Applied Energy 87 (2010) 3591–3605

Contents lists available at ScienceDirect

Applied Energy

journal homepage: www.elsevier .com/locate /apenergy

Analysis of active islanding detection methods for grid-connected microinvertersfor renewable energy processing

C.L. Trujillo a,b, D. Velasco a,*, E. Figueres a, G. Garcerá a

a Grupo de Sistemas Electrónicos Industriales del Departamento de Ingeniería Electrónica, Universidad Politécnica de Valencia, Camino de Vera S/N, C.P. 46022, Valencia, Spainb Departamento de Ingeniería Electrónica, Universidad Distrital Francisco José de Caldas, Carrera 7 N� 40-53 Piso 5, Bogotá, Colombia

a r t i c l e i n f o

Article history:Received 30 March 2010Received in revised form 6 May 2010Accepted 10 May 2010Available online 8 June 2010

Keywords:Distributed generationRenewable energy sourcesMicroinvertersActive islanding detection methods

0306-2619/$ - see front matter � 2010 Elsevier Ltd. Adoi:10.1016/j.apenergy.2010.05.014

* Corresponding author.E-mail addresses: cltrujillo@udistrital.edu.co (C.L. Tr

es (D. Velasco), efiguere@eln.upv.es (E. Figueres), ggarcURL: http://www.gsei.upv.es (D. Velasco).

a b s t r a c t

This paper presents the analysis and comparison of the main active techniques for islanding detectionused in grid-connected microinverters for power processing of renewable energy sources. These tech-niques can be classified into two classes: techniques introducing positive feedback in the control ofthe inverter and techniques based on harmonics injection. Accurate PSIMTM simulations have been car-ried out in order to perform a comparative analysis of the techniques under study and to establish theiradvantages and disadvantages according to IEEE standards.

� 2010 Elsevier Ltd. All rights reserved.

1. Introduction

In recent years, a major global priority is the development ofrenewable energy. These energy sources produce lower pollutionin terms of CO2 emissions than conventional fossil fuels. From thispoint of view the distributed generation concept takes importanceand it represents a paradigm shift from centralized power genera-tion [1,2].

Distributed generation can be defined as small-scale generatorsinstalled near the loads with the ability of interacting with the gridimporting or exporting energy [3].

Under this scheme, autonomous low power converters calledmicroinverters [4] have been developed. The microinverters havethe ability of operating both in grid connected mode by injectingenergy from renewable sources (solar energy, wind energy, fuelcells, among others) to the grid, and in islanding mode feedinglocal loads without grid connection. Besides, they can be connectedto other inverters with similar characteristics to supply a highernumber of loads, being easy to expand [5,6].

When a distributed generator (DG) is injecting power to thegrid, one feature that should be taken into account is the islandingcondition. The condition of ‘‘islanding” in DGs is an electrical phe-nomenon that occurs when the energy supplied by the power grid

ll rights reserved.

ujillo), davede@posgrado.upv.era@eln.upv.es (G. Garcerá).

is interrupted due to various factors and the DGs continue energiz-ing some or the entire load. Thus, the power grid stops controllingthis isolated part of the distribution system, which contains bothloads and generation. Therefore, islanding operation of grid-con-nected inverters may compromise security, restoration of serviceand the reliability of the equipment [7].

In the case of several DGs connected to a low-voltage powergrid, it is possible that the amount of energy generated by the dis-tributed system agrees with the amount of energy consumed bythe loads on the grid. Under this situation, there is no energy flowtowards the grid and the distributed systems may fail to detect apossible power grid disconnection, so that the DGs may continuefeeding the loads leading to an ‘‘islanding” condition. In addition,when the islanding condition happens, there is a primary securitycondition which forces the generator system to disconnect fromthe de-energized grid without taking into account the connectedloads.

The ‘‘islanding” effect in inverters may result from a failure de-tected by the grid and the consequent switch opening, accidentalopening of the electrical supply because of equipment failure, sud-den changes in the electric distribution systems and loads, inten-tional disconnection for maintenance services either on thenetwork or in the service, human error, vandalism or acts of nature.

There are many reasons why islanding should be anticipated inthe distributed generation systems connected to the grid. The mainreasons are safety, liability and maintenance of the quality of thesupplied energy.

For the above reasons, islanding detection is a mandatory fea-ture that should be taken into account in distributed generation

3592 C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605

systems. Several algorithms have been proposed in the last fewyears to solve it.

Islanding detection techniques can be divided into remote andlocal ones, and also into passive and active techniques [8]. Activetechniques resident in the inverter, which are discussed in thisarticle, introduce disturbances in the output of the inverter in orderto affect a certain parameter that comes out of range in an islan-ding situation.

There are different active methods based on positive feedbackin published literature [8–10]. Some of these methods are: varia-tion of active and reactive power [11], Sandia voltage shift (SVS)and Sandia frequency shift [12,13], slip-mode frequency shift(SMS) [14], active frequency drift (AFD) [15], and general electricfrequency schemes (GEFS) [16], among others.

The paper is organized as follows. First, the grid-connectedmicroinverter used to perform the study will be described. Second,different active resident methods will be presented. Next, thesemethods will be evaluated with RLC loads according to the stan-dards. Finally, some conclusions will be presented.

2. Description of the microinverter system

Fig. 1 shows the scheme of the microinverter that has been usedto evaluate the islanding detection algorithms. It is fed by two par-allel connected 220 W photovoltaic panels. The panels voltage is inthe range 26–36.7 V in the maximum power point (MPP), so that aDC–DC stage is needed to raise the voltage supplied by the panelsto suitable levels (400 V) for grid connection of the power process-ing system. The power stage chosen for this purpose is a push–pullDC–DC converter. This topology is a good choice for low input volt-age and medium power. Its advantage when compared to full-bridge DC–DC converters is that only two power transistors areneeded. When compared to a half-bridge DC–DC converter, thepush–pull does not need a capacitive input voltage divider withhigh RMS currents. The main drawback of the push–pull topologyis that the power transistors withstand twice the input voltage, butthat is not a problem in low input voltage applications. For DC–ACconversion a current-controlled H-bridge bipolar PWM inverter[17] with an output LCL filter [18,19] has been chosen. Note thatthe use of a push–pull topology provides galvanic isolation withoutthe need of bulky and heavy low frequency transformers.

As it can be observed from Fig. 1, the control of the push–pullconverter starts from a maximum power point tracker (MPPT),

Fig. 1. Block diagram of

implemented by a P and O (perturb and observe) algorithm[20–22]. The output of the MPPT, Vg_ref, is the reference of thePV string voltage, Vg. The error between both signals is amplifiedby a PI voltage controller whose output is the control voltage, Vc,of the peak current mode control loop (CIC) [23]. This controlvoltage limits the peak value of the current through the primaryof the push–pull transformer, avoiding its saturation. In the peakcurrent mode control PWM modulator an external stabilizationramp, Se, is added to the sensed transformer current (Sn). FM isthe PWM modulator gain. This value should be adjusted properlyin order to guarantee the stability of the current loop. Fig. 2shows the aspect of a conventional CIC loop applied to a push–pull converter.

The PWM inverter has an inner current controller based on aharmonic compensator [24] in order to comply with the IEEE929-2000 standard [25], in terms of the THD of the current injectedto the grid. The reference of the current loop is made of the sum oftwo terms: a power-feedforward term used to achieve a fast re-sponse of the inverter control with regard to changes in the powergenerated by the panels [26], and the output of the PI controller ofthe inverter DC link voltage. The resulting sum is the amplitude ofthe inverter reference current, that is synchronized with the funda-mental component of the grid voltage by means of a phase lockedloop (dqPLL) implemented using the synchronous rotating refer-ence frame technique [27,28]. Fig. 3 shows the dqPLL blockdiagram.

Table 1 summarizes the nominal values of the microinverter.Table 2 shows the expression of the chosen regulators for both

the inner current loops and the outer loop, crossover frequenciesand phase margins, for the microinverter studied.

Apparently, the islanding detection algorithms could be alsoevaluated by simplifying the power inverter by eliminating theMPPT algorithms and the dc voltage regulation loop of the inverter(in that case, a constant DC link voltage should be assumed). How-ever, those algorithms can be affected by the islanding detection,and vice versa. Therefore, they have been taken into account in thisstudy.

3. Active islanding detection methods

These techniques intentionally introduce disturbances at theoutput of the inverter in order to determine whether they affectvoltage, frequency or impedance parameters, in which case it is

the microinverter.

Fig. 2. Conventional CIC scheme of push–pull converter.

Fig. 3. Block diagram of the dqPLL.

Table 1Parameters of the microinverter under study.

Parameter Values

Panel voltage variation (Vg) 24–36.7 VInjected power by panel (Ppv) 430 WDC link voltage (VDC) 400 VInverter output voltage (VO) 230 VRMS

Inverter output frequency (fg) 50 HzPush–pull inductance (LX) 11 mHPush–pull input capacitor (CIN) 470 lFPush–pull transformer turns ratio (N) 20Push–pull switching frequency (fs) 20 kHzDC Link capacitor (CDC_LINK) 1 mFInverter inductance (L) 19.1 mHInverter output capacitor (C) 600 nFDamping resistance (Rd) 50 XGrid inductance (Lg) 1.91 mHInverter switching frequency (fsi) 20 kHz

C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605 3593

assumed that the grid is disconnected and the inverter must be iso-lated from the load.

Active techniques have the advantage of remarkably reducingor even eliminating the so-called non-detection zone (NDZ) [7],but in order to achieve their purpose they may deteriorate thequality of the grid voltage or even they may cause instability.

Active methods can be classified into: positive feedback basedmethods and harmonic injection based methods. In the followingsections, these methods will be discussed.

3.1. Active methods based on positive feedback

Active methods based on positive feedback present the particu-larity to affect, in a controlled way, the voltage, the frequency andthe remaining parameters associated with the inverter. In the casewhen there is a change that produces that the aforementionedvariables are outside the assigned threshold and the grid is not

Table 2Summary of the expressions of the chosen regulators, crossover frequencies andphase margins.

Controller Expression in thecontinuous time

Crossoverfrequencies

Phasemargins(�)

Push–pullPWM modulator

gainFM ¼ 1

ðSnþSeÞ�Ts¼ 1ð2181þ30;533Þ�50l 2.63 kHz 66.4

Voltagecontroller

kp þ kis ¼ 1:5þ 300

s98.4 Hz 89.6

InverterCurrent controller kp þ kh �Bh �s

s2þBh �sþðx2hÞ ¼ 1:04þ 100�ð2pÞ�s

s2þ2p�sþð100pÞ21.28 kHz 58.6

Voltage controller kp þ kis ¼ 2:5þ 0:04

s10 Hz 87

3594 C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605

connected, because of the positive feedback the system reacheslevels of over/under voltage (OUV) or over/under frequency(OUF) [11], so that the islanding situation may be detected.

Some islanding detection techniques based on positive feedbackare presented in the following.

3.1.1. Variation of active power and reactive powerThis method [11] involves injection of active power and reactive

power from the inverter to the grid system. In an islanding condi-tion the power flow from the inverter to the load affects currentand voltage in the PCC. The voltage variation with regard to the ac-tive power injected by the inverter to the load during an islandingcondition is obtained starting from the active power generated bythe DG in an islanding condition, following (1). R expresses theequivalent resistance of the load, PDG is the distributed generatorpower, PLOAD is the load power and V is the voltage in the pointof common coupling (PCC).

PDG ¼ PLOAD ¼V2

Rð1Þ

Differentiating PDG with regard to the voltage and expressing itin terms of power it results:

dPDG

dV¼ 2 � V

R¼ 2 �

ffiffiffiffiffiffiffiffiPDG

R

rð2Þ

Hence the voltage variation as a function of power can be ex-pressed as:

DV ¼ DPDG

2�

ffiffiffiffiffiffiffiffiR

PDG

sð3Þ

Since both R and PDG are constant, the voltage variation is di-rectly proportional to the variation of active power. Hence, in orderto bring the amplitude of the voltage outside the normal operatingrange for allowing islanding detection, a possible strategy is to varythe active power injected by the inverter.

It is necessary to choose carefully when the power is injectedbecause continuous variations of the injected power can perturbthe Maximum Power Point Tracking (MPPT) algorithms. For thesereasons, this method involves the injection of active power onlywhen the voltage measured at the PCC exceeds a certain thresholdvalue (Vs).

The time needed by the algorithm to detect a fault can be ad-justed with a Kv factor that increases o decreases dP proportionallyto the voltage variation. This value should be chosen large enoughto detect the islanding situation avoiding overcurrents which maydamage the system elements. An initial value of Kv can be obtainedfrom Eq. (2), following Kv_initial = 2�p(PDG/R). The reference for theinverter current control loop can be calculated as follows:

Iref ¼dP þ PDG

Vð4Þ

where dP = Kv(V–Vn), Vn and V being the amplitude of the grid nom-inal voltage and the measured actual value, respectively.

Similarly to the relationship between voltage and active power,a strong dependence between frequency and reactive power exists,which may be used to develop another method of islanding, basedon measuring the grid frequency. Since Std. 929-2000 recommendsDG operation near unity power factor, the generated reactivepower must be zero in normal operation. At the beginning of islan-ding, Eq. (5) remains valid, so that the frequency depends on thevalues of the inductive and capacitive components of the load.Therefore, the variation of Q with x follows Eq. (6), where x0 isthe resonance frequency of the equivalent RLC circuit, Eq. (7).

QDG ¼ QLOAD ¼ 0 ¼ V2 � 1x � L�x � C� �

ð5Þ

Q � 0x�x0

¼ dQdxjx¼x0

ð6Þ

x0 ¼1ffiffiffiffiffiffiffiffiffiL � Cp ð7Þ

Solving Eq. (6), in order to determine the variation of the fre-quency as a function of DQ and setting the frequency of the loadat the resonance frequency, it is obtained:

Dx ¼ �V2 � x20 � L

x20 � L � C þ 1

� �� DQ ð8Þ

As observed from Eq. (8), the frequency variation is directly pro-portional to the changes of reactive power. It can be obtained theexpression of Df as a function of the quality factor, Qf = R�C�x0,the resonance frequency, f0, and the active power, PDG, in an oper-ation point, following Eq. (9).

Df ¼ � f0

2 � PDG � QfDQ ð9Þ

The reference frequency for the inverter control can be calculated asfollows:

fref ¼ fg þ Kf � ðf � fgÞ ð10Þ

where fg is the nominal grid frequency, f is the measured frequencyand Kf is a factor that allows to accelerate the islanding detection. Kf

may be calculated following Eq. (6).Fig. 4 shows the whole block diagram of the method, where

both active and reactive channels are disturbed.The disadvantage of this method is that it can generate false

detections of islanding when several inverters are connected tothe same point of the grid. Moreover, instability problems may ap-pear because the inverter is continuously injecting disturbancesinto the grid.

3.1.2. Sandia voltage shift and Sandia frequency shiftThis method [12,13] uses a positive feedback loop of the point

of common coupling (PCC) voltage amplitude, being similar tothe active power variation technique. If the voltage amplitude(usually it is measured its RMS value) decreases, the inverter re-duces its output current and thus the output power. Taking into ac-count Eqs. (1)–(3) derived above, it is possible to express the activepower variation in terms of the RMS voltage at the PCC and of itsvariation as:

DPDG ¼ 2 � DV � VR

ð11Þ

The response time of the algorithm can be adjusted by a factorKv that increases or decreases the inverter current proportionallyto the voltage variation. This value should be chosen following

Fig. 4. Block diagram of P and Q injection method.

C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605 3595

the same considerations that were described in the active powervariation method. The reference for the inverter current controlloop can be calculated as follows:

Iref ¼dP þ PDG

V¼ kV � DV þ PDG

Vð12Þ

Sandia frequency shift (SFS) is based on the positive feedbackof the frequency of the PCC voltage. When the grid is connected,the method detects and tries to amplify small changes in fre-quency, but the presence of the grid avoids it. When the grid isdisconnected, the frequency changes produce a phase error. Thepositive feedback makes, in an iterative process, that this errorpushes the fundamental frequency of the inverter beyond thethresholds of OUF. When the method is implemented, the refer-ence frequency for the inverter is calculated as a function of bothits value at the previous iteration (n) and its variation Df. This isexpressed as:

fnþ1 ¼ fn þ Kf � Df ð13Þ

where fn+1 is the reference frequency for the inverter in the (n + 1)thiteration cycle, fn is the frequency in the nth cycle and Kf is a con-stant that allows to accelerate the islanding detection. Finally Dfn

is the frequency variation in each cycle. Sometimes, the resonancefrequency of the load may be within the thresholds established todetect islanding, so that islanding is not detected because of thenatural tendency of the system to approach the load resonance fre-quency. Kf is designed to push the inverter frequency outside thenon-detection limits, so that islanding is detected even in the previ-ously described situation.

In the (n + 1)th cycle the inverter injects a current with a certainfrequency. Therefore, the load introduces a sliding phase angle /corresponding to an interval of time Tps following Eq. (14).

Tps ¼/

2 � p Tnþ1 ¼/

2 � p � fnþ1ð14Þ

Kf is chosen to maintain the sign of the calculated frequencyshift, fn+1 � fn, in order to avoid the natural tendency of the systemto approach the load resonance frequency. In principle, this condi-tion implies that, for an increasing shift, the frequency in the(n + 1)th cycle must be greater than that one of the nth cycle,which yields the following condition:

fnþ1 ¼1

fn þ Kf � Dfn� /

2 � p � ðfn þ Kf � DfnÞ

� ��1

P fn ! Kf

>�/ � fn

2 � p � Dfnð15Þ

Another issue for choosing the value of Kf is the desired time re-sponse. This can be accomplished by imposing a minimum timevariation called shift time, Tf. Therefore, it is obtained:

fnþ1 ¼1

fn þ Kf � Dfn� /

fn þ Kf � Dfn

� ��1

P1fn� Tf

� ��1

ð16Þ

For frequencies close to the grid frequency, the term 2�p�fn�Tf isappreciably greater that the phase-shift angle /, whereas the term(1 � fn�Tf) is near one. With these simplifications it is obtained Eq.(17).

Kf P fn �j/j þ 2 � p � fn � Tf

ð1� fn � Tf Þ � 2 � p � Dfn! Kf P

f 2n � Tf

Dfnð17Þ

These methods are easy to be implemented. They provide agood compromise between effectiveness of islanding detection,output power quality and reduced effects of the system transientresponse.

One of the main disadvantages of this method is that it slightlyreduces the power quality at the inverter output. The other disad-vantage of SVS and SFS is that these methods may have small im-pacts on the utility system transient response. However, theseeffects can be avoided by decreasing Kv and Kf. Fig. 5 shows a blockdiagram the SVS and the SFS method.

3.1.3. General electric frequency schemes (GEFS)This method [16] injects a disturbance into the current that is

generated by the inverter and evaluates the effects on the PCC.The disturbance is added to the control signals in a synchronousreference frame (SRF), usually known as DQ frame. The activepower is proportional to the D axis component and the reactivepower is proportional to the Q axis component.

Because the microinverter is a single-phase system, the imple-mentation in a DQ reference framework is not as obvious as in athree-phase system. However, single-phase quantities can betransformed in DQ coordinates by the creation of a virtual Q axisas shown in Fig. 6.

3596 C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605

In the frequency diagram of General Electric, a bandpass filter(BPF) is used to obtain the variation of the iq component. This isdone to cancel out the signal noise at low frequencies and the DCoffset at high frequencies, caused by the positive feedback effect.The bandwidth of the filter must be chosen taking into accountthat standards suggest a time for islanding protection of approxi-mately 2 s. Therefore, a bandwidth into the range 1–10 Hz is suit-able to filter out the undesired components while respecting theresponse time suggested by the standards. When an islanding con-dition happens, the injected disturbance should be big enough todrive the system to instability, without significantly affecting theclosed loop stability when the system is connected to the grid.The final element to consider is the limiter, which allows specify-ing the maximum injected current based on two premises: themaximum overcurrent capability of the inverter and the minimumpower displacement factor that might be achieved. In this method,both the magnitude and the phase of the reference current are af-fected, as described by the following equations:

I ¼ffiffiffiffiffiffiffiffiffiffiffiffiffii2d þ i2

q

qand / ¼ tan�1 iq

id

� �ð18Þ

i�ref ¼ I � cosðhþ /Þ ð19Þ

Fig. 5. Block diagram of the S

Fig. 6. Block diagram of the GEFS metho

This method has the advantages of being easy to implement ona microcontroller, of a reduced NDZ, a low impact on power qual-ity, and a high robustness against grid disturbances. Nevertheless,injection of the disturbance signals (frequency and voltage) shouldbe as small as possible to preserve power quality.

3.2. Active methods based on harmonic injection

Active methods based on harmonic injection have the particu-larity of monitoring changes in grid impedance after the injectionof a particular harmonic or a sub-harmonic [29–31]. They arebased on the idea that when the grid is connected, its impedanceis much lower than that of the load. Therefore, the injected har-monic flows into the grid. However, when the grid is disconnected,it flows through the load and produces a voltage distortion that canbe detected. The method, as it has been implemented in [32], is ex-plained in the following.

A second-order harmonic current is injected by acting on theangle calculated by the PLL for the inverter synchronization. Thisis achieved by injecting a sinusoidal signal synchronized cycle bycycle which forces to smoothly modify the angle of the inverter

VS and the SFS method.

d applied to single-phase systems.

C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605 3597

current. A feedback signal is extracted from the PCC voltage (Vq

component) as a result of the injected harmonic signal.Fig. 7 shows a block diagram of this anti-islanding method.The injected harmonic signal, Sinj, is defined as:

Sinj ¼ k � sinð2 � hPLLÞ ð20Þ

where k is the gain used to choose the amount of disturbance nec-essary for the islanding detection, and hPLL is the angle calculated bythe PLL. As it can be observed in Eq. (20), a second harmonic hasbeen entered which will be added to the PLL phase and then scaledby the magnitude of the reference current of the system, as shownby Eq. (21).

iref ¼ Iref � cosðhPLLÞ þ k � sinð2 � hPLLÞ½ � ð21Þ

The value of k should be chosen taking into account three pre-mises. The first one is that k has to be big enough to detect an islan-ding condition. The second issue is that its value should not be sogreat to increase the THD of the injected current, beyond the limitsset by the standards. The third premise is that it should not affectin a significant way the amplitude of the reference current. In Eq.(21) it can be observed that the perturbation can be treated interms of a percentage of the reference current. In this case it wasused a perturbation with an amplitude of 1% (k = 0.01), whichmeets the above conditions.

When the inverter is connected to the grid, the consequence ofadding k�sin(2�hPLL) is observed as a feedback signal in the voltageat the PCC. The feedback signal has the double of the fundamentalfrequency and its amplitude is related to the grid impedance value.This feedback signal can be extracted from the voltage at PCC afterthe Park Transformation (from vq). After the transformation, thefeedback signal frequency will become half of the frequency, asshown below:

vq ¼ � sinðhÞ � va þ cosðhÞ � vb ð22Þ

Considering a system of orthogonal voltages of the feedbackvoltage (va�y�vb) due to the injected signal and considering a andb voltages, they can be represented as:

va ¼ cosð2 � hÞ and vb ¼ cos 2 � h� p2

� �¼ sinð2 � hÞ ð23Þ

Replacing the values of Eq. (23) in Eq. (24) and solving, it isobtained:

Fig. 7. Block diagram of harm

vq ¼ sinðhÞ � cosð2 � hÞ cosðhÞ � sinð2 � hÞ ð24Þvq ¼ sinðhÞ ð25Þ

The feedback voltage signal can be calculated after Park Trans-formation through a second order band pass filter tuned at the fun-damental frequency. After that, the absolute value of the peaks ofvq signal is detected, and the average value in a period of 5 Hzand of 50 Hz is calculated. Finally, the difference between thetwo averaged values is taken to determine whether it is inside oroutside a certain threshold value. If it is outside the threshold va-lue, it is activated a delay to establish if the conditions persists,and in case it happens, islanding condition is activated.

The main advantages of this method are: it does not affect thezero crossings of the current waveform, the disturbance amplitudeis so small that it can be injected without significantly affecting thecurrent THD, and it presents a very low NDZ. The method draw-backs are: it slightly degrades the quality of the injected powerby having a constant disturbance, and it can have trouble detectingfalse islanding during certain grid disturbances whose effectsmight match those of the injected harmonic.

4. Simulation results

This section presents an accurate simulation study of the differ-ent methods described above. These simulations were carried outby means of PSIMTM software [33] and tested on the microinverterthat was presented in Section 2, following the requirements of IEEEStd. 929-2000 (type of load and quality factor). In all cases it wasused an RLC load with a quality factor of 2.5. The values of theRLC load are: R = 120 X, L = 153 mH and C = 67 lF.

The microinverter system has been simulated at an MPP of thePV array of Vpv = 29.25 V, Ipv = 14.7 A (Ppv = 430 W). The refreshingfrequency of the MPPT algorithm is much lower that than of thecurrent and voltages loops. All the control loops of the push–pullDC–DC converter and of the inverter are operating when the islan-ding situation occurs at t = 0.5 s. The simulation results show theresponse of the most important magnitudes of the microinverterduring the islanding detection.

Fig. 8a shows the results of the islanding detection methodbased on the variation of active power through the use of voltagefeedback. The value of Kv = 15 determines the trip time. In this

onic injection method.

Fig. 8a. PCC voltage (up), RMS voltage (middle) and islanding detection (down), with active power injection.

Fig. 8b. Transient response of the most important electrical variables of both the DC–DC converter and of the inverter, presented when the active power injection method isoperating.

3598 C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605

Fig. 9a. PCC voltage (up), frequency (middle) and islanding detection (down), with reactive power injection.

Fig. 9b. Transient response of the most important electrical variables of both the DC–DC converter and of the inverter, presented when the reactive power injection method isoperating.

C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605 3599

Fig. 10a. PCC voltage (up), frequency (middle) and islanding detection (down), with SVS method.

Fig. 10b. Transient response of the most important electrical variables of both the DC–DC converter and of the inverter, presented when the SVS method is operating.

3600 C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605

Fig. 11a. PCC voltage (up), frequency (middle) and islanding detection (down), with SFS method.

Fig. 11b. Transient response of the most important electrical variables of both the DC–DC converter and of the inverter, presented when the SFS method is operating.

C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605 3601

Table 3Comparison of the different active methods for islanding detection.

Method Trip time THDi

(%)Reactivepower(VAR)

Active powervariation

0.245 s with kv = 15 3.33 1.37

Reactive powervariation

0.230 s with Kf = 4 3.30 0.68

SVS 0.231s with kv = 5.5 3.40 1.64SFS 0.228 s with Kf = 3.5 3.50 0.82GEFS 0.1–0.2 s as a function of the size of

the disturbance3.30 3.75

Impedancedetection

0.1–0.2 s as a function of the size ofthe disturbance

3.37 1.30

3602 C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605

graph it is possible to observe how the microinverter disconnectionis produced in 245 ms, complying with the time set by the stan-dards (maximum trip time = 120 cycles) [25].

Fig. 8b shows the transient response of the most importantelectrical variables of both the DC–DC converter and of theinverter:

– Ipv � current supplied by the PV array,– Vg � voltage at the PV array,– ILx � current through the output inductor of the push–pull

converter,– VDC � inverter DC link voltage,– Iinv � inverter output current supplied to the PCC.

It is observed from Fig. 8b that the islanding detection based onvariation of the active power produces transient voltages and cur-rents in the power converters that are not dangerous for the powersemiconductors and filtering components. The distortion of thecurrent injected by the inverter to the PCC, THDi, is less than3.33% at full power.

Fig. 9, corresponding to the islanding detection method basedon variation of reactive power through the use of frequency feed-back, is analogous to Fig. 8. The value of Kf = 4 determines the triptime. It is observed from Fig. 9a that the trip time is 230 ms, wellinside the maximum time defined by standards.

It is observed from Fig. 9b that the transients during islandingdetection are acceptable for the power stage. The distortion of

Fig. 12a. PCC voltage (up), frequency (middle) and

the current injected by the inverter to the PCC, THDi, is less than3.33% at full power, and the reactive power is less than 0.68VARwhen the islanding situation is presented.

Figs. 10a, 10b, 11a and 11b show the results of the SVS and SFSmethods, which have similar characteristics to those presented inthe variation of active and reactive power methods.

The values of the THDi, reactive power and trip signal for both,SVS method and SFS method are presented in Table 3.

The power injection method and the SVS method have the prob-lem that the DC link voltage level may have high and dangerousvalues for the components of the microinverter. This level dependson the size of the injected disturbance.

Fig. 12 shows the results of the GEFS method. It can be seen thatwhen islanding appears there is a variation in the Iq componentthat makes the reference current to vary in both magnitude andangle so that a feedback effect happens. This effect pushes themicroinverter frequency (or voltage) outside the established limits.

It is observed from Fig. 12b that the transients during islandingdetection are acceptable for the power stage. The distortion of thecurrent injected by the inverter to the PCC, THDi, is less than 3.3%at full power, and the reactive power is less than 3.75VAR whenthe islanding situation is presented.

Fig. 13 shows the results achieved by the impedance detectionmethod. Note that when islanding situation occurs, the componentvq changes as a result of the impedance variation at the frequencyof the injected harmonic, so that the islanding condition may bedetected.

It is observed from Fig. 13b that the transients during islandingdetection are acceptable for the power stage. The distortion of thecurrent injected by the inverter to the PCC, THDi, is less than 3.37%at full power, and the reactive power is less than 1.30VAR when theislanding situation is presented.

It is worth to point out that, for all the methods previously ana-lyzed, the THDi never exceeds 3.5% at nominal power. The methodthat has most effect about the reactive power is the GEFS methodshowever the amount reactive power produced during the islandingcondition only was 0.87% of the active power.

Moreover, the methods based on harmonic injection (GEFSmethod and impedance detection method) have a shorter trip timethan the methods based on positive feedback. Additionally thesemethods present a good transitory response similar to reactivepower variation method and SFS method.

islanding detection (down), with GEFS method.

Fig. 12b. Transient response of the most important electrical variables of both the DC–DC converter and of the inverter, presented when the GEFS method is operating.

Fig. 13a. PCC voltage and islanding detection, with impedance detection method.

C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605 3603

Fig. 13b. Transient response of the most important electrical variables of both the DC–DC converter and of the inverter, presented when the impedance detection method isoperating.

3604 C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605

Table 3 provides information to compare the trip time, THDi andreactive power of the different methods.

5. Conclusions

In this paper an analysis of several active inverter-residentmethods for islanding detection in single phase photovoltaic micr-oinverters was presented. An important novelty of this study isthat no simplifications of the power stages (DC–DC con-verter + inverter) control structures have been performed. The sim-ulation results have been obtained with the MPPT algorithm, thecontrol loops of the DC–DC converter and all the control loops ofthe inverter under operation. Therefore, the effect of the islandingdetection methods on the transient response of the electrical mag-nitudes of the power stages can be evaluated.

For the same test conditions as established by the IEEE 929 allmethods meet the detection time. However, positive feedbackbased methods have a longer trip time than those based on har-monic injection. The reason for that is that positive feedback meth-ods need a considerable amount of time to reach the UOV or UOF,whereas the methods based on harmonic injection detect varia-

tions in the impedance of the grid, which allows working withsmaller detection thresholds.

It has been observed that no detection method produces dan-gerous transient overvoltages or overcurrents in the power con-verters and in their filtering components.

Although all the methods under study are based on disturbingthe grid, potentially degrading power quality, if the design param-eters are carefully chosen, their effects on power quality are notsignificant.

Acknowledgement

This work was supported by the Spanish Ministry of Scienceand Innovation under Grant ENE2009-13998-C02-02.

References

[1] Ruan Y, Liu Q, Zhou W, Firestone R, Gao W, Watanabe T. Optimal option ofdistributed generation technologies for various commercial buildings. ApplEnergy 2009;86:1641–53.

[2] Mancarella P, Chicco G. Global and local emission impact assessment ofdistributed cogeneration systems with partial-loads models. Appl Energy2009;86:2096–106.

C.L. Trujillo et al. / Applied Energy 87 (2010) 3591–3605 3605

[3] Bakos GC. Distributed porwer generation: a case study of small scale PV powerplant in Greece. Appl Energy 2009;86:1757–66.

[4] Erickson RW, Rogers AP. A microinverter for building-integrated photovoltaics.In: Applied power electronics conference and exposition, 2009. APEC 2009.Twenty-fourth annual IEEE. 15–19 February; 2009. p. 911–7.

[5] Calais M, Myrzik J, Spooner T, Agelidis VG. Inverters for single-phase gridconnected photovoltaic systems – an overview. In: Power electronicsspecialists conference, 2002. pesc 02. 2002 IEEE 33rd annual, vol. 4; 2002. p.1995–2000.

[6] Kang F, Park S, Kim J. Photovoltaic power interface circuit incorporated with abuck-boost converter and a full-bridge inverter. Appl Energy 2005;82:266–83.

[7] Task V. Evaluation of islanding detection methods for photovoltaic utility-interactive power systems. Tech. Rep. IEAPVPS T 5-09:2002; March 2002.

[8] Mahat P, Zhe C, Bak-Jensen B. Review of islanding detection methods fordistributed generation. In: Third international conference on electric utilityderegulation and restructuring and power technologies; 2008, p. 2743–8.

[9] Francesco De M, Marco L, Antonio DA. Overview of anti-islanding algorithmsfor pv systems. Part II: active methods, In: Proceedings of EPE-PEMC; 2006. p.1884–9.

[10] Balaguer IJ, Heung-Geun K, Peng FZ, Ortiz EI. Survey of photovoltaicpower systems islanding detection methods, In: Industrial electronics, 2008.IECON 2008. 34th Annual conference of IEEE, m10-13; November 2008. p.2247–52.

[11] Francesco De M, Marco L, Antonio DA, Alberto P. Overview of anti-islandingalgorithms for PV systems. Part I: passive methods. In: Proceedings of EPE-PEMC; 2006. p 1878–83.

[12] Stevens J, Bonn R, Ginn J, Gonzalez S, Kern G. Development and testing of anapproach to anti-islanding in utility interconnected photovoltaic systems.Sandia report SAND 2000-1939; August 2000. <http://photovoltaics.sandia.gov/docs/PDF/0800steve.pdf>.

[13] Sanchis P, Marroyo L, Coloma J. Design methodology for the frequency shiftmethod of islanding prevention and analysis of its detection capability, ProgPhotovolt: Res Appl 2005;13(5):409–28. <http://dx.doi.org/10.1002/pip.613>.

[14] Lopes LAC, Sun H. Performance assessment of active frequency driftingislanding detection methods. IEEE Trans Energy Convers 2006;21(1):171–80.

[15] Ropp ME, Begovic M, Rohatgi A. Analysis and performance assessment of theactive frequency drift method of islanding prevention. IEEE Trans EnergyConvers 1999;14(3):810–6.

[16] Ye Z, Walling R, Garces L, Zhou R, Li L, Wang T. Study and development of anti-islanding control for grid-connected inverters. General Electric GlobalResearch Center, Niskayuna, New York, USA; 2004. <http://www.nrel.gov/docs/fy04osti/36243.pdf>.

[17] Mohan N, Undeland T, Robbins W. Power electronics: converters, applications,and design. John Wiley & Sons; 1995.

[18] Liserre M, Blaabjerg F, Teodorescu R, Chen Z. Power converters and control ofrenewable energy systems, ICPE 2004, Busan, Korea, Invited paper.

[19] Liang J, Green TC, Weiss G, Zhong Q. Hybrid control of multiple inverters in anisland-mode distribution system, PESC ‘03. In: IEEE 34th annual conference onpower electronics specialist, vol. 1; June 2003. p. 61–6.

[20] Esram T, Chapman PL. Comparison of photovoltaic array maximum powerpoint tracking techniques. IEEE Trans Energy Convers 2007;22(2):439–49.

[21] Kim IS. Sliding mode controller for the single-phase grid-connectedphotovoltaic system. Appl Energy 2006;83:1101–15.

[22] Gonzalez LG, Figueres E, Garcerá G, Carranza O. Maximum-power-pointtracking with reduced mechanical stress applied to wind-energy-conversionsystems. Appl Energy 2010;87(7):2304–12.

[23] Sable DM, Ridley RB. Comparison of performance of single-loop and current-injection control for PWM converters that operate in both continuous anddiscontinuous modes of operation. IEEE Trans Power Electron 1992;7(1):136–42.

[24] Kazmierkowski MP, Krishnan R, Blaabjerg F. Control in powerelectronics. London: Academic Press; 2002.

[25] IEEE recommended practice for grid interface of photovoltaic (PV) systems,IEEE Std. 929; 2000.

[26] Ciobotaru M, Teodorescu R, Blaabjerg F. Control of single-stage single-phasePV inverter. In: European conference on power electronics and applications2005; 2005. p. 10.

[27] Ciobotaru M, Teodorescu R, Blaabjerg F. A new singlephase PLL structure basedon second order generalized integrator. In: Record of IEEE PESC 2006, Jeju,Korea; 2006. p. 1511–6.

[28] Liserre M, Timbus A, Teodorescu R, Blaabjerg F. Synchronization methods forthree phase distributed power generation systems. An overview andevaluation, PESC 2005, Recife, Brasile; 2005.

[29] Zhang C, Liu W, San G, Wu W. A novel active islanding detection method ofgrid-connected photovoltaic inverters based on current-disturbing. In: Powerelectronics and motion control conference, 2006. IPEMC 2006. CES/IEEE 5thinternational, vol. 3; August 2006. p. 1–4.

[30] Ropp ME, Begovic M, Rohatgi A. Prevention of islanding in grid-connectedphotovoltaic systems. Prog Photovolt: Res Appl 1999;7(1):39–59. <http://www3.interscience.wiley.com/cgi-bin/fulltext/40007275/PDFSTART>.

[31] Timbus AV, Teodorescu R, Blaabjerg F, Borup U. Online grid measurement andENS detection for PV inverter running on highly inductive grid. IEEE PowerElectron Lett 2004;2(3):77–82.

[32] Ciobotaru M, Agelidis V, Teodorescu R. Accurate and less-disturbing activeanti-islanding method based on PLL for grid-connected PV inverters. In: Powerelectronics specialists conference, 2008. PESC 2008. IEEE; June 2008. p. 4569–76.

[33] PSIM 7.0 users guide. Powersim Inc.; 2006.