608 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 2, APRIL 2014

Online Optimal Control of Reactive Sources inWind Power Plants

HoanV. Pham, Student Member, IEEE, José L. Rueda, SeniorMember, IEEE, and István Erlich, Senior Member, IEEE

Abstract—This paper suggests a novel approach to the problemof online optimal control of reactive sources in wind power plants(WPPs). Based on measurements tracking the actual conditions ofeach wind generator and the settings of transformers and compen-sation equipment within the WPP facility, the key idea is to in-corporate reactive power optimization into a global WPP’s controlloop so that it can be used online to determine the optimal distribu-tion of reactive power, which is needed to meet grid code require-ments, among the available Var sources. The optimization problemis handled using an enhanced version of the mean-variance map-ping optimization algorithm. A test case, based on a representativeoffshore WPP, is presented to demonstrate the effectiveness of theproposed control strategy.

Index Terms—Mean-variance mapping optimization, onlineadaptive control, reactive power management, wind power plant.

I. INTRODUCTION

W IND power has firmly positioned itself as one of themost important renewable energy sources over the past

two decades. As of this writing, the share of wind power in rela-tion to the overall installed capacity has increased significantlydue to the policy incentives adopted by several countries to sup-port renewable energy development, and this trend is in all like-lihood set to continue. In some countries, the share of wind inrelation to the overall installed capacity is already approachingthe 50% mark [1].With the increasing integration of wind power plants (WPPs),

grid utilities require extended reactive power supply capability,not only during voltage dips, but also in steady state operation.According to the grid codes [2], the steady state reactive powerrequirements are defined alternatively in terms of the powerfactor, the amount of reactive power supplied or the voltage atthe point of common coupling (PCC). Typically, the availablereactive power sources within the WPP are wind genera-tors, conventional compensation elements, or some version ofFACTS devices. In addition, the control systems, which modernwind generators employ, are characterized by fast responsetime and thus open up additional unconventional options toprovide extra reactive power support [3]. Hence, coordinatedreactive power control strategies have been suggested based

Manuscript received April 30, 2012; revised December 17, 2012 and June28, 2013; accepted July 02, 2013. Date of publication August 02, 2013; dateof current version March 18, 2014. This work was supported by the UniversityDuisburg-Essen and the German Academic Exchange Service.The authors are with the Department of Electrical Power Systems, Univer-

sity Duisburg-Essen, Duisburg, 47057, Germany (e-mail: vanhoan.pham@uni-due.de; jose.rueda@uni-due.de; istvan.erlich@uni-due.de).Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/TSTE.2013.2272586

on the operational chart of variable speed wind generators [1],[4]. The optimal utilization of reactive sources in WPPs hasalso been addressed in recent literature, in which the optimalreactive power dispatch problem (ORPDP) was formulated asminimum power loss or minimum voltage deviation as targets[5], [6]. In [3] and [7], a predictive control approach, where theORPDP also accounts for minimization of the cost of operationassociated with the number of on-load tap changes (OLTC), issuggested.Mathematically, the ORPDP falls into the category of

mixed-integer nonlinear optimization problems. Classicalgradient-based optimization algorithms may fail to solve suchproblems due to the difficulties in handling nonconvex anddiscontinuous problems as well as discrete variables. More-over, the accuracy of the solution is quite sensitive to the initialpoints [5]. In recent years, an ever-increasing research efforthas been dedicated to the solution of the ORPDP based on theapplication of a variety of heuristic optimization algorithmssuch as genetic algorithm [6], particle swarm optimization [3],differential evolution [8], evolutionary programming [9], antcolony optimization [10], and bacterial foraging optimization[11]. These techniques have indeed demonstrated effectivenessin overcoming the disadvantages of classical algorithms. Par-ticularly, particle swarm optimization and differential evolutionhave received great attention from researchers due to theirsearching power. Nevertheless, some pitfalls for the use ofthese techniques should be considered in order to avoid pre-mature convergence and local stagnation since their searchingcapability is highly dependent on appropriate parameter set-tings as evidenced in several applications [10]–[13].The main objective of this paper is to introduce a mean-vari-

ance mapping optimization (MVMO)-based controller for on-line optimal control of reactive power sources in WPPs, whichcan be implemented as an extension to the existing WPP con-trol structures. Common controllers include a proportional inte-gral (PI) control unit for offsetting the difference between gridcode reactive power requirement and the reactive power cur-rently supplied at the point of common coupling (PCC). Theoutput of the PI block constitutes the reactive power that needsto be supplied on top of the current setting which should be dis-tributed between the available Var sources in an optimal manner.Based onmeasurements and tracking the current settings of eachwind turbine, transformer, and compensation elements, the op-timization, for any given operating point, will result in optimaldistribution of reactive power between each of the sources cur-rently in service leading to minimum losses while at the sametime satisfying mandatory reactive power supply.

1949-3029 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

PHAM et al.: ONLINE OPTIMAL CONTROL OF REACTIVE SOURCES IN WIND POWER PLANTS 609

Fig. 1. Implementation procedure for slow reactive power control at WPPs.

MVMO is a novel heuristic optimization algorithm whichwas originally proposed in [14]. Its most salient feature is thatit uses a special mapping function applied for mutating the off-spring on the basis of the mean and variance of the -best pop-ulation attained so far. Motivated by the success gained in pre-vious applications to different power system optimization prob-lems [5], [7], the second objective of the paper is to introducean enhanced variant of MVMO that exploits the asymmetricalproperties of the algorithm’s mapping function so that it canfind the optimum more quickly with minimum risk of prema-ture convergence.The remainder of this paper is organized as follows. Section II

gives an overview of the proposed control strategy, discussingthe main implementation issues. In Section III, a test case isdeveloped and evaluated. Finally, conclusions and outlook forfuture work are presented in Section IV.

II. PROPOSED CONTROL STRATEGY

The implementation aspects of the proposed strategy for on-line reactive power control in WPPs are summarized in Fig. 1.Considering the availability of a data acquisition system to pro-vide measurements related to actual status of all wind genera-tors (WGs), transformers, and compensation devices within theWPP, the adopted control approach continuously fulfills the gridcode requirement at PCC (e.g., ) by means of a slow re-sponse controller which allows the optimummanagement of theavailable WPP’s Var sources during normal (i.e., steady-state orquasi-steady-state) conditions. Although such a control is cou-pled to a local fast control scheme at every Var source, it hasa slow response to small operational changes (i.e., time frameof 10 s to a few minutes) and does not provide any fast re-action during large disturbances in order to avoid undesirableinteractions.It is emphasized that the proposed controller is exclusively

conceived (from WPP operation point of view) for continuousfulfillment (in an optimal manner) of grid code requirement atPCC and not for system-wide reactive power control purposes.So analysis with large-scale system modeling beyond the PCC

Fig. 2. WPP control schema including online optimization.

is not needed to illustrate the implementation and the test resultsprovided in this paper.

A. Wind Power Plant Benchmark Layout

The implementation of the proposed control strategy is illus-trated henceforward by considering the WPP layout shown inFig. 2. It should, however, be pointed out that due to its ver-satility and simplicity, this control strategy can be straightfor-wardly adapted to other configurations. The layout resemblesthe commonly used topology for offshore WPP, which is nor-mally connected to the main grid using long cables with step-uptransformers at both ends. Due to large charging currents of ca-bles, line reactors are connected at one or both ends of the ca-bles. In the example presented in this paper, a switched reactoris connected to the grid side bus bar. Although not consideredin this paper, FACTS devices are sometimes included to pro-vide fast and continuous Var control. Also for very long cablesit is common to connect permanently connected shunt reactorsdirectly to both ends of the cable which are switched always to-gether with the cable.

B. WPP Var Control Considerations

The proposed controller, as depicted in Fig. 2, is intended forreactive power control by considering load flow and topologychanges. The control task is to be performed only during normal(i.e., steady-state or quasi-steady-state) conditions. Thus, it re-quires a slow response to adapt the overall WPP response (i.e.,by optimally adjusting the long-term reactive power referenceinputs for all available controllable Var sources) to changingsteady-state requirements. A time frame of 10 s to a few min-utes is commonly sufficient [15]. Hence, the controller does notprovide fast response during grid faults in order to avoid trig-gering of oscillations and unnecessary control actions. Such afast control, which is beyond the scope of this paper, can onlybe implemented at the level of individual wind generators andshould react to fast voltage changes within 20–30 ms withoutaltering the longer-term settings of the slow controller. It is alsoworth mentioning that, in case of a voltage drop, there shouldbe a predominance of fast control on an individual generator

610 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 2, APRIL 2014

Fig. 3. Example for grid code requirement at PCC: Minimum requirementsfor reactive power compensation of a generation plant with unrestricted activepower output.

whereas slow control at the overall plant level should be priori-tized under steady-state conditions.

C. Implementation of the Controller

At the WPP level, a certain degree of coordination is neces-sary between different devices within the WPP facility to pro-vide proper control reactions with respect to the PCC. Thus,the task of the WPP controller is to control the reference in-puts (i.e., set points) of the individual wind generators and pos-sible additional active or passive reactive power compensationequipment, so that control requirements at the PCC are met. Thetransmission system operators (TSOs) commonly define suchrequirements in their grid codes for the respective voltage levels.Grid codes may depend, to some extent, on the specific condi-tions of the TSO and, therefore, they can differ from companyto company even within the same country [16]. A typical re-quirement during normal operating conditions, which was builtbased on data in [17], is illustrated in Fig. 3.The WPP must be able to operate at any point within the area

in the diagram. Besides, the requirements can alternatively beformulated in terms of power factor or reactive power referenceat the PCC, with the latter being used in this paper. Thus, thesubsequent analysis considers that the WPP should always at-tempt to supply the required reactive power at the PCC .From Fig. 2, note that the proposed WPP controller includes

a PI control block, which is used to continuously adapt the gen-erated reactive power to the reference at the PCC. It is worthrecalling that the controller is characterized by a slow reaction(e.g., within a time frame of around 10 s to a few minutes) toadapt the overallWPP response to changing steady-state require-ments (cf. Section II-B). Therefore, for design purposes, the pro-portional gain of the PI block can be considered relatively smalland the integral time constant can be set within 10–20 s. Theoutput signal of the PI unit constitutes the additional reactivepower needed to meet the requirement at the PCC, which is dis-tributed to the individualWPPVar sources based on an optimiza-tionmodule that assigns the optimal set point to each source withthe ultimate goal of operating the WPP with minimum losseswhile fulfilling the reactive power requirements at the PCC.

D. Optimization Module

The optimization module is used for efficient operation ofthe WPP as per grid reactive power requirements. It optimizespower flow in such a way that the total losses of the wind en-ergy system are minimized. The resulting set points are pro-vided to the Var sources. Some sources, which can play a role inminimizing system losses, if operated optimally, can be variedcontinuously (e.g., wind generator reactive power output) whileothers allow variation only stepwise (e.g., transformer tap posi-tion). Thus the task which has to be solved represents a mixed-integer optimization problem. The set points of wind generatorsare normally controlled by the WPP controller by allocating therequired reactive power (output of the WPP controller) equallyto each generator. In contrast, the optimization suggested in thispaper can modify the reference settings by deviating from theuniform Var distribution when necessary. The resulting Var ref-erences are sent via communication link to the wind turbines.In current WPP operating practice, the transformers tap po-

sitions are controlled by separate and independent voltage con-trollers. However, according to the approach here, the controlof tap positions is part of the optimization. Assuming that theactual operational status of each wind generator (i.e., ON/OFFstatus, current output power) as well as the current tap settings oftransformer and compensation elements are available, the opti-mization problem can mathematically be formulated as follows:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

The bus voltage magnitude vector and its corresponding limitsare denoted by , , and , whereas and are the cur-rent and apparent power flow vectors in the branches with limitsdefined by and , respectively. and are the nodalactive and reactive power generation vectors whereas andare the nodal active and reactive power demand vectors.and stand for nodal active and reactive power injectionvectors.Recalling Fig. 2, the objective function, as defined in (1), en-

tails indirectly the minimization of total losses of the wind en-ergy system while satisfying the reactive power requirement atPCC, since the terms and in the equation equal the differ-ence between the injection from the WPP and the fictitious con-sumption associated to the load L1, which represents a dummyload whose real part corresponds with the nominal WPP activepower and the imaginary component with , respectively.Constraints (1) and (2) account for nodal balance, whereas the

PHAM et al.: ONLINE OPTIMAL CONTROL OF REACTIVE SOURCES IN WIND POWER PLANTS 611

constraint set (7)–(9) is composed of bounds on the decisionvariables, which include:1) The vectors of Var limits ( and ) for the windgenerators, which can be obtained from the active/reactivepower capability curves supplied by the manufacturers.

2) The vectors of transformer discrete tap change limits( and ).

3) ON/OFF switching status of the reactor .Again, it is worth emphasizing that the significance of the

minimum attainable value of the objective function, as definedin (1), is twofold:i) operation of the WPP with minimum losses;ii) meeting the required reactive power at the PCC.

Simultaneously the internal voltage profiles (e.g., at wind gen-erator terminals) will be kept within acceptable ranges.The reactive power contribution of each wind generator is

determined by

(10)

where is the number of wind turbines, is the distributionfactor assigned to the th wind generator, and is theoutput of the WPP controller. The control strategy, as suggestedin this paper, can be also interpreted as an implicit optimal ad-justment of the distribution factors (i.e., allocation) to changingoperating conditions. Nevertheless, the optimization modulecan also operate in an alternative mode which optimally coor-dinates the settings of transformer and compensation elementsonly while allowing uniform reactive power distribution amongwind generators. This may be favorable if the different uti-lization of wind turbines for Var generation will not result insignificant reduction of losses.During continuous operating regime of the WPP, the afore-

said measurements are supplied to the optimization module,which delivers the optimal decisions, obtained from the solu-tion to the optimization problem as described above, as controlsignals to the various Var sources at each time step for the givenoperating point. This kind of optimization can be repeated incertain time intervals, e.g., every 5–15 min. The determinationof the solution to the optimization problem is handled throughthe optimization module MVMO.

E. MVMO Algorithm

MVMO is a population-based stochastic optimization algo-rithm that has been recently developed and shown to have aremarkably better performance, compared to other basic andenhanced evolutionary algorithms, especially in terms of con-vergence behavior [18]. The basic theoretical background ofMVMO has been published in [14].MVMO operates on a single solution rather than a set of solu-

tions like in many evolutionary algorithms. In the initial stage,the real min/max boundaries of variables have to be normalizedto 0 and 1, since the internal search range of all variables isrestricted to [0, 1]. Hence, it is avoided that any componentof the solution vector violates the corresponding boundariesthroughout the optimum searching process. Nevertheless, thefunction evaluation is carried out always in the original scales.

Fig. 4. Exemplary shape of the MVMO mapping function.

The key feature of MVMO is a special mapping function de-scribed by mean and shape variables, which are derived fromthe n best solutions saved in an archive. In essence, the mappingfunction transforms a variable varied randomly with unitydistribution to another variable which is concentrated aroundthe mean value. The mapping is illustratively shown in Fig. 4.The distribution of the new variable does not correspond

with any of the well-known distribution functions even thoughthere is some kind of resemblance to the Gauss function. Thetransformation is as follows:

(11)

where the -function is defined as

(12)

where , , and are the outputs of the -function, basedon different inputs given by

(13)

Note that the output of (12) is always within the bounds [0, 1]for every generated .Recalling (12), it can be surmised that the shape of the -func-

tion is determined by the mean and the shape variables and. The effect of these parameters on the form of the function is

illustrated in Fig. 5, which leads to the statement that the searchspace can be fully explored by taking advantage of the asym-metry of the mapping function’s shape. Hence, the distinctiveproperty of MVMO is the ability to perform global search fo-cusing around the best solutions attained so far. This is shownfor two variables in Fig. 6. As can be seen, the search is focusedaround the mean values which are and inthis example. However, there are some samples also outside themean areas, i.e., the algorithm performs global search but theemphasis is around the mean values.Mean and shape variables are calculated from the archive,

where the best populations are stored, as follows:

(14)

(15)

612 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 2, APRIL 2014

Fig. 5. Effect of parameters on the mapping function.

with the variance

(16)

The variance is calculated only for different variables in thearchive. The factor can be used to change the shape of thefunction, e.g., when the accuracy needs to be improved (increase

) or more global search is required (decrease ).In contrast to the previous publication [5], [7], [14], [17], theshapes and of variable are not calculated directly from(16) but by using the following procedure:

(17)

The initial values of are set for all variables at the beginningof the optimization. Experience so far shows that values around1–5 are suitable to guarantee good initial performance. At everyiteration, each is scaled up or down with the factor . If

, the current is divided by which is always largerthan 1.0 and thus leads to reduced value of . In case ,

Fig. 6. Search space depending on the shape factor; number of variables, , shape for both variables.

Fig. 7. Effect of the algorithm shown in (17) on the mapping function.

will be multiplied by resulting in increased . In this way,will always oscillate around the current shape factor . By

using the factor instead of , which is a function of variance,a smoothing effect is achieved, as illustrated in Fig. 7.Furthermore, the asymmetrical properties of the mapping

functions are exploited to enhance the algorithm searchingperformance (i.e., robustness) and the zero variance handling.Zero variance can occur when all variables of in the archiveare identical. In this case, the previous nonzero value is usedfurther. The mean and variance are not calculated before the

PHAM et al.: ONLINE OPTIMAL CONTROL OF REACTIVE SOURCES IN WIND POWER PLANTS 613

Fig. 8. Flowchart of the MVMO algorithm.

archive is filled up completely. In this stage, searching is per-formed with which corresponds with a straightline between zero and one as the mapping function. Usuallyan archive size of 2–5 is sufficient. A larger archive size willresult in a rather conservative searching with orientation on thesaved best populations.The flowchart of the MVMO algorithm is sketched in Fig. 8.

As a parent of the new population, the best population savedin the archive (first position) is used. Then a few variables areselected and projected on the mapping function. Alternative se-lection methods are described in [5] and [14].

III. TEST RESULTS

The proposed online optimal control strategy is tested ona WPP whose layout is as shown in Fig. 2. It consists of

Fig. 9. Wind power variation.

Fig. 10. Reactive power requirement at the PCC.

18 generators each rated at 5 MW and is connected to the220-kV-power grid through two 100-MVA transformers andone 110-kV submarine cable km of about30 km in length. Both transformers are equipped with OLTC.The PCC is considered at the 220-kV-side of the transformerT1 ( % kV with 33 taps). The transformer T2 israted at % kV with 13 taps. The distances fromeach generator to the main collector of the WPP are uneven.The voltage at the wind turbine terminals is to be maintainedbetween 0.92 and 0.97 kV (i.e., 920–970 V) where the nom-inal value is 0.95 kV (i.e., 950 V). For all other nodes 5%range around the nominal voltage is prescribed. The excessivecharging current of the submarine cable is compensated byconnecting a shunt reactor on one side of the cable. The reactoris rated at 500 and is adjusted with ON/OFF control.The exemplary wind profile shown in Fig. 9 is used for

the simulation. To highlight the relevance of the online re-active power control problem, the grid code requirementscorresponding to the actual operating condition are defined asstepwise changes of the reactive power reference at PCC asshown in Fig. 10.Three cases are considered for comparison purposes:Case 1) Optimal adjustment of all Var sources to meet the

actual reactive power requirement at PCC includingboth OLTC settings, shunt reactor ON/OFF com-mands and individual reactive power of each windturbine (i.e., different distribution factors for eachwind turbines). The controller calculates the optimalsettings every 15 min.

614 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 2, APRIL 2014

Fig. 11. Reactive power output of wind generator 1.

Fig. 12. Reactive power of the reactor.

Fig. 13. OLTC tap positions—T1.

Case 2) Optimal adjustment of the shunt reactor, both OLTCand a uniform wind turbine reactive power to begenerated by all wind generators.

Case 3) The optimization is not implemented. The PCC con-troller output distributed equally to all operatingwind turbines. The OLTCs are controlling the busbar voltage at the 110- and 33-kV levels, respec-tively. This case represents the common operatingmode of wind farms currently implemented. The re-actor is switched at the predefined time taken fromCases 1 and 2.

Figs. 11–14 describe the dispatch curves of the WPP Varsources under the above-mentioned different cases. The ad-vantages of using the proposed controller are demonstrated inFigs. 15–18.

Fig. 14. OLTC tap positions—T2.

Fig. 15. Total active power losses of wind farm.

Fig. 16. Active power at PCC.

Fig. 17. Difference between demanded and supplied reactive power at PCC.

PHAM et al.: ONLINE OPTIMAL CONTROL OF REACTIVE SOURCES IN WIND POWER PLANTS 615

Fig. 18. Terminal voltage of wind generator 1.

From Fig. 11, it can be seen that the generator reactive poweroutput curves corresponding to Cases 1 and 2 are very close ateach time interval. This is reasonable since the adopted WPPlayout did not involve large asymmetrical distances betweengenerators, which was reflected in comparable distribution fac-tors in both cases. The wind farm is almost capable to meet theVar requirements except in the extreme situations shown at bothends of the curves in Fig. 17. By contrast, in Case 3, the op-timization tool is not used. The required reactive power fromthe PI controller is distributed equally to the wind turbines andthe OLTCs are controlling the respective bus bar voltages. Dueto the PI characteristic of the WPP controller, the Var require-ments are met in similar manner as in Cases 1 and 2 (Fig. 17)as long as the wind turbine’s maximum reactive power capa-bility is not reached. However, the voltage level controlled bythe OLTCs is not optimal in Case 3. As can be seen from Fig. 18,the terminal voltage of the wind turbines exceeds the prescribedrange of 0.92–0.97 kV considerably. Lower voltages will resultin higher losses, which is the case in the second half of the day inthe simulated scenario. This will also lead to smaller total powerbeing supplied at the PCC, as shown in Fig. 16. It seems fromFig. 15 that for positive reactive power demand (left-hand side),the losses are smaller in Case 3. However, this is due to the highvoltage value, which is higher than the allowed 0.97 kV. Smalltemporary violations of the voltage limits can also happen inCases 1 and 2 owing to the fact that the optimization is carriedout every 15 min and in the meantime the wind power fed-inmay change.From Fig. 12, note that the reactor is switched ON at the same

point in time for all cases due to the actual compensation needfrom this point. The step variation of its reactive power is attrib-utable to OLTC changes. Figs. 13 and 14 show the total numberof tap movements in the daily operation of the WPP for T1 andT2, respectively. Remarkably, more movements would be nec-essary in Cases 1 and 2 to meet grid code requirements corre-sponding to the actual measurements from the WPP.

IV. CONCLUSION

In this paper, a novel control strategy has been suggested andsuccessfully applied for online optimal control of wind powerplant reactive sources. It can be implemented as an extensionto the existing WPP control structures. Based on measurements

from the plant indicating the actual settings of its Var sourcesand the actual active power generated by the wind turbines, theoptimal utilization of Var sources and OLTC positions are deter-mined. The results of optimal wind turbine Var settings are in-corporated into the existing WPP controller by distribution fac-tors. The suggested approach guarantees not only optimal WPPoperation but is also robust. As backup option uniform distribu-tion factors can be used that corresponds with the current statusof implementations without the extension by the optimizer. De-pending on theWPP design, uniform distribution of Var genera-tion to the wind turbines may be sufficient. Results demonstratethat the incorporation of the controller at overall plant level en-tails optimal operation with smaller energy losses than the directlocal control of Var sources as well as continuous fulfillment ofoperational requirements. However, the simulation results haveshown that the required number of transformer tap changes willalso increase when the suggested optimal controller approach isused.Future research work is being directed towards inclusion of

predictive control issues where actions are taken on the basis ofa wind speed forecast in order to avoid unnecessary short-termOLTC tap changes.

REFERENCES[1] S. Engelhardt, I. Erlich, C. Feltes, J. Kretschmann, and F. Shewarega,

“Reactive power capability of wind turbines based on doubly fed in-duction generators,” IEEE Trans. Energy Convers., vol. 26, no. 1, pp.364–372, Mar. 2011.

[2] I. Erlich, W. Winter, and A. Dittrich, “Advanced grid requirementsfor the integration of wind turbines into the German transmissionsystem,” in Proc. 2006 IEEE Power Engineering Society GeneralMeeting, Montreal, Canada, Jun. 2006, pp. 1–7.

[3] V. S. Pappala, W. Nakawiro, and I. Erlich, “Predictive optimal controlof wind farm reactive sources,” in Proc. 2010 IEEE PES Transmissionand Distribution Conf. Exposition, Apr. 2010, pp. 1–7.

[4] S. Foster, L. Xu, and B. Fox, “Coordinated reactive power control forfacilitating fault ride through of doubly fed induction generator-andfixed speed induction generator-based wind farms,” IET Renew. PowerGenerat., vol. 4, no. 2, pp. 128–138, Mar. 2010.

[5] W. Nakawiro, I. Erlich, and J. L. Rueda, “A novel optimization algo-rithm for optimal reactive power dispatch: A comparative study,” inProc. 4th Int. Conf. Electric Utility Deregulation and Restructuringand Power Technologies, Weihai, Shandong, China, Jul. 2011.

[6] I. J. Fang, G. Li, X. Liang, and M. Zhou;, “An optimal control strategyfor reactive power in wind farms consisting of VSCF DFIG wind tur-bine generator systems,” in Proc. 2011 4th Int. Conf. Electric UtilityDeregulation and Restructuring and Power Technologies, Jul. 2011,pp. 1709–1715.

[7] I. Erlich,W. Nakawiro, andM.Martinez, “Optimal dispatch of reactivesources in wind farms,” in Proc. 2011 IEEE Power and Energy SocietyGeneral Meeting, Jul. 2011, pp. 1–7.

[8] M. Varadarajan and K. S. Swarup, “Network loss minimization withvoltage security using differential evolution,” Electric Power Syst.Res., vol. 78, no. 5, pp. 815–823, May 2008.

[9] Q. H.Wu and J. T. Ma, “Power system optimal reactive power dispatchusing evolutionary programming,” IEEE Trans. Power Syst., vol. 10,no. 3, pp. 1243–1249, Aug. 1995.

[10] A. Abbasy and S. H. Hosseini, “Ant colony optimization-based ap-proach to optimal reactive power dispatch: A comparison of variousant systems,” in Proc. 2007 IEEE Power Engineering Society Conf.and Exposition in Africa, July 2007, pp. 1–8.

[11] M. Tripathy and S. Mishra, “Bacteria foraging-based solution to op-timize both real power loss and voltage stability limit,” IEEE Trans.Power Syst., vol. 22, no. 1, pp. 240–248, Feb. 2007.

[12] J. Lampinen and I. Zelinka, “On stagnation of the differential evolutionalgorithms,” in Proc. 6th Int. Conf. Soft Computing., 2000, pp. 76–83.

[13] W. Nakawiro and I. Erlich, “A new adaptive differential evolution al-gorithm for voltage stability constrained optimal power flow,” in Proc.2011 Power Systems Computation Conf., Aug. 2011, pp. 1–7.

616 IEEE TRANSACTIONS ON SUSTAINABLE ENERGY, VOL. 5, NO. 2, APRIL 2014

[14] I. Erlich, G. K. Venayagamoorthy, and W. Nakawiro, “A mean-vari-ance optimization algorithm,” in Proc. 2010 IEEE World Congress onComputational Intelligence, Barcelona, Spain.

[15] E. ON Netz GmbH, Grid Code High and Extra High Voltage, Apr.2006.

[16] V. S. Pappala, S. N. Singh, M. Wilch, and I. Erlich, “Reactive powermanagement in offshore wind farms by adaptive PSO,” in Proc. 2007Int. Conf. Intelligent Systems Applications to Power Systems, Nov.2007, pp. 1–8.

[17] E.ONNetz GmbH, Requirements for Offshore Grid Connections in theE.O.N. Netz Network, Bayreuth, Germany, Apr. 2008 [Online]. Avail-able: http://www.eon-netz.com

[18] I. Erlich, F. Shewarega, C. Feltes, F. Koch, and J. Fortmann, “Determi-nation of dynamic wind farm equivalents using heuristic optimization,”inProc. 2nd 2012 IEEE Power &Energy Society General Meeting, SanDiego, Paper Accepted for Publication the, USA, accepted for publi-cation.

Hoan V. Pham (S’12) was born in 1984. Hereceived the B.E. degree in electrical engineeringfrom College of Technology, Can Tho University,Vietnam, in 2007, and the M.Sc. degree in electricalengineering from College of Engineering, JejuNational University, South Korea, in 2010. Heis currently working toward the Ph.D. degree atthe University of Duisburg-Essen, Germany withfinancial support from DAAD.His research interests include voltage stability in

power system, computational intelligence with em-phasis on heuristic optimization and its applications in renewable power sys-tems.

José L. Rueda (SM’12) was born in 1980. He re-ceived the Electrical Engineer diploma from the Es-cuela Politécnica Nacional, Quito, Ecuador, in 2004,and the Ph.D. degree in electrical engineering fromthe Universidad Nacional de San Juan, San Juan, Ar-gentina, in 2009.From September 2003 to February 2005, he

worked in Ecuador, in the fields of industrial con-trol systems and electrical distribution networksoperation and planning. Currently, he is a researchassociate at the Institute of Electrical Power Systems,

University of Duisburg-Essen. His current research interests include powersystem stability and control, system identification, power system planning,probabilistic and artificial intelligence methods, smart grids, heuristic opti-mization, FACTS devices, and wind power.

István Erlich (SM’08) was born in 1953. He re-ceived the Dipl.-Ing. degree in electrical engineeringand the Ph.D. degree from the University of Dresden,Dresden, Germany, in 1976 and 1983, respectively.From 1979 to 1991, he was with the Department

of Electrical Power Systems of the University ofDresden. In the period of 1991 to 1998, he workedwith the consulting company EAB, Berlin, Germany,and the Fraunhofer Institute IITB Dresden. Duringthis time, he also had a teaching assignment at theUniversity of Dresden. Since 1998, he has been a

Professor and head of the Institute of Electrical Power Systems at the Universityof Duisburg-Essen, Duisburg. Germany. His major scientific interest comprisespower system stability and control, modeling, and simulation of power systemdynamics, including intelligent system applications.Dr. Erlich is a member of VDE and the chairman of the International Feder-

ation of Automatic Control (IFAC) Technical Committee on Power Plants andPower Systems.