trans. Power Del.,vol. 19, no. 2, pp. 694701, Apr. 2004.

[3] K. R. Padiyar and A. M. Kulkarni, Control design and simulation ofunified power flow controller, IEEE Trans. Power Del., vol. 13, no. 4,pp. 13481354, Oct. 1998.

[4] B. S. Chen and Y. Y. Hsu, A minimal harmonic controller for a STAT-COM, IEEE Trans. Ind. Electron., vol. 55, no. 2, pp. 655664, Feb. 2008.

[5] P. Mitra and G. K. Venayagamoorthy, An adaptive control strategy forDSTATCOM applications in an electric ship power system, IEEE Trans.Power Electron., vol. 25, no. 1, pp. 95104, Jan. 2010.

[6] S. Mohagheghi, Y. del Valle, G. K. Venayagamoorthy, and R. G. Harley,A proportional-integral type adaptive critic design-based neuro controllerfor a static compensator in a multimachine power system, IEEE Trans.Ind. Electron., vol. 54, no. 1, pp. 8696, Feb. 2007.

[7] A. G. Loukianov, J. M. Caedo, L. M. Fridman, and A. Soto-Cota, High-order block sliding-mode controller for a synchronous generator with anexciter system, IEEE Trans. Ind. Electron., vol. 58, no. 1, pp. 337347,Jan. 2011.

[8] W. Qiao and R. G. Harley, Indirect adaptive external neuro-control for aseries capacitive reactance compensator based on a voltage source PWMconverter in damping power oscillations, IEEE Trans. Ind. Electron.,vol. 54, no. 1, pp. 7785, Feb. 2007.

[9] H. Ishibuchi and T. Nakaskima, Improving the performance of fuzzyclassifier systems for pattern classification problems with continuousattributes, IEEE Trans. Ind. Electron., vol. 46, no. 6, pp. 10571068,Dec. 1999.

[10] X. Duan, H. Deng, and H. Li, A saturation-based tuning method for fuzzyPID controller, IEEE Trans. Ind. Electron., vol. 60, no. 11, pp. 51775185, Nov. 2013.

[11] C. Yan and G. K. Venayagamoorthy, AIS-Based coordinated and adaptivecontrol of generator excitation systems for an electric ship, IEEE Trans.Ind. Electron., vol. 59, no. 8, pp. 31023112, Aug. 2012.

[12] M. Wei and T. Liu, Design and implementation of an online tuningadaptive controller for synchronous reluctance motor drives, IEEE Trans.Ind. Electron., vol. 60, no. 9, pp. 36443657, Sep. 2013.

[13] G. P. Chen, O. P. Malik, G. S. Hope, Y. H. Qin, and G. Y. Xu, Anadaptive power system stabilizer based on the self-optimizing pole shiftcontrol strategy, IEEE Trans. Energy Convers., vol. 8, no. 4, pp. 639645,Dec. 1993.

[14] R. Gokaraju and O. P. Malik, Radial basis function identifier and pole-shifting controller for power system stabilizer application, IEEE Trans.Energy Convers., vol. 19, no. 4, pp. 663670, Dec. 2004.

[15] A. Domahidi, B. Chaudhuri, P. Korba, R. Majumder, and T. C. Green,Self-tuning flexible ac transmission system controllers for power


oscillation damping: A case study in real time, IET Gener. Transm.Distrib., vol. 3, no. 12, pp. 10791089, Dec. 2009.

[16] R. Sadikovic, P. Korba, and G. Andersson, Self-tuning controller fordamping power system oscillations with FACTS devices, in Proc. IEEEPower Eng. Soc. Gen. Meet., 2007, pp. 16.

[17] K. Sen and M. L. Sen, Introduction to FACTS Controller: Theory, Model-ing, and Applications. New York, NY, USA: IEEE Press, 2009, p. 298.

[18] K. J. strm and B. Wittenmark, Adaptive Control, 2nd ed. New York,NY, USA: Dover, 2008.

[19] A. Gole, S. Filizadeh, R. Menzies, and P. Wilson, Optimization enabledelectromagnetic transient simulation, IEEE Trans. Power Del., vol. 20,no. 1, pp. 512518, Jan. 2005.

[20] R. Yang, G. Liu, P. Shi, C. Thomas, and M. V. Basin, Predictive outputfeedback control for networked control systems, IEEE Trans. Ind. Elec-tron., vol. 61, no. 1, pp. 512520, Jan. 2014.

Urvi Malhotra (S09) received the B.E. degreein electronics engineering from the University ofMumbai, Mumbai, India, in 2008 and the M.Sc.degree in electrical engineering from the Universityof Saskatchewan, Saskatoon, Canada, in 2011.

Since May 2011, she has been working as a Net-work Management Engineer (EIT) at SaskatchewanPower Corporation in Regina, SK, Canada. Her re-search interests include flexible ac transmission sys-tems and adaptive control.

Ramakrishna (Rama) Gokaraju (S88M00) re-ceived the B.S. degree in electrical and electronicsengineering from the Regional Engineering College,Trichy, India, in 1992 and the M.Sc. and Ph.D.degrees in electrical and computer engineering fromthe University of Calgary, Calgary, Canada, in June1996 and May 2000, respectively.

During 19921994, he worked with Larsen &Toubro-ECC, India (Graduate Engineer), RegionalEngineering College, Rourkela, India (Research En-gineer), Indian Institute of Technology, Kanpur,

India (Project Associate). During the period 2000 to 2002, he worked withthe IBM Toronto Lab as a Staff Software Engineer. He joined the Universityof Saskatchewan in 2003 and is presently an Associate Professor. His currentresearch work is in the area of real-time power systems simulations, adaptivecontrol and smart grids.

/ColorImageDict > /JPEG2000ColorACSImageDict > /JPEG2000ColorImageDict > /AntiAliasGrayImages false /CropGrayImages true /GrayImageMinResolution 300 /GrayImageMinResolutionPolicy /OK /DownsampleGrayImages true /GrayImageDownsampleType /Bicubic /GrayImageResolution 300 /GrayImageDepth -1 /GrayImageMinDownsampleDepth 2 /GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true /GrayImageFilter /DCTEncode /AutoFilterGrayImages false /GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict > /GrayImageDict > /JPEG2000GrayACSImageDict > /JPEG2000GrayImageDict > /AntiAliasMonoImages false /CropMonoImages true /MonoImageMinResolution 1200 /MonoImageMinResolutionPolicy /OK /DownsampleMonoImages true /MonoImageDownsampleType /Bicubic /MonoImageResolution 600 /MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000 /EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode /MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None ] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false /PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true /PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ] /PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier () /PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped /False

/Description > /Namespace [ (Adobe) (Common) (1.0) ] /OtherNamespaces [ > /FormElements false /GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks false /IncludeInteractive false /IncludeLayers false /IncludeProfiles false /MultimediaHandling /UseObjectSettings /Namespace [ (Adobe) (CreativeSuite) (2.0) ] /PDFXOutputIntentProfileSelector /DocumentCMYK /PreserveEditing true /UntaggedCMYKHandling /LeaveUntagged /UntaggedRGBHandling /UseDocumentProfile /UseDocumentBleed false >> ]>> setdistillerparams> setpagedevice

an add-on self-tuning control system

Documents