fully distributed coordination of multiple dfigs in a microgrid for load sharing

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON SMART GRID 1

Fully Distributed Coordination of Multiple DFIGs ina Microgrid for Load Sharing

Wei Zhang, Student Member, IEEE, Yinliang Xu, Student Member, IEEE, Wenxin Liu, Member, IEEE,Frank Ferrese, Member, IEEE, and Liming Liu, Member, IEEE

Abstract—When wind power penetration is high, the availablegeneration may be more than needed, especially for wind-pow-ered microgrids working autonomously. Because the maximumpeak power tracking algorithm may result in a supply-demandimbalance, an alternative algorithm is needed for load sharing.In this paper, a fully distributed control scheme is presented tocoordinate the operations of multiple doubly-fed induction gener-ators (DFIGs) in a microgrid. According to the proposed controlstrategy, each bus in a microgrid has an associated bus agent thatmay have two function modules. The global information discoverymodule discovers the total available wind generation and total de-mand. The load sharing control module calculates the generationreference of a DFIG. The consensus-based algorithm can guar-antee convergence for microgrids of arbitrary topologies undervarious operating conditions. By controlling the utilization levelsof DFIGs to a common value, the supply-demand balance can bemaintained. In addition, the detrimental impact of inaccurate andoutdated predictions of maximum wind power can be alleviated.The generated control references are tracked by coordinatingconverter controls and pitch angle control. Simulation results witha 5-DFIG microgrid demonstrate the effectiveness of the proposedcontrol scheme.

Index Terms—Cooperative systems, distributed control, micro-grid, wind power generation.

I. INTRODUCTION

I N RECENT years, environmental and economic concernshave significantly increased the demand for clean and

efficient power generation. Due to advances in relevant tech-nologies and commercial viability, wind power is gaining morepopularity compared with other types of renewable energyresources. The U.S. Department of Energy expects to derive20% of its national electrical supply from wind energy by2030 [1]. However, the intermittency of wind power poses newchallenges for power system operation and control, especiallyduring times of high penetration.One challenge with the control of a wind-powered micro-

grid is setting the generation references of the wind turbinegenerators under dynamic wind and loading conditions. Thisis achieved by “load sharing” control, whose objective is torealize supply-demand balance of the microgrid. This processis called “load sharing.” The popular maximum peak power

Manuscript received December 12, 2011; revised May 31, 2012; acceptedDecember 05, 2012. This work was supported by the U.S. National ScienceFoundation under Grant ECCS #1125776. Paper no. TSG-00684-2011.W. Zhang, Y. Xu and W. Liu are with the Klipsch School of Electrical and

Computer Engineering, New Mexico State University, Las Cruces, NM 88003USA (e-mail: [email protected]; [email protected]; [email protected]).F. Ferrese is with the Naval Surface Warfare Center-Carderock Division,

Philadelphia, PA 19112 USA (e-mail: [email protected]).L. Liu is with the Center of Advanced Power Systems, Florida State Univer-

sity, Tallahassee, FL 32310 USA (e-mail: [email protected]).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/TSG.2012.2234149

tracking (MPPT) algorithm may cause supply-demand imbal-ance when the maximum wind power is more than required,such as in an autonomous microgrid. To overcome this problem,some DFIGs can be controlled to work at MPPT mode and therest can evenly share the remaining demand. However, this typeof solution is susceptible to inaccuracy in available wind powerprediction. Another alternative way is utilizing energy storagedevices, such as pumped water, compressed air, or super-capac-itors, which can store the excessive power generated accordingto MPPT [2]–[4]. However, limited by current energy storagetechniques, it is very expensive to install high-capacity energystorage devices. Even if energy storage is available, its initialinstallation is usually limited and will run out of energy storagecapability easily. Thus, it is necessary to investigate the controlissue when there is insufficient or no energy storage devices inan autonomous microgrid.In [5], the authors presented a two-level control scheme

for automatic generation control of a wind farm. This controlscheme consists of a supervisory control level and a machinecontrol level. The supervisory control level decides the powersettings of different DFIGs, while the machine control levelensures that the set points are realized. Optimization algorithmsalso can be implemented to optimize the generation set points,such as in [6]. However, these centralized control schemesrequire complicated communication networks to collect infor-mation globally [7], as well as a powerful central controller toprocess the huge amount of data. Thus, these centralized solu-tions are costly to implement and susceptible to single-pointfailures [8]. Due to the intermittency of wind power, morefrequent generation updates are required. The above solutionsmay not be able to respond in a timely fashion when operatingconditions change rapidly and unexpectedly.Once the required active power generations have been de-

cided, the DFIGs must be controlled to realize the control ob-jectives. Because the required generations may be less than themaximum, the so-called deloading control strategy, as discussedin [6], [9], is needed. The deloading strategies presented in [6],[9] can only adjust generation accurately in a limited range. Allof the methods discussed in [10] have some limitations, such asresponse speed, range of operation, and maintenance costs. Forefficient deloading, the two deloading strategies of convertercontrol and pitch angle control must be well coordinated. In ad-dition to active power generation, reactive power generation isalso needed for voltage stabilization and grid support. Based onthe above analysis, deloading control of an individual DFIG isanother challenging component of wind power generation.To address the problems with centralized solutions, a

multi-agent system (MAS)-based distributed solution is pre-sented in this paper. As one of the most popular distributedcontrol solutions, MAS is flexible, reliable, less expensive toimplement, and has the ability to survive single-point failures.

1949-3053/$31.00 © 2013 IEEE


2 IEEE TRANSACTIONS ON SMART GRID

Over the past several years, MAS has been applied widely toload restoration and power system reconfiguration [11], [12].However, these solutions have limited applicability and lackrigorous stability analysis. Even though the MAS concept tendsto be oversold, the potential of MAS has not been exploredfully. Recent progress in consensus and cooperative controlmake advanced MAS-based designs possible, such as in [13],[14].According to the MAS-based fully distributed control

strategy presented in this paper, each bus agent may have twofunction modules. The global information discovery modulediscovers the total available wind power and total demands. Theload sharing control module calculates the desired utilizationlevel and then sets the generation reference of a DFIG. By con-trolling the utilization levels of the DFIGs to a common value,the supply-demand balance can be maintained. In addition, thesolution can mitigate the impacts of inaccurate and outdatedmaximum wind power predictions. The deloaded generationreference of a DFIG can be tracked by coordinating converterand pitch angle controls. A detailed introduction and analysisof the deloading control strategy for the DFIG is provided inthis paper. A 5-DFIG microgrid system is used to demonstratethe performance of the proposed control scheme.The main contribution of this paper is the introduction of a

novel fully distributed control solution for the coordination ofmultiple DFIGs in an islanded microgrid that can ensure thestatic supply-demand balance and good dynamic performance.To calculate the desired utilization level, the consensus basedglobal information discovery algorithm proposed in [13], [14]is introduced. The algorithm can guarantee convergence for mi-crogrids of any size and topology and can be implemented usingsimple communication network. To realize generation referencetracking, the deloading strategies for both converter control andpitch angle control of a DFIG are designed and the implemen-tation details are provided.The rest of the paper is organized as follows. Section II

introduces the proposed MAS based control strategy for loadsharing among multiple DFIGs. Section III describes thecontrol implementation of DFIGs and synchronous generator.Section IV presents the simulation results with a 5-DFIG mi-crogrid. Section V provides discussion and concluding remarks.

II. PROPOSED CONTROL STRATEGY

The fully distributed control strategy is illustrated in Fig. 1.The microgrid contains distributed DFIGs, as well as a reli-able conventional generator (CG), as shown in Fig. 1. The CGprovides reactive power for voltage regulation and generatesadditional active power during low wind conditions. The mi-crogrid is connected to the main grid through a circuit breaker(CB).Each bus in a microgrid has a corresponding bus agent (BA).

The MAS based control system is fully distributed in the sensethat no hierarchical framework or specialized agent is used tocoordinate the operations of the autonomous agents. As illus-trated in Fig. 1, the communication topology (blue dash line)is the same as the topology of the power network. This typeof design can utilize the power line communication techniques.Based on other considerations, such as cost, convenience, andetc., the topology of the communication network can be de-signed to be different from that of the power network.

Fig. 1. Distributed control architecture.

A bus agent may have two function modules for global in-formation discovery (GID) and load sharing control (LSC), re-spectively. The functions of a specific BA are decided by theproperties of the corresponding bus. If a bus has no generationcapability, such as bus 3 in Fig. 1, then its corresponding BAonly has the function of GID. If a bus has generation capability,such as buses 1, 2, 4, and in Fig. 1, the corresponding BA hasboth functions of GID and LSC. The GID module is responsiblefor measurement and prediction of local generation/load and in-formation exchange among agents. The LSC module decidesthe common utilization level with the help of the GID moduleand sets the active power generation references for the corre-sponding DFIG. The DFIG receives the generation referenceinformation from the corresponding bus agent and realizes ac-tive power reference tracking (deloading) by coordinating theconverter and pitch angle controls.Due to the reduced amount of data to process, the MAS based

solution is able to provide faster response compared with thecentralized solutions. Since the distributed solution requires nei-ther a powerful central controller to process the huge amount ofdata nor a complicated communication network, it is less expen-sive to implement. In addition, the distributed solution is flexibleand able to automatically adapt to changes of operating condi-tions.

A. Consensus-based Active Power Setting

A microgrid can operate in either grid-connected mode or is-landed mode. In grid connected mode, the power grid can eitherabsorb surplus power from the microgrid or inject power intothe microgrid to compensate for the power shortage. Thus, ingrid-connected mode, the supply-demand balance is maintainedby the power grid and there is no need to use the load sharingalgorithm. On the other hand, when a microgrid operates underislanded mode, supply-demand should be balanced within themicrogrid to ensure that the system operates autonomously. Thisrequires the DFIGs to coordinate with each other properly.In this paper, only the microgrid under islanded mode is con-

sidered. The targeted problem of the fully distributed MAS-based algorithm for active power setting is introduced below.The total active power demand of a microgrid can be

represented using:

(1)

where is the number of buses in the microgrid, is the localnet demand calculated based on the local load and generation,


ZHANG et al.: FULLY DISTRIBUTED COORDINATION OF MULTIPLE DFIGS IN A MICROGRID FOR LOAD SHARING 3

and is the active power loss in the microgrid and can beconsidered as a small percentage of .The total available wind power generation of the DFIGs

can be calculated using:

(2)

where denotes the predicted maximum wind power gen-eration of the DFIG.If is larger than , all DFIGs should be operated under

MPPT control. If the reliable supply from the CG or power gridis insufficient to compensate for the difference, non-vital loadscan be disconnected until . If is less than , a suitable de-loading strategy is required to share the demand among DFIGs.As discussed in this paper, this is achieved by controlling the uti-lization levels of DFIGs, maintaining them at a commonvalue, as described in (3) and (4):

(3)

(4)

where is the active power generation setting of theDFIG.The above deloading strategy can guarantee the supply-de-

mand balance using:

(5)

The remaining problem regards how to determine the utiliza-tion level in a fully distributed way. The distributed algorithmshould be able to discover and through distributedcommunications. In this paper, the consensus-based global in-formation discovery algorithm developed in [13], [14] is usedfor this purpose. For convenience, the algorithm is briefly intro-duced here.The global information is discovered iteratively, and the up-

dating rule is formulated as:

(6)

where and are the information discovered by agentsand at iteration , respectively, is the immediate up-date of , is the coefficient of the information exchangedbetween agents and , and is the number of agents partici-pating in information discovery.Based on a rigorous stability analysis, all will converge to

the same value, as shown in (7), as long as the coefficientssatisfy certain constraints [16], [17].

(7)

Thus, by initializing with local demand and local max-imum available wind power , the above algorithm candiscover the averages of the required global information, i.e.,

and . Once the required global

information has been obtained, the desired utilization level canbe calculated as:

(8)

Due to wind and load fluctuations, the utilization level cal-culated according to (8) will change constantly. To avoid theproblem of chattering, a dead zone is introduced to decrease thesensitivity of to changes in operating conditions, as in:

(9)

where is the immediate update of calculatedaccording to (8), and .

B. Deloading Strategies for Control Reference Tracking

1) Rotor Speed Adjustment: When the available wind powerexceeds the demand, deloading tracking should be used insteadofMPPT. The strategy used in [6] is applicable only when islarger than 0.8 due to the accuracy of the linearized model. Thecontrol scheme proposed in [9] tunes the pitch angle frequentlyeven during medium and low wind conditions, which causesthe blade to chatter or strain. To overcome these limitations,the deloading control strategy proposed in this paper integratesoverspeeding control and pitch angle control.The power extracted from wind can be calculated according

to the well-known wind turbine aerodynamic characteristics:

(10)

where is air density, is the effective area sweptby wind turbine blades with being the blade radius, isthe wind speed, is the power coefficient of the wind turbine,which is a function of the tip speed ratio and the turbine bladepitch angle , and is defined using [15]:

(11)where is defined according to (12), with being the rota-tional speed of the rotor and defined according to (13).

(12)

(13)

The maximum available wind power is determined using thewind speed measurement, as shown in (14), where is themaximum of , and is constant.

(14)

For deloading tracking, the output power should be controlledusing:

(15)

where is defined as the new power coeffi-cient.



Fig. 2. Illustration of the overspeeding control strategy.

According to (12)–(13), the generated power can be adjustedby controlling rotor speed and/or regulating turbine pitchangle . Tuning the rotor speed is preferable for two reasons.First, the rotor speed can be altered faster than the pitch anglebecause it can be controlled through power converters, whichrespond faster than the blades’ actuators. Second, tuning therotor speed can protect the pitch blade from suffering wear andtear [10]. However, when the rotational speed reaches the upperbound, changing the pitch angle becomes the only option. Themechanism driving the overspeeding-based deloading strategyis described as follows.For a specific deloading condition, the power coefficient

and the rotor speed can be determined using (11). For ex-ample, when and , the maximum power coef-ficient under MPPT , the rotor speed is . As-sume is required for deloading. According to (15),the new power coefficient will be .Substituting with in (11) produces two solutions( and ). Based on (12), the correspondingrotor speed of the DFIG can be determined according to

, which is or . Thismeans that the deloading objective can be realized by either in-creasing or decreasing the rotor speed, as illustrated in Fig. 2.For example, when the wind speed is 10 m/s, the maximum

wind power can be harvested by setting the rotor speed to. For the deloading operation mentioned above, two can-

didate rotor speeds can be selected. For open loop control, op-erating point A is a stable equilibrium, while B is an unstableone. Studies also show that overspeeding can improve smallsignal stability [16] and facilitate frequency regulation. Due tothese advantages, an overspeeding-based deloading strategy isselected in this paper.2) Pitch Angle Adjustment: Pitch angle control is another

method of active power control. It is activated when the rotorspeed exceeds the predefined speed threshold. In this paper, 1.3p.u. is used as that threshold. When the rotor speed exceedsthis value, the pitch angle control system sets 1.3 p.u as therotor speed reference in order to protect the generator, whichcan also shed the extra energy according to the deloading com-mand. Fig. 3 illustrates the pitch angle control strategy. As-suming that the DFIG operates first at point C with a maximumpower extraction of wind speed of 12 m/s, it then receives com-mand from the bus agent, which requires it to operate under thedeloading condition with . According to the over-speeding deloading strategy, if it is not subject to speed limi-tations, the DFIG will accelerate following the tracking curveuntil it reaches the stable equilibrium C’. However, when therotor speed hits the upper boundary, at point D, the pitch anglecontrol is activated. Then, the DFIG will follow section DE by

Fig. 3. Illustration of the pitch angle control strategy.

Fig. 4. Control of a DFIG under deloading mode.

increasing the pitch angle until it reaches point E, where thepitch angle equals 1.608 .For deloading tracking, the electrical power converters (over-

speeding) control and the mechanical wind turbine (pitch angle)control need to be coordinated. If overspeeding control alone isenough for deloading, the rotor will increase the speed until itreaches the equilibrium (point A in Fig. 2). If the rotor speedhits the threshold (1.3 p.u. in Fig. 3), the pitch angle control willbe activated and a PI controller will be utilized to regulate thepitch angle until it reaches the equilibrium (point E in Fig. 3).

III. CONTROL IMPLEMENTATION

To implement the control strategy outlined above, the DFIGunits in a microgrid must be well controlled, which presentsa challenge. This section explains how the proposed controlstrategy is implemented through the controls of DFIGs and thesynchronous generator (SG).

A. Control of DFIGs

Once the references for active power generation havebeen decided, the control reference can be tracked by control-ling the DFIG. In this way, the overall system’s active powersupply-demand balance can be maintained, and the operation ofmultiple DFIGs can be coordinated. In addition to active power,reactive power and the voltages of the terminal bus and dc-linkvoltage also must be controlled. As illustrated in Fig. 4, the con-trol of a DFIG has three modules for electrically controlling thetwo converters and mechanically controlling the pitch angle.The active power is controlled through the rotor side con-

verter (RSC). The reactive power injection can be obtained andadjusted by controlling the RSC, the grid side converter (GSC),or both. For the voltage control, the rotor-side converter is pre-ferred [17]. Thus, RSC is selected in this paper to control bothactive power and reactive power, and GSC is used only to stabi-lize the dc-link voltage. More information about the three con-trol components is provided below.



Fig. 5. Control of the rotor side converter.

1) RSC Control: In this paper, the field-oriented control(FOC) strategy is used for the DFIG converter controls. Theload sharing control does not impose any constraints on theFOC strategy. Simulation studies show that this conventionalvector control has no problem tracking the control referencesgenerated by load sharing control module. The FOC enablesdecoupled controls of P and Q by regulating the componentsof the rotor currents [18] through the RSC. Two main FOCcontrol strategies are enabled by setting or to zero,respectively [19]. In this paper, is set to zero; thus, theactive power is controlled by d-axis rotor current , while thereactive power is controlled by q-axis rotor current . Theblock diagram of the RSC controller is shown in Fig. 5.The active power reference generated by the load sharing

control module is adjusted by to generate the actualpower reference is used to compensate the inaccura-cies that cause power imbalance, which includes: 1) the mea-surement errors, 2) the error due to simplifications and assump-tions during the development of the speed-power characteristicsshown in Fig. 2, 3) the inaccuracy due to the outdated informa-tion resulted from the delay with the decision-making process,and 4) the estimation error of power loss in the microgrid, whichis approximated as a fixed percentage (3%) of the total load. Asshown in Fig. 5, the slight active power imbalance can be com-pensated by a PI controller taking frequency deviation asinput. Then, the difference between the output active power ofDFIG and reference value forms the error signal thatis processed by the PI controller to produce rotor current .Similarly, the difference between rotor current and referencevalue generates another error signal for the PI controller toproduce rotor voltage . A compensator term, , is addedto the PI controller to improve the dynamic performance [17].The two modes of reactive power control for DFIGs are

voltage regulation mode and reactive power control mode.Both modes regulate q-axis rotor current . In voltage controlmode, can be divided into magnetizing componentand terminal voltage control component . The magne-tizing component compensates for the no-load reactive powerabsorbed by the DFIG, and the voltage control component iscontrolled in response to voltage fluctuations. The control loopfor voltage regulation also is shown in Fig. 5. Thesignal controls the operational mode of the DFIG. The voltagedifference between and reference value forms thereference through a PI controller, compensated by , to

Fig. 6. Control of the grid side converter.

Fig. 7. Scheme of pitch angle control.

Fig. 8. Control of the synchronous generator.

generate the q-axis reference value of rotor current . The-axis voltage reference is generated using a PI controller and-axis voltage compensator [17].For reactive power control, the difference between the reac-

tive command and the reactive power output formsrotor current reference through a PI controller. The reactivepower control loop also is shown in Fig. 5.2) GSC Control: Fig. 6 shows the overall control scheme

of the GSC. The dc-link voltage is controlled by regulatingthe d-axis grid side current , which can regulate the activepower that the dc link exchanges with the grid to maintain thedc voltage [20], [21].3) Pitch Angle Control: The pitch angle controller used in

this paper is depicted in Fig. 7, it consists of the PI controller andthe pitch angle actuator. The threshold speed is set to 1.3 p.u, andthe is set to zero. The maximum pitch angle change rate islimited by and , which are set to .

B. Control of the Synchronous Generator

In this paper, a synchronous generator (SG) is used to illus-trate the control of the conventional generators. Tomaximize theenergy efficiency, the SG is controlled only to generate reactivepower for voltage regulation when the wind energy is sufficient.When the wind energy is insufficient, the SG should also gen-erate active power to compensate for the active power shortage.The control logic of the SG is shown in Fig. 8. In order to modelthe ramp rate of the SG, a rate limiter is added in thecontrol loop, as in Fig. 8. The governor system is mainly a PIcontroller, which is similar to one of the models proposed in[22]. The excitation system block is implemented with a dc ex-citer, as described in [23].



Fig. 9. Configuration of a 6-bus microgrid.

Fig. 10. Global information discovery processes. (a) Average available windpower discovery. (b) Average load demand discovery.

IV. SIMULATION STUDIES

The proposed cooperative control method is tested with a6-bus microgrid system, as shown in Fig. 9. During simula-tion studies, both MAS and microgrid are implemented usingSimulink and the Simpowersystems toolbox. The microgridcontains six loads, five DFIGs, and one SG. DFIGs 1, 2, and4 are controlled in var regulation mode, and DFIGs 3 and 5are controlled in voltage regulation mode, as introduced inSection II. The active power output of the SG is maintained atzero when the wind power is sufficient. When the wind poweris insufficient, the SG is controlled to increase the output forregulating the frequency. The ramp-up and ramp-down rates ofthe SG are both set to 0.02 p.u./s.In this section, an example of the information discovery

process is provided to explain the calculation of the activepower references of the DFIGs. The control scheme is testedunder two operating conditions, i.e., constant winds and loads,and variable winds and loads. The first operating condition,while unrealistic, is easier to understand due to its simplicity.

A. Consensus-Based Information Discovery

The active power references of the DFIGs are calculatedbased on the consensus-based load sharing algorithm. Ac-cording to the algorithm, the averages of the total availableactive power generations and total loads are first discovered. Anexample of the information discovery process for the averagesis shown in Fig. 10.In Fig. 10, the available DFIG wind generations are 1.37,

4.50, 4.50, 6.00 and 5.45MW, and the loads are 3.00, 2.05, 4.50,6.00, and 4.69 MW, respectively. The local active power gener-ation and the local demand are both set to 0 for the bus to whichthe SG is connected. Thus, the total available wind generationis 21.82 MW, and the total load is 20.24 MW. After informa-tion discovery, the average available wind power converges to

, and the average total load converges to. The desired utilization level can be calcu-

lated according to . Then, the activepower generation references of the DFIGs can be set accordingto .In the example above, the available wind power is more than

required. Thus, the active power reference of the SG is set to

zero, and the SG only produces reactive power for voltage reg-ulation. If the available wind power is insufficient, the SG’s ac-tive power control will be activated. Load shedding might benecessary if the gap exceeds the capacity of the SG. Some sim-ulation results will be provided later in this section.The speed of the information discovery process is evaluated

using a number of iterations instead of time. The actual timeis decided by the complexity of the system and the techniquesused to implement both software and hardware. According tothe analysis in [13], the algorithm can converge within 0.15second for the IEEE 162-bus system. Thus, the proposed algo-rithm has no problem to realize the 0.5 second updating intervalof the active power generation references. In addition, one canconclude that the proposed control solution is scalable and canbe applied to a fairly large system.In this paper, the interval for reference updates is set to 0.5 s.

Due to the simplicity of the algorithm, the control reference canbe updated much more frequently. However, excessively fre-quent updates might make the system unable to follow. Thus,the time interval of reference update needs to be coordinatedwith the response speed of a DFIG. For a closed loop controlsystem, the response speed is not only decided by the time con-stant of a single component, bus also the parameter of the con-trollers. Simulation studies show that the value of 0.5 s is largeenough for the system to follow and can render good dynamicperformance.

B. Constant Winds and Constant Loads

During this test, the demands of the loads are held constantto obtain a total load of 20.24 MW and 6.40 MVar. In this testscenario, constant power load model is used. The wind speedsat the DFIGs are 10 m/s, 13.5 m/s, 14 m/s, 13 m/s, and 11 m/s,respectively, which may generate as much as 21.82MWof windpower. The SG’s active power output is set to 4 MW beforeislanding. An islanding event is simulated at 60 s to study thedynamic performance of the proposed control solution duringtransition of different operating modes.The output of the SG can be any value within its generation

limits.When themicrogrid is operating in grid-connected mode,the generation is decided by grid requirements. In this test, theinitial output of the SG is intentionally set to a high value. Theobjective is to create sufficient disturbances. During controlledislanding, the set point of the SG can be set to a value as closeas possible to the value for autonomous operation. In this paper,the available wind power is assumed to be more than requiredwithin the microgrid. For this particular situation, the generationof the SG can be adjusted to zero before deploying the islandingoperation.Fig. 11 shows the updating process of the utilization level

profile. The DFIGs are controlled using the MPPT algorithmwhen the microgrid is connected to the power grid, so the initialutilization level is 100%. After islanding, the utilization levelcontinues to update based on instant generations and loads andwill finally converge to . A zoomed-insegment of the utilization level updating process is shown inFig. 12. The values of the dead zone parameters (0.01) and theupdating interval (0.5 s) can be observed clearly. The smallerthe two parameters are, the more sensitive the algorithm is tochanges in operating conditions. However, over-sensitivity willcreate unnecessary disturbances that will decrease the system’s



Fig. 11. Updating process of the utilization level during islanding.

Fig. 12. Zoomed updating process of the utilization level.

Fig. 13. System’s frequency responses.

Fig. 14. Active and reactive power generations of the synchronous generator.

stability. If the values of the parameters are too large, the algo-rithmwill not be able to provide a timely response. For a specificsystem, the two parameters should be selected by trial and error.The system’s frequency response is shown in Fig. 13. As can

be seen in Fig. 11, the total available wind power is more thantotal load demand before islanding. Since the DFIGs are con-trolled at MPPT mode initially, the total generation is more thanload demand in the microgrid and the excessive active powergeneration is absorbed by the main grid. Immediate after is-landing, the active power generation is more than load demand.Thus, an increase in frequency can be observed as can be seenin Fig. 13. However, the size of the overshoot is very small andfalls within the allowable range of operation. Within a short pe-riod of time, the frequency response is able to converge to thenominal value. Similar phenomena can be observed in Fig. 27.As noted previously, performance can improve if the generationof the SG can be adjusted to a suitable level during intentionalislanding.The active and reactive power generations of the SG are

shown in Fig. 14. The active power will eventually decrease

Fig. 15. Deloading tracking of DFIG 1. (a) Available and generated activepowers. (b) Rotor speed and pitch angle.

Fig. 16. Deloading tracking of DFIG 4. (a) Available and generated activepowers. (b) Rotor speed and pitch angle.

Fig. 17. Terminal voltages responses of the DFIGs and SG.

Fig. 18. Zoomed terminal voltage responses of the DFIGs and SG.

Fig. 19. Load profiles with different load models.

to zero with the SG’s ramp-down rate of speed. The reactivepower out is nonzero due to voltage regulation requirements.Figs. 15 and 16 show the dynamic responses of DFIGs 1 and

4, respectively, demonstrating that both of the active power gen-erations converge to a value below the maximum availability.The utilization levels, if calculated, will be the same as the valuecalculated previously. Fig. 15(b) indicates that deloading re-quires overspeeding. Because the rotor speed is always less thanthe triggering value (1.3 p.u.), pitch angle control will be deac-tivated. For DFIG 4, overspeeding will cause the rotor speed toexceed 1.3 p.u. Thus, pitch angle control will be reactivated. In



Fig. 20. Voltage profiles with dynamic load model.

Fig. 21. Wind profiles of the DFIGs.

Fig. 22. Utilization level profiles.

Fig. 23. Active power tracking of DFIG 4.

Fig. 24. Rotor speed and pitch angle responses of DFIG 4.

the end, the rotor speed will stabilize at 1.3 p.u., and the pitchangle will converge to a larger value.The terminal voltage responses of the DFIGs and the SG

are shown in Figs. 17– 18. Since some DFIGs are operated atvoltage regulation mode, their terminal voltages respond fasterthan those of other buses, as can be observed from Fig. 18.Considering the complexities of loading conditions, it is nec-

essary to test the proposed control scheme with different loadmodels. Thus, in addition to the constant PQ load model used inprevious study, a dynamic load model described in [24] is alsoused in the following study. According to the dynamic loadmodel, the active and reactive powers absorbed by a load are

Fig. 25. Active and reactive power generations of the SG.

Fig. 26. System’s frequency response.

Fig. 27. Terminal voltages responses of the DFIGs and SG.

functions of the corresponding bus voltage. Fig. 19 comparesthe load profiles with different load models during islanding.It can be seen that both static and dynamic load responses aredifferent. Fig. 13 compares the microgrid’s frequency responseswith different load models. Fig. 20 shows the voltage responseswith the dynamic load models. By comparing Figs. 18 and20, one can say that different load models only slightly affectsystem’s dynamic responses. In another word, the proposedcontrol scheme is robust against changes in loading conditions.

C. Variable Wind and Fluctuated Load

This test utilizes practical measured wind speed data. Tocreate sufficient diversity, different segments of the measureddata are used to model the wind speeds of different DFIGs, asshown in Fig. 21. In addition, load fluctuation is simulated bydecreasing all loads by a total of 2 MW at 150 s and revertingback to the original load at 200 s. Similar to the previous test,the islanding event occurs at 60 s, and the active power outputof the SG is set to 4 MW before islanding.Fig. 22 shows the utilization level profiles of the DFIGs with

respect to the total available wind power and total load. The uti-lization level decreases when available wind power increasesand increases when available wind power decreases. From 60s to 200 s, the utilized level is less than 1, which means thesystem is operating under high wind condition. From 200 s to240 s, the system is operating under low wind condition, whichmeans wind generation alone is insufficient to support all loads.In that case, all DFIGs operate at MPPT mode and the genera-tion deficiency is compensated by the conventional SG.Fig. 23 compares the actual active power generation with the

available/maximum wind power of DFIG4. Note that the DFIGoperates in MPPT mode during grid connection and when the



total available wind power is insufficient. Otherwise, the ac-tual generation is less than the available wind power. Similarresponses can be observed for other DFIGs. The demands areshared based on the capabilities of the DIFGs, so a DFIG usuallydoes not need to work at full capacity. This can significantly re-duce disturbances caused by inaccurate maximum wind powerpredictions.Fig. 24 shows the rotor speed and pitch angle response of

DFIG4. Note that the pitch angle control was activated whenthe rotor speed exceeded 1.3 p.u. Increasing the pitch angle de-creases the rotor speed. If the rotor speed falls below 1.3 p.u.,the pitch angle control will be deactivated. Between 195 s and278 s, the pitch angle falls to 0, which is the used in thepaper.Fig. 25 shows the active and reactive power generation of the

SG. After islanding, the SG’s active power generation decreasesto zero because there is sufficient wind generation. When a 2MW load is connected back to the microgrid at 200 s, the SG iscontrolled to generate active power to compensate for the powershortage. Because the original active power generation duringgrid-connected mode is set to a large value intentionally, a rela-tively larger disturbance can be observed during islanding. Thedecreasing slope of active power generation is decided by theSG’s parameters. A faster SG may improve the dynamic re-sponse of the system. In autonomous mode, the SG’s reactivepower generation is nonzero due to the need for voltage stabi-lization.The voltage and frequency responses are of immediate in-

terest to customers, so it is necessary to evaluate those usingstandards and regulation codes. According to IEEE Std. 1547,normal frequencies should range from 59.8 Hz to 60.5 Hz [25].In steady state, ANSI/NEMA C84.1 specifies that the frequencydeviation should not exceed , and the voltage shouldfall between 0.95 p.u. and 1.05 p.u. The system frequency re-sponse shown in Fig. 26 indicates that the maximum frequency(60.38 Hz) is less than 60.5 Hz. As shown in Fig. 27, the voltageresponse is in the range of (0.948 p.u., 1.035 p.u.). Both re-sponses satisfy the above standards. The largest deviations ap-pear during islanding, and the system is quite stable against loadfluctuations.From Fig. 26, it can be seen that frequency can be controlled

using DFIGs only if available wind power is enough to supportall loads, such as the case from 150 s to 200 s.

V. CONCLUSION AND DISCUSSION

This paper presented a fully distributed algorithm for coor-dinating multiple DFIGs in microgrids. Due to the distributedproperty, the solution is stable, reliable, adaptive, scalable, andcost efficient. Simulation studies have demonstrated the effec-tiveness of the proposed solutions. Suitable algorithms can beadopted to realize load sharing control for different types ofrenewable energy resources. Thus, the proposed algorithm canbe applied widely to various microgrids consisting of differenttypes of distributed energy resource units.The load sharing algorithm aims at maintaining the active

power supply-demand balance within an islanded microgrid,which will regulate frequency. To compensate the inaccuraciesthat cause power imbalance, PI controls with frequency devia-tion as input are introduced, as shown in Figs. 5 and 8. Sincethe PI control is only used to compensate the inaccuracy of theload sharing control, the PI control is secondary in term of loadsharing.

Unlike the inertia of a synchronous generator that can be rep-resented using “ ,” the inertia of the microgrid under closedloop control is hard to quantify. Usually, the inertia of an is-landed wind-powered microgrid is much smaller than that of apower system composed of traditional synchronous generators.The control of a system with reduced inertia is much more diffi-cult. According to the simulation results, it can be seen that theproposed controller is effective and the control performance issatisfactory.Just like other existing solutions, the control scheme has cer-

tain requirements on operating conditions, such as the SG is ableto compensate the generation deficiency of the DFIGs and thedeloading strategies can regulate the DFIG’s output to a desiredvalue in a timely manner. Under certain extreme operating con-ditions, such as gust wind, the proposed control strategy mighthave trouble in reducing the generation of a DFIG to a desiredvalue or responding fast enough. For these extreme situations,other control options, such as shedding load(s) or disconnectingDFIG from the microgrid, can be adopted.

VI. APPENDIX

Below are the parameters of the 6-bus microgrid.The parameters of the synchronous machine are:

, , ,, , , ,

, , , ,, .

The parameters of 5 DFIGs are:DFIG 1: , , ,

, , ,, , , and active power

output at wind speed of 12 m/s is 0.7865 p.u.DFIGs 2 and 3: , ,

, , ,, , ,, and active power output at wind speed of

12 m/s is 0.7865 p.u..DFIG 4: , , ,


output at wind speed of 12 m/s is 0.7865 p..DFIG 5: , , ,


output at wind speed of 12 m/s is 0.7865 p.u..The control parameters of the RSC controller (Fig. 5) are:The active power regulator: , .The current regulator ( -axis): , .The reactive power regulator: , .The voltage regulator: , .The current regulator ( -axis): , .The control parameters of the GSC controller (Fig. 6) are:The voltage regulator: , .The current regulator ( -axis): , .The current regulator ( -axis): , .The control parameters of the pitch angle controller (Fig. 7)

are:, , the maximum rotor speed is 1.3

p.u., degrees, degrees, the rate limiter is, and actuator time constant is 0.1 s.



REFERENCES

[1] U.S. Dept. Energy, “20% Wind Energy by 2030: Increasing wind en-ergy’s contribution to U.S. electricity supply,” U.S. Dept. Energy Rep.,2008.

[2] W. Leonhard and E. M. Grobe, “Sustainable electrical energy supplywith wind and pumped storage—A realistic long-term strategy orutopia?,” in Proc. IEEE Power Eng. Soc. Gen. Meet., Jun. 2004, vol.2, pp. 1221–1225.

[3] A. Daneshi, M. Khederzadeh, N. Sadrmomtazi, and J. Olamaei, “Inte-gration of wind power and energy storage in SCUC problem,” in Proc.World Non-Grid-Connected Wind Power Energy Conf. (WNWEC),Nov. 2010, pp. 1–8.

[4] L. Qu and W. Qiao, “Constant power control of DFIG wind turbineswith supercapacitor energy storage,” IEEE Trans. Ind. Appl., vol. 47,no. 1, pp. 359–367, Jan./Feb. 2011.

[5] J. L. Rodriguez-Amenedo, S. Arnalte, and J. C. Burgos, “Automaticgeneration control of a wind farm with variable speed wind turbines,”IEEE Trans. Energy Convers., vol. 17, no. 2, pp. 279–284, Jun. 2002.

[6] R. G. de Almeida, E. D. Castronuovo, and J. A. P. Lopes, “Optimumgeneration control in wind parks when carrying out system operatorrequests,” IEEE Trans. Power Syst., vol. 21, no. 2, pp. 718–725, May2006.

[7] H. Xin, Z. Qu, J. Seuss, and A. Maknouninejad, “A self-organizingstrategy for power flow control of photovoltaic generators in a distribu-tion network,” IEEE Trans. Power Syst., vol. 26, no. 3, pp. 1462–1473,Aug. 2011.

[8] M. Wooldridge, An Introduction to Multiagent Systems. Hoboken,NJ, USA: Wiley, 2002.

[9] P. Moutis, E. Loukarakis, S. Papathanasiou, and N. D. Hatziargyriou,“Primary load-frequency control from pitch-controlled wind turbines,”Proc. 2009 IEEE Bucharest PowerTech, pp. 1–7.

[10] E. Loukarakis, I. Margaris, and P. Moutis, “Frequency control sup-port and participation methods provided by wind generation,” in Proc.Elect. Power Energy Conf. (EPEC) 2009, pp. 1–6.

[11] J. M. Solanki, S. Khushalani, and N. N. Schulz, “Amulti-agent solutionto distribution systems restoration,” IEEE Trans. Power Syst., vol. 22,no. 3, pp. 1026–1034, 2007.

[12] K. Huang, D. A. Cartes, and S. K. Srivastava, “A multiagent-based al-gorithm for ring-structured shipboard power system reconfiguration,”IEEE Trans. Power Syst., vol. 37, no. 5, pp. 1016–1021, 2007.

[13] Y. Xu andW. Liu, “Novel multi agent based load restoration algorithmfor microgrids,” IEEE Trans. Smart Grid, vol. 2, no. 1, pp. 140–149,Mar. 2011.

[14] Y. Xu, W. Liu, and J. Gong, “Multi-agent based load shedding algo-rithm for power systems,” IEEE Trans. Power Syst., vol. 26, no. 4, pp.2006–2014, Nov. 2011.

[15] S. Heier, Grid Integration of Wind Energy Conversion Sys-tems. Chicester, U.K.: Wiley, 1998.

[16] N. A. Janssens, G. Lambin, and N. Bragard, “Active power controlstrategies of DFIG wind turbines,” Proc. IEEE Power Tech, Lausanne,pp. 516–521, Jul. 2007.

[17] O. Anaya-Lara, N. Jenkins, J. Ekanayake, P. Cartwright, and M.Hughes, Wind Energy Generation: Modeling and Control. Hoboken,NJ, USA: Wiley, 2009.

[18] A. Luna, F. K. de Araujo Lima, D. Santos, P. Rodriguez, E. H.Watanabe, and S. Arnaltes, “Simplified modeling of a DFIG for tran-sient studies in wind power applications,” IEEE Trans. Ind. Electron.,vol. 58, no. 1, pp. 9–20, Jan. 2011.

[19] P. Krause, O. Wasynczuk, and S. Sudhoff, Analysis of Electric Ma-chinery and Drive Systems, 2nd ed. New York: Wiley-IEEE Press,2002.

[20] R. P. S. Chandrasena, A. Arulampalam, J. B. Ekanayake, and S. G.Abeyratne, “Grid side converter controller of DFIG for wind powergeneration,” in Proc. Int. Conf. Ind. Inf. Syst. (ICIIS), Aug. 2007, pp.141–146.

[21] E. Tremblay, A. Chandra, P. J. Lagace, and R. Gagnon, “Study of grid-side converter control for grid-connected DFIG wind turbines underunbalanced load condition,” in Proc. IEEE Int. Symp. Ind. Electron.,Jul. 2006, vol. 2, pp. 1619–1624.

[22] “IEEE committee report, Dynamic models for steam and hydro tur-bines in power system studies,” IEEE Trans. Power App. Syst., vol.PAS-92, no. 6, pp. 1904–1915, 1973.

[23] Recommended Practice for Excitation System Models for PowerSystem Stability Studies, IEEE Std. 421.5-1992, Aug. 1992.

[24] A. Borghetti, R. Caldon, A. Mari, and C. A. Nucci, “On dynamic loadmodels for voltage stability studies,” IEEE Trans. Power Syst., vol. 12,no. 1, pp. 293–303, Feb. 1997.

[25] IEEE Standard for Interconnecting Distributed Resources with ElectricPower Systems, IEEE Std. 1547-2003, 2003.

Wei Zhang (S’11) received the B.S. and M.S. de-grees both in power system engineering from HarbinInstitute of Technology, Harbin, China, in 2007 and2009, respectively. Currently, he is working towardthe Ph.D. degree at the Klipsch School of Electricaland Computer Engineering of New Mexico StateUniversity, Las Cruces.His research interests include distributed control

and optimization of power systems, renewable en-ergy and power system state estimation, and stabilityanalysis.

Yinliang Xu (S’10) received the B.S. and M.S.degrees both in control science and engineeringfrom Harbin Institute of Technology, Harbin, China,in 2007 and 2009, respectively. Currently, he isworking toward the Ph.D. degree at the KlipschSchool of Electrical and Computer Engineering ofNew Mexico State University, Las Cruces.His research interests include intelligent control

and optimization of microgrids.

Wenxin Liu (S’01-M’05) received the B.S. andM.S.degrees from Northeastern University, Shenyang,China, in 1996 and 2000, respectively, and thePh.D. degree in electrical engineering from MissouriUniversity of Science and Technology (formerlyUniversity of Missouri-Rolla) in 2005.From 2005 to 2009, he was an Assistant Scholar

Scientist with the Center for Advanced PowerSystems of Florida State University, Tallahassee. Heis currently an Assistant Professor with the KlipschSchool of Electrical and Computer Engineering of

NewMexico State University, Las Cruces. His research interests include powersystems, control, computational intelligence, and renewable energy.

Frank Ferrese (M’04) received the B.S. degreein electrical engineering from Drexel University,Philadelphia, PA, and the M.S. degree in computerengineering from Villanova University, RadnorTownship, PA.He is a Research Engineer at the Naval Surface

Warfare Center, Carderock Division, Philadelphia.He currently does research in the area of controlsand automation.

Liming Liu (M’08–SM’11) received the Ph.D. de-gree in electrical engineering from Huazhong Uni-versity of Scientist and Technology, China, in 2006.He is currently an Assistant Scientist in the Center

for Advanced Power Systems, Florida State Uni-versity, Tallahassee. His research interests includerenewable energy conversion systems, modelingand control of multilevel inverter applications,motor drive control with hybrid energy storages, andflexible ac transmission system.

fully distributed coordination of multiple dfigs in a microgrid for load sharing

Documents