A Dual-Active-Bridge based Bi-DirectionalMicro-Inverter with Integrated Short-Term Li-Ion

Ultra-Capacitor Storage and Active PowerSmoothing for Modular PV Systems

Shahab Poshtkouhi1, Miad Fard1, Husam Hussein1, Lucas Marcelino Dos Santos1,Olivier Trescases1, Mihai Varlan2 and Tudor Lipan2

1University of Toronto, 10 King’s College Road, Toronto, ON, M5S 3G4, Canada2Solantro Semiconductor, Ottawa, ON, K2E 7Y1, Canada

E-mail: shahab.poshtkouhi@utoronto.ca

Abstract— This work targets modular nanogrids for remote lo-cations, where photovoltaic modules can be gradually introducedto grow the renewable energy capacity at minimal capital cost,while reducing diesel fuel consumption. Today’s grid-tied micro-inverters provide a modular solution for ac power generation innanogrids but battery storage remains centralized, requiring anadditional ac-dc converter. The main contribution of this workis a new micro-inverter platform and control scheme with bi-directional power flow between the nanogrid, the photovoltaicmodule and integrated short-term storage, using new highenergy-density Lithium-Ion Capacitor technology. A real-timepower smoothing algorithm is also proposed and the performanceof the new 100 W micro-inverter is experimentally verified undervarious closed-loop dynamic conditions.

I. INTRODUCTION

There is a growing demand for stand-alone flexible acnanogrids in areas ranging from underpopulated remote com-munities and research stations, to developing nations that lacka power infrastructure. Off-grid power systems are tradition-ally based on diesel generators; however the high cost ofmaintenance and fuel transport makes the integration of solarenergy attractive [1]. For example, [2] presents a 70 kWoff-grid photovoltaic (PV) system in Canada. Dynamic loadshedding is a key strategy for energy management in off-grid systems having minimal storage [3]. This work targetssmall-scale remote nanogrids (up to tens of kW) where PVmodules are used either as the primary energy source, oras a supplement to diesel generators. In these applications,there is a need for modular power electronics with minimalcapital cost, while allowing gradual expansion of generationcapacity. The intermittent nature of PV, and thus the need forstorage, is a major challenge in these applications, even ifpower-quality and up-time requirements are reduced comparedto conventional grids. Low-power micro-converters (MIC) [4],[5] and micro-inverters (MIV) [6]–[8] provide high-granularitydistributed Maximum Power Point Tracking (MPPT) [9]–[11]at the module or sub-string level. This leads to increased ro-bustness to clouds, dirt, and aging effects as well as irradianceand temperature gradients. Existing MIVs satisfy the need for

low capital-cost and expandable ac generation, but not storage.A typical off-grid MIV based ac power system is shown inFig. 1(a).

The Energy Storage System (ESS) is usually based on alarge centralized bi-directional ac-dc converter, connected toa battery bank or a flywheel [12], [13]. The ESS generallyincludes a sophisticated Battery Management System (BMS),with cell-level monitoring and balancing circuits. The objec-tive of this work is to (1) introduce a new bi-directional MIVarchitecture with integrated ESS, as shown in Fig. 1(b), and(2) demonstrate a PV power smoothing algorithm that takesadvantage of the integrated storage to improve power qualityand reduce stress on the generator. The new MIV hardware canincorporate either (1) short-term storage using new Lithium-Ion Capacitor (LIC) technology, for power smoothing and loadtransient mitigation, or (2) long-term storage using batteries(ac-battery). Aside from injecting real-power into the nanogrid,the MIV architecture can also provide reactive power on de-mand. The short-term storage can also be used to mitigate therelatively slow diesel generator start-up [14]. While the MIVcost is slightly increased due to the bi-directional capability,the improved system modularity has tremendous value in thetarget applications.

This paper is organized as follows. The chosen short-termstorage technology, Lithium-Ion Ultra-Capacitors (LICs), isdiscussed in Section II. The proposed MIV architecture andcontrol scheme are discussed in Section III. An averaging-based power smoothing algorithm is discussed in Section IV.Experimental results are reported in Section V, demonstratingboth the dc-dc and dc-ac stages operating in closed loop. Thetests are performed on a 100 W PV module, which is a suitablecandidate for high modular nanogrids with low incrementalcost.

II. LI-ION ULTRA-CAPACITOR AS SHORT TERM STORAGE

Electrochemical Double-Layer Capacitors (EDLC), alsoknown as ultra-capacitors (u-caps), have a symmetric inputand output specific power in the range of 0.5-25 kW/kg [15],

Vpvm

+

-

Igrid

MIV

ac

Loads

Ipvm

Generator

AC Module m

Central

ESS

Vpv1

+

-

MIV

Ipv1

Vgrid

+

-

AC Module 1

+

-

VpVV vm

+

-

MIV

IpII vm AC Module m

VpVV v1

+

-

MIV

IpII v1 AC Module 1

(a)

+Vpvm-

Bidirectional

MIV

ac

Loads

Ipvm

Generator

AC Module m

Distributed

ESS

DAB

Synchronous Buck

ConverterFull-Bridge

UnfolderBi-directional

Boost+

Vpv1-

+

VLIC-

ILIC

Ipv1

+

Vgrid,rect-

I grid

Dc-dc stage+ESS Dc-ac stage

Vgrid

+

-

AC Module 1

+

Vbus-

DAB

Synchronous Buck

ConverterFull-Bridge

Unfoff lderBi-directional

Boost+

VpVV v1

-+

VLVV ICII

-

ILII ICII

IpII v1

+

VgVV rgg id,dd rect

-

Dc-dc stage+ESS Dc-ac stage

AC Module 1

+

VbVV us

-

+VpVV vm

-

Bidirectional

MIV

IpII vmAC Module m

Distributed

ESS

(b)

Fig. 1. The diesel-PV nanogrid (a) with central ESS, and (b) distributedESS, with the proposed MIV architecture.

which is at least ten times higher then typical Lithium-Ion (Li-Ion) batteries [16]. U-caps also offer higher cycle-life, lowerEquivalent Series Resistance (ESR) and reduced susceptibilityto high depth-of-discharge. A LIC is a hybrid device thatcombines the intercalation mechanism of traditional Li-Ionbatteries with the cathode of EDLCs, as shown in Fig. 2(a)[17]. The cathode exhibits an activated carbon material, whilecharge is stored at the interface between the carbon and theelectrolyte. The anode is generally pre-doped with Lithiumions, resulting in a lower anode voltage, and a higher cellvoltage of 3.8-4 V versus 2.5-2.7 V for conventional EDLCs[17]. Unlike EDLCs, LICs have a minimum operating voltage,VLIC,min, which limits their energy density. Overall, LICshave 2-4× higher energy density than EDLCs, as shown inFig. 2(b). LICs are well suited to the distributed PV ESS,where relatively high energy for power smoothing on thetime-scale of minutes, and high cycle-life are required. Unlikebatteries, LICs are expected to match the required 20-25 yearlifespan of PV systems.

III. DUAL-ACTIVE-BRIDGE BASED MICRO-INVERTERWITH INTEGRATED STORAGE

The detailed two-stage MIV architecture composed of dc-dcand dc-ac stages is shown in Fig. 3 and Fig. 4, respectively.Two dc-dc converters interface to the PV voltage, Vpv , as

(a)

0.01

0.1

1

10

100

0.1 1 10 100 1000

Specific

Pow

er

(kW

/kg)

Specific Energy (Wh/kg)

ucap 2.5 V

ucap 2.7 V

Battery discharge

Battery charge

LIC 3.8 V

EDLC U-capsBattery:Discharge

Battery:Charge

High-power batteries

High-energybatteries

LICs: Used in this work

(b)

Fig. 2. (a) Physical structure of Lithium-Ion Capacitor [17]. (b) Ragone plotfor different storage technologies [18].

shown in Fig. 3(a). The synchronous boost connected tothe LIC is operated in duty-cycle control to regulate Vpv tothe reference voltage, Vpv,ref , which is set by the MPPTblock. The second converter is an isolated Dual-Active-Bridge(DAB) that interfaces with the high-voltage dc link, Vbus. TheDAB topology was selected based on soft-switching operationand simple phase-shift power control [19]–[21]. Both dc-dcconverters are bi-directional, such that the LIC can transferenergy to/from other elements in the nanogrid. The averagepower from Vpv to Vbus, PDAB , is

PDAB =VpvVbus

nωsLDABϕ(1− ϕ

π), (1)

where n is the transformer turns ratio, LDAB = Lext+Lleak

is the DAB inductance, ϕ is the phase-shift between the twobridges, and ωs = 2πfs, where fs is the switching frequency.

PDAB can be regulated through adjusting the phase shift,ϕ, between the corresponding primary and secondary bridges(for example c2 and c6), which are driven at a duty cycle ofD1 = 50%. As a result, ϕ is used to indirectly regulate the LICcurrent, ILIC , to the reference, ILIC,ref , for power smoothing,as described in Section IV, while MPPT is achieved by regu-lating Vpv to MPP voltage. This is obtained through adjustingthe duty cycle of the LIC synchronous boost converter, D2.This control process is illustrated in Fig. 3(b) and Fig. 4(b)

LDAB=Lext+Lleak1:n

Phase Shifted by φ

Ipv

LLIC

+

Vpv

-

+

VLIC

-

+

Vbus

-

ILIC

c1 c2

c3 c4

c5 c6

c7 c8

c9

c10

To dc-ac

stageCbusVx1

Vx2

ILDAB

Cin

(a)

ILIC

ILIC,ref

Gc1(s)PWM

Generator

c1-4

Phase-Shift

Block

c5-8φ

-

+

Vpv

Gc2(s)PWM

Generatorc9-10

-+Vpv,refMPPT

Block

Vpv

Ipv

Generated by

the power-

smoothing loop

D2

(b)

Fig. 3. (a) Architecture, and (b) simplified control diagram for the dc-dcstage.

respectively. These two inner loop compensators are designedbased on the separation of time constants; hence the bandwidthof Gc1(s) is much lower than Gc2(s) to ensure stability.

The ac-dc inverter stage consists of a soft-switching syn-chronous buck converter, followed by a low-frequency un-folder, as shown in Fig. 4(a). The high-voltage buck converterregulates the dc link voltage, Vbus, while shaping the PLL-synchronized grid current to achieve unity power factor, unlessreactive support is requested. The inductor current, ILbuck,is regulated in Boundary Conduction Mode (BCM) for soft-switching and high-efficiency [22]. The required on-time, Ton,of the high-side (low-side) switch in positive (negative) realpower flow operation is

Ton = Lbuck∆I

Vbus − Vgrid,rect, (2)

where Lbuck is the buck converter inductance, Vgrid,rect isthe rectified grid voltage, and ∆I is the current ripple ofLbuck. The simplified control diagram for the dc-ac stage withzero reactive power is shown in Fig. 4(b). Reference valuesfor Ton,ref throughout the half line-cycle are pre-calculatedand stored in a Look-Up Table (LUT), such that the averageinductor current < ILbuck >Ts is sinusoidal. The PI controllerscales Ton to adjust the power transfer, based on the Vbus

regulation loop. The average power can be positive or negativedepending on the PI controller Gc3(s) output. The power flowdirection is set accordingly by driving c11 and c12 in thecorrect order by using the sign of PI controller output, K.

Lbuck

Cbuck

Lfilter +Vgrid-

c13 c16

c15 c14To dc-dcstage

c11

c12

ILbuck

+

Vbus

-

+

Vgrid,rect

-

Lfilter

(a)+-

Vgridc15-16

c13-14Gc3(s)-+

Vbus

Vbus,ref

K

LUT Free-wheeling

counterclk

reset

+-

Ton

ILbuck

+

Ton,ref

+

+-

c11

c12 DQILbuck

t

Tgrid/2

Ton

K Mode

K > 0

K < 0

1

0

Mode

Mode

(b)

Fig. 4. (a) Architecture and (b) simplified control diagram for the dc-acstage.

IV. POWER SMOOTHING ALGORITHM

The short-term storage can be used to filter the PV module’soutput power, Ppv , which can improve the nanogrid powerquality by reducing the dP/dt factor [23], as well as thefuel economy in the diesel generator [24], [25]. In fact,taking full advantage of the distributed ESS at the systemlevel requires load monitoring, intelligent load-shedding andcoordinated MIV control, which is beyond the scope of thispaper and considered as future work. The LIC energy capacityfor each MIV is chosen based on providing/absorbing the ratedPV power for approximately five minutes, which provides asufficient buffer for startup and typical cloud variations.

The smoothing process is illustrated in Fig. 10(a). Ppv issampled every 10 s and passed to a moving averaging windowof approximately 5 minutes, based on the low-latency Hulldigital filter [26], generating Ppv,ave. At each sample point,the Least Square Estimation (LSE) method is used to generatethe polynomial s(θ) that best fits this curve, by finding theoptimization vector variable θ:

θ = argminθ

m∑n=1

(Ppv,ave(n)− s(n,θ))2, (3)

where m is the number of previous samples which are usedfor the prediction. In this case, an averaging window of 5minutes is considered, and thus m = 30, assuming that themeasurements are done every 10 s. To simplify the solution,it is generally assumed that the signal (Ppv,ave in this case)can be modelled with a polynomial, in which case the estimates(θ) can be expressed as

s(θ) = Hθ, (4)

where H is any matrix of size m × (p + 1), and p is theorder of polynomial which is used in the estimation. In thiswork, p = 8 is chosen. A higher order polynomial increasesthe prediction accuracy resulting in a smoother power curve,however the computation effort increases substantially with p.A common expression for H is as follows

Hm,p+1 =

1 1 · · · 11 21 · · · 2p

......

. . ....

1 m1 · · · mp

. (5)

It can be shown that θ = (HTH)−1HTP pv,ave, and thuss(θ) can be constructed using (4). By evaluating s(m+1,θ),Ppv,ave(m + 1) is forecasted. In order for Ppv to follow theaverage power curve, the difference power, Pdiff , must besupplied or absorbed by the LIC:

P ∗LIC,ref (m+ 1) =

{Pdiff

ηLICPdiff ≥ 0,

Pdiff .ηLIC Pdiff < 0,(6)

where ηLIC is the LIC dc-dc converter efficiency, andP ∗LIC,ref (m + 1) is the calculated LIC power reference at

sample-time m+ 1. Pdiff is defined by the following

Pdiff = s(m+ 1,θ)− Ppv,ave(m+ 1). (7)

Vpv

Ipv

Hull

Moving

Averaging

PpvLSE

Ppv,ave s(θ)

+

-

P*LIC,ref

LIC SOC

Estimation

+

SOCref

Gc4(s)

+

I*LIC,ref

I**LIC,ref

ILIC,ref

Power Smoothing Loop

SOC Management Loop

+

To the LIC

Converter

LICV

1

-

Fig. 5. Power smoothing and LIC SOC control.

The smoothing process is illustrated in Fig. 5. The SOCmanagement loop is implemented to gradually steer the LICState-of-Charge (SOC) to ≈50% by generating the offset LICcurrent command, I∗∗LIC,ref . The SOC is estimated by the LICvoltage, by considering the well-known quadratic formula forstored energy of a capacitor, E = 1

2CLICV2LIC . Maintaining

the SOC near the mid-range is essential to maximize thepotential swing. The LIC current command at sample timem+ 1, is therefore

ILIC,ref (m+ 1) =P ∗LIC,ref (m+ 1)

VLIC+ I∗∗LIC,ref (m+ 1).

(8)

TABLE IMIV PROTOTYPE SPECIFICATIONS

Parameter Value Units

Dc-dc Stage Switching Frequency, fs 156 kHz(DAB and LIC Converter)

Input Capacitance, Cin 450 µF

Bus Capacitance, Cbus 270 µF

Bus Voltage, Vbus 450 V

DAB Inductance, LDAB 4.7 µH

LIC Converter Inductance, LLIC 10 µH

Dc-ac stage Inductance, Lbuck 500 µH

Transformer Turns Ratio, n 19

V. EXPERIMENTAL RESULTS

A 100 W MIV prototype was fabricated on a custom PrintedCircuit Board (PCB). The prototype parameters are listed inTable I. A 100 W Building-Integrated PV (BIPV) module withthe specifications listed in Table II was used for testing. Thedc-dc and dc-ac stages are digitally controlled by an on-boardFPGA and 16-bit micro-controller, respectively. The DABtransformer was designed as a custom planar magnetic elementin order to reduce the weight and profile of the prototype.

A. MIV Operation

75

80

85

90

95

100

0 20 40 60 80 100

Efficiency (%

)

Power (W)

DAB Converter

Dc-ac stage

LIC Converter

Fig. 6. Measured efficiency of the DAB, LIC and dc-ac stages.

The measured efficiency of the DAB, LIC and inverterstages are shown in Fig. 6. These efficiency curves are nearlysymmetrical and thus the data is shown for the positivepower flow. The converter stages achieve a peak efficiency of95.1%, 97.2%, and 96.3%, for DAB, LIC and dc-ac stages,respectively. A new switching technique is currently underdevelopment to improve the light-load efficiency of the DABconverter. The steady-state DAB waveforms at the rated powerof 100 W are shown in Fig.7. The dynamic response of thePV voltage regulation loop for a step change in Vpv,ref andILIC,ref are shown in Fig. 8(a) and (b), respectively. In both

c5

c2

ILDAB

ᵩ Ts

DAB

buspv

L

n

VV +

DAB

buspv

L

n

VV +-

Fig. 7. Steady-state waveforms for the DAB converter (ILDAB :5 A/div).

cases, ILIC and Vpv are well regulated to their respectivereference values. The inverter dynamic response for an inputpower step is shown in Fig. 9.

B. Smoothing Algorithm Performance Evaluation

TABLE IILIC AND PV SPECIFICATIONS

Parameter Value Units

Nominal MPPT PV Voltage, Vmpp 10 V

Nominal MPPT PV Current, Impp 10 A

Total Minimum LIC Voltage, VLIC,min 4.4 V

Total Maximum LIC Voltage, VLIC,max 7.6 V

Total LIC Capacitance, CLIC 2200 F

Total LIC Energy Capacitance, ELIC 8.8 Wh

# of Series-Connected LIC Modules 2

The measured power curve of the BIPV module is shownin Fig. 10(a). The test was run on November 16, 2013, for arooftop installation at the University of Toronto. A relativelycloudy day was chosen to illustrate the power smoothingprocess. The activation of the PV module’s internal sub-stringbypass diodes at the start of the day is clearly visible, leadingto fast variations in power at sunrise. The power smoothingalgorithm depicted in Fig. 5 was separately implemented ona PC running NI LabVIEW, allowing a flexible choice offiltering parameters, while all other controls are implementedlocally in the micro-controller and FPGA. Two 4.4 Wh LICsare connected in series, resulting in a voltage range of 4.4 V≤VLIC ≤7.6 V. The specifications for the PV and LICs used inthis experiment are listed in Table II. The net generated power,s(θ), is approximately 7× smoother than Ppv , as shown inFig. 10(b). The LIC voltage swing throughout the cloudy dayis shown in Fig. 10(c), which demonstrates the proper sizing ofthe storage module for the considered averaging period. Theslow SOC regulation loop correctly steers the LIC voltagetowards 6.2 V throughout the day. The level of smoothingcan be optimized by tuning the filter parameters to utilize awider range of the LIC’s capacity. There is clearly a trade-off

Vpv

Ipv

ILIC

11 V

9 V

10 A

6.2A

-5.6 A

3.8 ms 2.2 ms

(a)

ILIC = -4.6 A

Ipv

Vpv 11.2 V

4 A

ILIC = 4.4 A

16 ms 27 ms

(b)

Fig. 8. Step changes in (a) Vpv,ref , and (b) ILIC,ref (ILIC :2 A/div).

between the amount of power smoothing and the energy yieldof MIV, based on the LIC converter efficiency.

VI. CONCLUSIONS

A new bi-directional MIV architecture and simple controlscheme were developed for modular low-cost nanogrid ap-plications. The MIV architecture introduces a new integratedshort-term storage solution using LIC technology as opposedto a conventional central storage solution. Furthermore, a lag-free averaging scheme is introduced which decreases the realpower variations by up to 7× on a typical cloudy day inToronto by utilizing the short-term storage. This can poten-tially benefit the stability of the nanogrid and saves on fuel

Vgrid

Igrid

678 V

0.22 A 0.6 A

Fig. 9. MIV dynamic response at unity power factor with two consecutivePpv steps, 40 W → 16 W → 40 W (Igrid:0.2 A/div).

costs as well as improving generator life-time and reducingmaintenance costs. Stable MIV operation was achieved byincorporating a dual loop approach in the dc-dc stage. TheDAB indirectly regulates the LIC’s current, while MPPT isimplemented by the LIC dc-dc converter. The four-quadrantdc-ac stage operates in BCM current control which ensures ahigh efficiency of up to 96.3%, and can be programmed toprovide reactive power. Further work is required to improvethe light-load efficiency of the DAB. The trade-off betweenthe degree of power smoothing and the nanogrid efficiencyalso needs to be further explored at the system level.

ACKNOWLEDGEMENT

This work was supported by Solantro Semiconductor, theOntario Centres of Excellence, the Natural Sciences and En-gineering Research Council of Canada, Auto21, the CanadianFoundation for Innovation and the Ontario Research Fund. Theauthors also thank Ray Orr and Ben Bacque for discussionsrelated to nanogrids and micro-inverters.

REFERENCES

[1] R. Wies, R. Johnson, A. Agarwal, and T. Chubb, “Economic analysisand environmental impacts of a pv with diesel-battery system for remotevillages,” in IEEE Power Engineering Society General Meeting, vol. 2,2004, pp. 1898–1905.

[2] S. Pelland, D. Turcotte, G. Colgate, and A. Swingler, “Nemiah valleyphotovoltaic-diesel mini-grid: System performance and fuel saving basedon one year of monitored data,” IEEE Transactions on SustainableEnergy, vol. 3, no. 1, pp. 167–175, 2012.

[3] J. Byun, I. Hong, and S. Park, “Intelligent cloud home energy manage-ment system using household appliance priority based scheduling basedon prediction of renewable energy capability,” IEEE Transactions onConsumer Electronics, vol. 58, no. 4, pp. 1194–1201, 2012.

6 7 8 9 10 11 12 13 14 15 16 17 18−20

−10

0

10

20

30

40

50

60

70

80

90

100

t

Pow

er

(W)

Ppv

s(θ)

Ppv,ave

PLIC

(hour)

(a)

6 7 8 9 10 11 12 13 14 15 16 17 18−4−3−2−1012345678

t

dP/dt(W/s)

dPpv/dt

ds(θ)/dt

(hour)

(b)

6 7 8 9 10 11 12 13 14 15 16 17 185.25.35.45.55.65.75.85.9

66.16.26.36.46.56.66.7

t

VL

IC (

V)

(hour)

SOC = 50%

(c)

Fig. 10. (a) Ppv , the running average, Ppv,ave, and the fitted polynomial,s(θ), and (b) dPpv

dt, ds(θ)

dt, and (c) VLIC for a typical cloudy day.

[4] R. K. Hester, C. Thornton, S. Dhople, Z. Zhao, N. Sridhar, and D. Free-man, “High efficiency wide load range buck/boost/bridge photovoltaicmicroconverter,” in IEEE Applied Power Electronics Conference andExposition, 2011, pp. 309 –313.

[5] B. York, W. Yu, and J.-S. Lai, “An integrated boost resonant converterfor photovoltaic applications,” IEEE Transactions on Power Electronics,vol. 28, no. 3, pp. 1199–1207, 2013.

[6] R. Erickson and A. Rogers, “A microinverter for building-integratedphotovoltaics,” in IEEE Applied Power Electronics Conference andExposition, 2009, pp. 911 –917.

[7] “Directgrid dga smart series,” Island Technologies Datasheet, 2010,available http://www.directgrid.com.

[8] “Enphase m190 microninveter,” Enphase Datasheet, 2009, availablehttp://enphaseenergy.com.

[9] L. Linares, R. Erickson, S. MacAlpine, and M. Brandemuehl, “ImprovedEnergy Capture in Series String Photovoltaics via Smart DistributedPower Electronics,” in IEEE Applied Power Electronics Conference andExposition, 2009, pp. 904 –910.

[10] N. Femia, G. Lisi, G. Petrone, G. Spagnuolo, and M. Vitelli, “Distributedmaximum power point tracking of photovoltaic arrays: Novel approachand system analysis,” IEEE Transactions on Industrial Electronics,vol. 55, no. 7, pp. 2610 –2621, July 2008.

[11] S. Poshtkouhi, V. Palaniappan, M. Fard, and O. Trescases, “A generalapproach for quantifying the benefit of distributed power electronicsfor fine grained mppt in photovoltaic applications using 3-d modeling,”IEEE Transactions on Power Electronics, vol. 27, no. 11, pp. 4656–4666,2012.

[12] L. Xu and D. Chen, “Control and operation of a dc microgrid withvariable generation and energy storage,” IEEE Transactions on PowerDelivery, vol. 26, no. 4, pp. 2513–2522, 2011.

[13] G. Suvire, M. Molina, and P. Mercado, “Improving the integrationof wind power generation into ac microgrids using flywheel energystorage,” IEEE Transactions on Smart Grid, vol. 3, no. 4, pp. 1945–1954, 2012.

[14] T. Theubou, R. Wamkeue, and I. Kamwa, “Dynamic model of diesel gen-erator set for hybrid wind-diesel small grids applications,” in 25th IEEECanadian Conference on Electrical Computer Engineering (CCECE),2012, pp. 1–4.

[15] “48 V ultracapacitor specifications,” MaxwellTechnologies Inc. datasheet, 2011, available athttp://maxwell.interconnectnet.com/ultracapacitors/.

[16] “U-charge xp lithium iron magnesium phospate battery,” Valence Tech-nology datasheet, 2010, available at http://www.valence.com/energy-storage/xp-12v-19v-lithium-phosphate-battery-module.

[17] “Lithium Ion Capacitor,” JSR Micro, 2012, available at

http://www.jsrmicro.com/index.php/EnergyAndEnvironment/.[18] O. Laldin, M. Moshirvaziri, and O. Trescases, “Predictive algorithm for

optimizing power flow in hybrid ultracapacitor/battery storage systemsfor light electric vehicles,” IEEE Transactions on Power Electronics,vol. 28, no. 8, pp. 3882–3895, 2013.

[19] F. Krismer and J. Kolar, “Efficiency-optimized high-current dual activebridge converter for automotive applications,” IEEE Transactions onIndustrial Electronics, vol. 59, no. 7, pp. 2745–2760, 2012.

[20] ——, “Accurate small-signal model for the digital control of an auto-motive bidirectional dual active bridge,” IEEE Transactions on PowerElectronics, vol. 24, no. 12, pp. 2756–2768, 2009.

[21] H. Qin and J. Kimball, “Generalized average modeling of dual ac-tive bridge dc-dc converter,” IEEE Transactions on Power Electronics,vol. 27, no. 4, pp. 2078–2084, 2012.

[22] R. Erickson and D. Maksimovic, Fundamentals of Power Electronics,Second Ed. Springer, 2001.

[23] M. Datta, T. Senjyu, A. Yona, T. Funabashi, and C.-H. Kim, “Afrequency-control approach by photovoltaic generator in a pv-diesel hy-brid power system,” IEEE Transactions on Energy Conversion, vol. 26,no. 2, pp. 559–571, 2011.

[24] J.-H. Lee, S.-H. Lee, and S.-K. Sul, “Variable-speed engine generatorwith supercapacitor: Isolated power generation system and fuel effi-ciency,” IEEE Transactions on Industry Applications, vol. 45, no. 6,pp. 2130–2135, 2009.

[25] M. Datta, T. Senjyu, A. Yona, T. Funabashi, and C.-H. Kim, “Acoordinated control method for leveling pv output power fluctuationsof pv-diesel hybrid systems connected to isolated power utility,” IEEETransactions on Energy Conversion, vol. 24, no. 1, pp. 153–162, 2009.

[26] “Hull Moving Average (HMA),” Alan Hull, available athttp://www.justdata.com.au/Journals/AlanHull/.