Evaluating Unintentional Islanding Risks for a HighPenetration PV Feeder

Mohammad Nikkhah Mojdehi∗ and Prasanta Ghosh†∗Energy and Microgrid Group, O’Brien and Gere Engineers, 333 W Washington St, Syracuse, NY 13202

Email: mohammad.mojdehi@obg.com†Department of Electrical Engineering and Computer Science, Syracuse University, Syracuse, NY 13210

Email: pkghosh@syr.edu

Abstract—The continued adoption of DGs is quickly alteringthe composition of distribution systems (DSs). As the penetrationand saturation of these new distributed generators (DGs) escalate,new challenges in the planning and operations of DSs are arising.At current penetration level, DGs represent a rare or minimalthreat to the stable DS operation. However, in a fully maturemarketplace in the future, high saturation of DG can createthe need for more diligent review and analysis. Commonly-used software applications in utility industry are not capableof analyzing the challenges, especially those requiring time-based analysis. This paper presents a simulation platform whichenables utilities to evaluate the risk of unintentional islandingof interconnected DGs. The platform uses available data fromutilities’ existing database to generate an accurate model of DSand calculates the run-on time of interconnected DGs after theloss-of-main grid. Results of a real case study shows the accuracyand functionality of the platform. The developed platform can beused by the utilities’ engineers to facilitate the interconnection ofDGs.

I. INTRODUCTION

Distribution systems (DSs) have been designed based ona vertical operating framework which is predicated on largecentral generations producing energy through transmissionand substations to end-users. However, existing DSs mayneed to be prepared for accommodating distributed generators(DGs), implementing a non-vertical operating approach, andenabling end-users to participate in different market-basedactivities. Mandatory policies supporting energy resiliency andutilization of renewable energy resources, emphasize the needto upgrade or build capable DSs. Along those policies, anew vision for DS planning and operation is necessary toaddress weaknesses and potential solutions. As can be inferred,the key stakeholders in this context are utilities. However,current utility practices limit capabilities as the industry andDS’ transform. One of these limitations is the lack of acomprehensive analytical tool for modeling and managing theimpact of any change/upgrade on DSs.

The next generation DSs represent multi-domain complexsystems and there are no comprehensive analytical tools torespond to all concerns and challenges simultaneously [1].Utilities’ analytical demand varies based on the nature ofthe application. For instance, steady-state applications such asstatic power flow based analysis have been utilities’ traditionalanalytical practices for planning and operating purposes. How-ever, moving toward the next generation DS, new concerns andchallenges drive the needs to evaluate transient-state and real-time applications. Descriptions and list of available tools toanalyze transient-state and real-time applications can be found

in [2]-[3]. Among all developed tools, three major deficits canbe identified. First, there are still gaps in developed tools tocover all transient-state applications (such as islanding) andreal-time applications (such as island operation). Second, thereis not a comprehensive tool which can cover all applications:steady-state, transient-state, and real-time. Third, utilizing anew tool in utility industry imposes significant cost due topersonnel training and adapting existing data base. As men-tioned earlier, utilities commonly use software applications,like CYME/CYMDIST, which are primarily used for steadystate studies. Those software applications can provide alreadystructured database of load information, layout of the electricaldistribution system, and the different electrical parameters ofthe network equipment. It would be desirable to link existingelectric power simulation tools with innovative approaches andperform required analysis.

Anti-islanding or loss-of-main grid protection is an im-portant aspect of protection related challenges of DGs inter-connection. Islanding can be defined as a condition in whicha portion of the utility system contains both load and DG,remains energized while being isolated from the remainderof DS [4]. Due to safety and transient stability concerns,IEEE 1547 (2003) requires that the loss-of-main-grid must bedetected by DGs within 2 seconds and must lead to immediatetrip of the DGs from DS. Although DGs are equipped byanti-islanding schemes, those schemes are not 100% reliable.They present an inherent operating region, characterized bysmall power imbalances in an islanded system, where theyare not able to detect islanding condition in a timely manner[5]. The corresponding system operating conditions are callednon-detection zone (NDZ) and the islanding detection failureis named unintentional islanding [6].

From the utility stand point, identifying NDZs is veryimportant because they may violate the IEEE 1547 standardrequirements for unintentional islanding. Therefore, an analyt-ical tool which can perform in-depth and sufficiently rigorousislanding analysis, could help utilities to assess the risk ofunintentional islanding in more reliable and cost-effectivefashion and also facilitate interconnection of DGs in a safermanner. In this paper, a simulation platform is presented whichcan help utility companies to assess the risk of unintentionalislanding of interconnected DGs. The platform is simple toimplement and fast, which means savings in cost and time ofislanding studies. The platform can be used in a plug and playfashion with any kind of anti-islanding schemes used by DGs’manufacturer.

II. INTERCONNECTION SCREENING

Interconnection procedures may vary depending on thestate or federal jurisdiction, and implementation practices maychange by the utility. Several screening procedure have beenreported in the literature, such as the Sandia guideline [7]. Theguideline, based on a four-step procedure, indicates when therisk of islanding may not be negligible.

All screening procedures are used to avoid in-depth ad-ditional study to assess the risk of unintentional islandingof a proposed DG system. However, considering the rapidsaturation of DGs, it is expected that existing simplifiedscreening approaches could not address the risk. Therefore anin-depth study seems necessary. In order to reduce the time andcost of DG systems’ installation, the study should be effectiveand easy to implement. An analytical tool which can leveragecurrent utilities’ practices and software applications to conductislanding studies can perform the task in a cost-effective man-ner. The analytical tool must be able to model all componentsof the DS including distribution feeder, transformers, loads,and DGs. The following section presents the approach used inthe developed analytical tool to model the distribution feeder.

III. DISTRIBUTION FEEDER MODELING

DGs are connected to distribution feeders, therefore gen-erating an accurate model of the feeder is the first step inthe simulation platform. However, the size of the feeders andnumber of elements in the feeders make this step complex andtime consuming. In this section a simple algorithm is presentedto model the distribution feeder from the substation to the endnodes of the feeder.

The utilities’ commonly-used software applications, suchas CYME/CYMDIST, use the feeder data for different simula-tions. The data has been defined and structured specifically foreach software application. Therefore, integrating the feeder’sraw data into the platform results in an analytical tool whichcan interact with many software applications. To do so, the toolmust translate the raw data into the accurate feeder model. Theraw data includes nodes, lines (three-phase or single phase linewhich could be underground or overhead cables), transformers,capacitor banks, regulators, and loads. Each component isidentified by a unique ID. Creating the feeder topology anddetecting the nodes’ and other components’ connectivity is thekey element in distribution feeder modeling.

A simple algorithm has been used in the developed tool todetect distribution feeder topology, nodes’ connectivity, and allpaths from the substation to the end nodes of the feeder [8].The algorithm which can be applied to radial and normallyopen-loop DSs with any size and number of nodes, worksbased on the graph theory. A radial DS can be considered asa tree graph G which is connected and has no loop. Thereforethe distribution feeder can be modeled as a connected directedtree graph with n vertices and n-1 edges. A color can then beassigned to each node, using the coloring approach by Welshand Powell [9]. The algorithm then detects the connectivity ofall nodes and generates the accurate topology of the feeder bytracking the assigned colors of nodes.

At this point, the tool has modeled the distribution feeder.The next step is to model interconnected DGs.

IV. DG MODELING

Among all DG technologies, photovoltaic (PV) systemshave attracted considerable attention and investment in severalcountries such that a significant penetration of PV energyinto the DS is anticipated. Therefore, the focus of this paperis on PV systems. A PV system includes a PV generator(consisting of many PV panels) along with a boost DC/DCconverter connected to the DC side of a three-phase (3ph)DC/AC converter, a LC low-pass filter which is connected tothe AC side of the 3ph DC/AC converter, and an isolationtransformer to connect the rest of the system to the grid. TheDC/AC converter is considered as a voltage-sourced converter(VSC) in this paper.

For faster simulations, considering the complexity of theDS, an equivalent dynamic average-value model was used. Insuch a model, which is known as averaged model, instead ofapplying switching scheme, terminal variables of the VSC areapproximated by their respective per-switching-cycle movingaverage values [10]. An important feature of a grid-connectedPV system is the synchronization which identifies the angle ofthe grid voltage and can be achieved with variety of methods. Aphase locked loop control system based on the model providedby [11], was used for synchronization purpose. Assuming afixed input DC power at DC side of the VSC, the variationin sun irradiations and its effect on DC-link voltage could beignored for islanding study purposes. Also based on currentpractices in utilities, PVs are not allowed to perform voltageregulation, instead voltage and frequency is regulated by thegrid. Therefore, the real and reactive power control schemepresented by [10] was used in this paper. It is worth noting thatdue to the ongoing cost-performance improvement in powerelectronics and controls, IEEE 1547A (2014) standard allowsDGs to regulate voltage by changing their active and reactivepower, ride through abnormalities of grid voltage, and providemodulated power as a function of frequency. In response tothe standard, few utility companies considering the voltageregulation by DGs. In those cases, the PV model can bemodified to enable DGs for voltage regulation.

Islanding detection method (IDM) is another importantfeature of PV systems. Local IDMs are classified as passiveand active techniques. Passive IDMs rely on non-perturbedmeasurements such as frequency and voltage to detect island-ing condition. Over/under frequency protection (OFP/UFP)and over/under voltage (OVP/UVP) are examples of commonlyused passive methods. On the other hand, a perturbation is usedin active IDMs to drift the frequency or voltage beyond certainthreshold values.

In voltage (frequency) feedback approach, when theinverter-sensed output voltage (frequency) is increasing, theanti-islanding feedback will command the inverter active (re-active) power output to be increased. In the absence of thestiff grid, the voltage (frequency) will keep increasing in orderto balance the active (reactive) power. The increased voltage(frequency) will further drive the inverter active (reactive)power up due to the anti-islanding feedback. As a result,the voltage (frequency) will be eventually out of the nominalranges so that the islanding can be detected. Similar but oppo-site destabilization occurs when the sensed voltage (frequency)is decreasing initially. The most widely used active IDMsare Sandia frequency shift (SFS) and slip-mode frequencyshift (SMS). The basic idea behind SFS and SMS IDM is to

introduce a small increase or decrease in the current frequencyof the inverter. The positive feedback mechanism enables theinverter to push the frequency in the same direction as thedisturbance. In the absence of the stiff grid, the deviation inthe voltage frequency measured at the terminal increases whichresults in indication of the occurrence an islanding event [12].

A. SFS IDM

SFS is based on the use of one zero-current segment (tz)per line semicycle [13]. A positive feedback is used to increasethe chopping factor (cf ), which is defined as the ratio of thezero time tz to half of the period of the voltage waveformTv/2. Thus cf = 2tz/Tv with increasing the deviation ofthe frequency away from nominal. The increasing deviationis usually selected to be a linear function of the frequency ofthe voltage at the point of common coupling (PCC) which isdenoted by f . The function is as follows [14]:

cf = cf0 +K(f − fg), (1)

where cf0 is the chopping fraction when there is no frequencyerror, K is an accelerating gain, and fg is the grid frequency.The inverter angle for the SFS IDM (θSFS) can be calculatedas follows [15]:

θSFS(f) =ωtz2

=πcf(f)

2. (2)

The non-detection zone (NDZ) of the SFS IDM in the qualityfactor, Qf , versus resonant frequency, f0, space is derivedusing the phase criteria [15]:

tan−1

[Qf (

f0fis− fisf0

)

]=π

2[cf0 +K(f − f0)] . (3)

B. SMS IDM

SMS method applies positive feedback to the phase of thevoltage at the PCC as a means of shifting the phase and hencethe short-term frequency [16]. The phase angle of the currentis controlled as a function of the deviation of the frequencyof the last cycle from the nominal operating frequency of theutility grid. The phase angle of the PV inverter current, θSMS ,is calculated as follows [15]:

θSMS = θm sin

(π

2

f − fgfm − fg

), (4)

where fm is the frequency at which the maximum phase shiftθm occurs. After grid disconnection, the frequency of theislanded system will drift from the rated value fg if [15]:

dθloaddf|f=fg <

dθSMS

df|f=fg . (5)

Thus, the following equation can be derived for the SMS activefrequency drifting [16].

θmfm − fg

≥ 12Qf

π2, (6)

where fm−fg is usually taken as 3 Hz so that stable operatingpoints lie outside the normal frequency range.

The NDZ of the SMS IDM in the Qf versus f0 space isderived using the phase criteria [17]:

tan−1

[Qf

(f0fis− fisf0

)]= θm sin

(π

2

f − fgfm − fg

). (7)

After understanding the feeder and DG model, next theprocedure used in the developed platform to calculate the NDZis presented.

V. EVALUATION OF RISK OF UNINTENTIONAL ISLANDING

Considering the PV output active and reactive power as Pand Q, respectively, and the load active and reactive power asPL and QL, respectively, Fig. 1 shows the schematic diagramof a distribution feeder. As can be seen from the figure, themismatch between generation and consumption in the feeder,shown by ∆P + j∆Q, is compensated by the grid.

PV

RLC

P jQ

L LP jQ

Load

P j Q PCC

Utility

breaker Grid

Fig. 1. Schematic diagram of the distribution feeder.

The frequency at PCC can be calculated as follows [17]:

f =1

2π√LC

√( QL

QfPL

)2

+ 4− QL

QfPL

. (8)

The behavior of the system at the time of utility disconnec-tion will depend on ∆P and ∆Q just before the disconnection.If ∆P = ∆Q = 0 when the utility is disconnected, there mightnot be sufficient change in the voltage amplitude or frequencyat PCC to activate local IDMs. Calculation of NDZ in termsof ∆P and ∆Q might not provide a transparent view of thefeeder laoding condition. Therefore, the NDZ is defined basedon the load fraction (LF ) and power factor (PF ) of the feeder.LF denotes the fraction of the feeder’s load over the feeder’speak load.

Based on the formed island, its load, and interconnectedDGs, one can find the balance point at which the active/reactivepower generation matches the active/reactive power consump-tion. This balance point is denoted by LF ∗ and PF ∗ andcalculated as follows:

LF ∗ =

√P 2 + (Q+Qcap)2

P 2L +Q2

L

, (9)

PF ∗ =P√

P 2 + (Q+Qcap)2, (10)

where Qcap stands for reactive power injection by capacitorbanks. Notice that in the above equations, P and Q representthe aggregated DG plants’ generation capacity in the case ofseveral interconnected DGs.

Fig. 2 shows a simple presentation of the developed plat-form. After generating the feeder model and calculating the

balance point (LF ∗, PF ∗), a batch-mode coarse resolutionsweep is run over the expected range of LF and PF (theloop in the figure). For all pairs of LF and PF in the batch, asimulation is run in which an island is formed due to the loss-of-main grid and the resulting run-on time of all DG plantsin the island is recorded. Notice that the run-on time of a DGplant is defined as the time lag between occurrence of theloss-of-main grid and the DG plant disconnection caused byits islanding protection. The NDZ is defined as the range ofloads over which the run-on times of the DG plant are longerthan the IEEE 1547 limit of 2 sec. Once any NDZ is located,batches utilizing finer resolution are run to determine the peakrun-on time values and refine the prediction of the shape of theNDZ in the LF -PF plane. Once the NDZ is known, utilityengineers can decide whether the NDZ is such that the riskof islanding is negligible or it represents a realistic feeder’sloading scenario and additional mitigation is needed.

Generate Feeder Model Using the Feeder Raw DataGenerate Feeder Model Using the Feeder Raw Data

LF=LF* and PF=PF*LF=LF* and PF=PF*

Update the feeder load by LF and PF Update the feeder load by LF and PF

Apply loss-of-main grid and calculate the run-on time of all DGsApply loss-of-main grid and calculate the run-on time of all DGs

Identify the NDZ for the island and considered LF and PFIdentify the NDZ for the island and considered LF and PF

Change LF and PF to analyze a new feeder’s loading conditionChange LF and PF to analyze a new feeder’s loading condition

Fig. 2. Flowchart of developed analytical tool for islanding study.

VI. RESULTS AND DISCUSSION

The developed analytical platform was tested on an actualcase study. The case study represents a distribution feederin National Grid USA operating region in Northeastern U.S.The feeder is a four-wire multi-grounded neutral overheaddistribution feeder operated at 13.2 kV, as shown in Fig. 3.The voltage levels for step down transformers in the figureare in kV. The feeder contains 1438 nodes, 1437 branches, 2fixed shunt capacitor banks, and 4 transformers. The feedersmeasured peak daytime load during the past twelve monthswas approximately 9.469 MVA, and the daytime minimumload was measured to be approximately 3.0 MVA, which isslightly less than 1/3 of the peak load. Two PV plants, PVplant 1 and PV plant 2, were connected to the feeder. PVplant 1 is composed of 6 inverter modules, each with 500kW capacity. The total capacity of PV plant 1 is 3 MW. PVplant 2 is composed of 4 inverter modules, each with 500 kWcapacity. The total capacity of PV plant 2 is 2 MW. Both PVplants are connected to the feeder with a step up 0.32 kV/13.2kV transformer. Table I lists the set-point values for UFP/OFPand UVP/OVP of the PV plants. It is assumed that both PVplants are equipped by SFS IDM.

The feeder raw data used for simulations was acquired fromthe existing feeder’s CYME file. The data was then processedand the accurate model of the feeder with all components wasgenerated in the developed platform. MATLAB/SIMULINKwas utilized as the main platform.

Fig. 4 shows the run-on time over a range of LF andPF . As can be seen from the figure, the maximum run-on

Substation

200 kVAr/phase

Cap. Bank

400 kVAr/phase

Cap. Bank

Step Down

Transformer

13.2/4.8

Step Down

Trans.

13.2/4.8

Step Down

Trans.

13.2/4.8

Step Down

Trans.

13.2/4.8

PV Plant 1

PV Plant 2

Fig. 3. Single line diagram of the distribution feeder.

TABLE I. VOLTAGE AND FREQUENCY TRIP SET-POINTS OF THE PVPLANTS IN THE CASE STUDY.

Element Pickup Range Time DelayUnder Voltage 0.5 pu 160 msUnder Voltage 0.88 pu 2 sOver Voltage 1.1 pu 1 sOver Voltage 1.2 pu 160 ms

Under Frequency 57 Hz 160 msOver Frequency 60.5 Hz 160 ms

time is 1.8 second. The range of LF and PF in which themaximum run-on time occurs are 0.57 to 0.61 and 0.968 to0.972, respectively. The LF range for 1.8 second run-on timeis higher than the feeder minimum LF and the PF range for1.8 second run-on time is a realistic range to happen. Thereforethe risk of experiencing 1.8 second run-on time upon the loss-of-main grid is high. Since 1.8 second run-on time is tooclose to 2 second requirement, a conservative approach mightrequire additional protection in order to avoid any possibilityof unintentional islanding.

Fig. 4. Run-on time (in Second) of the PV plants.

To illustrate the behavior of each PV plants after the loss-of-main grid, an operating point of LF=0.6 and PF=0.9802 isconsidered. Fig. 5 demonstrates the frequency at PCC of PVplants for considered LF and PF . It is worth emphasizingthat the rate of change of frequency at the PCC after the loss-of-main grid is lower when SFS IDM is not applied. In theabsence of an active IDM, for the considered LF and PF , both

PV plants will remain energizing for longer than 2 seconds.

after the loss-of-main grid

Fig. 5. Frequency AT PCC of PV plants for LF=0.6 and PF=0.9802.

Fig. 6 shows the voltages at PCC of PV plant 2, afteroccurrence of the loss-of-main grid. Based on the figure andthe set-points for UVP and OVP, it can be inferred that thereason for tripping the PV plants is the frequency.

after the loss-of-main grid

Fig. 6. Three-phase voltage (phase A with blue, phase B with green, andphase C with red) at PCC of PV PLANT 2 with LF=0.6 and PF=0.9802.

From the simulation results, it can be observed that thedeveloped analytical tool provides a platform for evaluatingthe risk of unintentional islanding of interconnected DGs. Theplatform incorporated the feeder raw data from the commonly-used utilities’ software applications and generated the accuratemodel of the feeder. Run-on time of each DG was then calcu-lated for different loading conditions of the feeder. Calculatedrun-on time is then used by the utilities’ engineers to decideon additional protection in order to avoid any possibility ofunintentional islanding.

VII. CONCLUSION

An analytical platform was presented in this paper whichcan provide an analytical tool for utilities to evaluate the riskof unintentional islanding of interconnected DGs. The platformutilizes the feeder raw data from utilities’ commonly-usedsoftware applications and generates an accurate model of thefeeders in an automated fashion. The platform then calculatesthe run-on time of each interconnected DGs after the loss-of-main grid under different feeder’s operating conditions. Basedon the calculated run-on time, the platform then finds possiblefeeder’s loading conditions under which the DGs may violatethe IEEE 1547 requirement. Results can then be used by theutilities’ engineers to determine either the risk of islanding isnegligible or it represents a realistic feeder’s loading scenario

and additional mitigation is needed. The platform was used ina real case study and results indicate the simplicity, efficiency,and accuracy of the platform.

ACKNOWLEDGMENT

The authors would like to thank National Grid USA fortheir support.

REFERENCES

[1] M. A. Moffet and D. Beauvais, ”Review of open source code powergrid simulation tools for long-term parametric simulations”, NaturalResources Canada, Tech. Report 2011-137, 2011.

[2] California Energy Commission, Electric distribution system simulationand analysis tools, status and research gap for integration of renewablesand electric vehicles, Report CEC-500-10-055, 2013.

[3] K. Mahmud and G. E. Town, ”A review of computer tools for modelingelectric vehicle energy requiremnets and their impact on power distribu-tion networks”, Applied Energy, vol. 172, pp. 337–359, 2016.

[4] L. A. C. Lopes and Z. Yongzheng, ”Islanding detection assessment ofmulti-inverter systems with active frequency drifting methods,” IEEETransactions on Power Systems, vol. 23, no. 1, pp. 480486, 2008.

[5] W. Xu, K. Mauch, and S. Martel, ”An assessment of distributed gen-eration islanding detection methods and issues for Canada,” CANMETEnergy Technology Center, Varennes, Nat. Resources Canada, TechnicalReport 2004074, 2004.

[6] D. Salles, W. Freitas, J. C. M. Vieira, and W. Xu, ”Non-detection index ofanti-islanding passive protection of synchronous distributed generators,”IEEE Transactions on Power Delivery, vol. 27, no. 3, pp. 15091518,2012.

[7] M. Ropp and A. Ellis, ”Suggested guidelines for assessment ofDG unintentional islanding risk,” Sandia National Laboratory, ReportSAND2012-1365, 2012.

[8] M. Nikkhah Mojdehi and J. Viglione, ”Automated conversion fromsteady-state to transient-scale models of distribution systems with DERsfor protection analysis,” Submitted to IEEE Power and Energy SocietyGeneral Meeting, 2017.

[9] D. J. A. Welsh and M. B. Powell, “An upper bound for the chromaticnumber of a graph and its application to timetabling problems,” TheComputer Journal, vol. 10, no. 1, pp. 8586, 1967.

[10] A. Yazdani, A. Rita Di Fazio, H. Ghoddami, M. Russo M. Kazerani, J.Jatskevich, K. Strunz, S. Leva, and J. Martinez, ”Modeling guidelines anda benchmark for power system simulation studies of three-phase singlestage photovoltaic systems,” IEEE Transactions on Power Delivery, vol.26, no. 2, pp. 12471264, 2011.

[11] A. Yazdani and R. Iravani, ”Voltage-sourced converters in powersystems,” John Wiley & Sons, Inc, 2010.

[12] A. M. Massoud, K. H. Ahmed, S. J. Finney, and B. W. Williams,”Harmonic distortion-based island detection technique for inverter-baseddistributed generation,” IET Renewable Power Generation, vol. 3, no. 4,pp. 972973, 2009.

[13] W. Bower and M. Ropp, ”Evaluation Of Islanding Detection MethodsFor Photovoltaic Utility-Interactive Power Systems,” Rep. IEA-PVPS T5-09:2002, Mar. 2002.

[14] E. J. Estebanex, V. M. Moreno, A. Pigazo, M. Liserre, and A.Dell’Aquila, ”Performance evaluation of active islanding-detection al-gorithms in distributed-generation photovoltaic systems: two inverterscase,” IEEE Transactions on Industrial Electronics, vol. 58, no. 4, pp.1185-1193, 2011.

[15] L. A. C. Lopez and H. Sun, ”Performance assessment of activefrequency drifting islanding detection methods,” IEEE Transactions onEnergy Conversion, vol. 21, no. 1, pp. 171–180, 2006.

[16] P. Du, Z. Ye, E. E. Aponte, J. K. Nelson, and L. Fan, ”Positive-feedback-based active anti-islanding schemes for inverter-based distributed gener-ators: basic principle, design guideline and performance analysis,” IEEETransactions on Power Electronics, vol. 25, no. 12, pp. 29412948, 2010.

[17] S. Wang, Q. Wei, and S. Cai, ”Microgrid islanding detection methodbased on frequency-reactive power feedback,” International Journal ofGrid and Distributed Computing, vol. 6, no. 4, pp. 107116, 2013.