Effective Use of Film Capacitors in Single-PhasePV-inverters by Active Power Decoupling

Fritz Schimpf and Lars NorumNorwegian University of Science and Technology (NTNU), Department of Electrical Power Engineering

O.S. Bragstads Plass 2E, 7491 Trondheim, NorwayEmail: Fritz.Schimpf@elkraft.NTNU.no, Lars.Norum@elkraft.NTNU.no

Abstract—The lifetime and reliability of PV-inverters can beincreased by replacing electrolytic capacitors by film-capacitors.Film-capacitors have a lower capacitance per volume ratio;therefore a direct replacement leads to very large and expensivesolutions, especially for single-phase applications.

This paper presents an active circuit which acts as an interfacebetween the DC-link of a PV-inverter and an additional storagecapacitor. The voltage ripple in the storage capacitor can beincreased compared to the DC-link capacitor, allowing a moreefficient use of the stored energy and thus a massive reductionof the overall installed capacitance.

An especially promising application can be found in module-integrated PV-inverters, because here the most efficient andcheapest topologies suffer from big electrolytic capacitors whichdeteriorate the lifetime.

The paper focuses on different possible control schemes ofthe decoupling circuit. Results from simulations are used fordiscussing the proposed control methods. Also results froman experimental efficiency comparison between systems withelectrolytic and film-capacitors are given. Finally a lab-prototypeof the decoupling-circuit is presented which will be used forfurther experiments.

I. INTRODUCTION

It is a well known problem that in single-phase invertersthe DC-link-capacitance needs to be relatively large (typicallyca. 0,5 mF per kW of output power) to decouple input andoutput of the inverter. The DC-source at the input (i.e. thePV-generator) delivers a constant power, while the AC-outputleads to a fluctuating power with double grid frequency. TheDC-link-capacitor acts as a buffer and delivers or receives thedifference in instantaneous power.

Normally electrolytic capacitors are used in the DC-linkbecause of their good capacitance per volume ratio and lowprice. By using electrolytic capacitors it is easy and affordableto install very high capacitances for decoupling betweeninput and output of the inverter. This comes with a price:A severe drawback of this type of capacitor is a limitedlifetime. Electrolytics are affected by ageing effects more thanother electronic components and are therefore a bottleneckfor inverter reliability and lifetime. Over long time the liquidelectrolyte evaporates through the rubber seals of the capacitor,degrading the capacitance. The effect can be compensated byoversizing the capacitors by design, but a limit in lifetime willstill exist.

Many efforts have been made to replace electrolytic capac-itors by film capacitors (metalized polyester or polypropylene

films) because this type has a much higher lifetime and caneven be self-healing in case of minor isolation breakdowns.The disadvantage of film-capacitors is a low capacitanceper volume ratio (approximately 20 times lower than forelectrolytics) and a much higher price. A direct replacementis therefore not feasible in terms of cost and size.

A change from single-phase to three-phase topologies is asimple way to reduce the required capacitance for the DC-link, because for a given input power the output power willbe constant. Therefore the decoupling capacitor can be muchsmaller than in the single-phase case. But for low-powerPV-inverters, especially module-integrated inverters for AC-modules this would be a costly solution. Additional currentsensors, power semiconductors and increased overall com-plexity could annihilate the lifetime advantage of the filmcapacitor. Therefore single-phase topologies are advantageousin the power range of several hundred watts up to some kW.The problem to be solved is to reduce the required capacitancewhile still having sufficient power decoupling between inputand output.

In PV-applications the voltage ripple at the inverter inputhas to be kept small in order to assure stable operationin the maximum power point of the PV-modules. This isespecially critical in single stage inverters, there the DC-linkis connected directly to the input from the PV-panel(s) anda large DC-link capacitance is required. It cannot be reducedwithout compromising the overall efficiency by de-stabilizingthe operating point.

Several solutions have been proposed to solve the problemof large capacitances in single-phase PV-inverters. Some ofthem are based on known topologies like flyback or push-pullconverters with additional switches and storage elements toactively reduce the voltage ripple at the input. Examples aredescribed in [1] (flyback converter with an additional powerdecoupling circuit) and [2] (push-pull type with additionaldecoupling paths).

Another interesting solution is to use the DC/DC stage ofa two-stage inverter for decoupling between the input and theDC-link between the two stages. This is described in [3].

A general problem of low power inverters is their lowefficiency compared to inverters at higher power levels. Tomake them competitive, high efficiency is the key factor inaddition to acceptable lifetime and reliability. The highestefficiencies in the PV-market are reached with transformerless

~

I PV I load

DC

AC

I inverter

1S

Idecouple

CDCl

2S

D 1

D 1

CstoreU

storecircuit

Decoupling−

Fig. 1. Concept and topology for parallel decoupling

single-stage inverters. In addition to their very good efficiencythese inverters are simple, light and relatively cheap.

If the capacitance requirement of single-stage single-phaseinverters can be reduced to allow the use of film capacitorsfor increasing the lifetime, they could get an ideal solution forPV-systems in the lower power range, too. Especially for high-voltage PV-modules where a lot of PV-cells are connected inseries in one module their use would become very reasonable.

A very general solution for decreasing the capacitancewhich is relatively independent of the inverter topology is theparallel active filter presented in [4] and [5]. Figure 1 showsthe circuit, which will also be the circuit being considered inthis paper.

The main principle of the circuit is that the DC-link-capacitor is separated into two parts, both with relatively lowcapacity. The capacitors are connected via a bi-directionalDC/DC-converter, allowing a different voltage at both of them.The DC/DC-converter is operated in a way which keeps thevoltage at the DC-link constant, while the voltage of thesecond capacitor Cstore can have a high ripple. This allowsusing a larger part of the stored energy in the storage capacitor.

The DC/DC-converter is bidirectional. When switch S1 infigure 1 is operated, the circuit becomes a buck converter anddelivers energy from the storage capacitor Cstore to the DC-link CDCl. In this case D2 is used for freewheeling. WhenS2 is operated, the circuit operates as a boost converter, andcharges Cstore via the diode D1. The voltage at the storagecapacitor will always be higher or equal the voltage at theDC-link.

In [4] the decoupling circuit is operated as a controlledcurrent source, achieving good results in eliminating the ripplecurrent in the DC-link. But when only controlled by a currentcontroller, the voltage in the storage capacitor Cstore can dropto the DC-link voltage while it is being discharged. Then themarginal PWM operation occurs in the buck operation of thecircuit, leading to instability and high ripple in the DC-linkvoltage. [5] presents a possible solution for the problem byrecharging the capacitor via a small transformer and a dioderectifier from the grid side. Another solution is to use more

sophisticated control method which also provides additionaladvantages as presented in the next section.

II. CONTROL OF THE ACTIVE POWER DECOUPLING

The control of the decoupling circuit shown in figure 1 iscritical, because if the DC-link capacitor is reduced to smallvalues, the voltage has to be stabilized fast and effectively.Otherwise a huge ripple will occur; or worse, the DC-link-voltage will drop below the peak value of the grid voltageand deform the grid current.

To compare different control strategies, simulations in Mat-lab/Simulink with the additional software package PLECSwere performed. The first control strategy, current feed-forward, is taken from the description in [4]. A secondsimulation was done for the configuration with an additionalrecharge circuit like described in [5]. Then two new conceptsare evaluated, leading to a ”virtual capacitance”-control, whichoperates stable without any input of reference values. Thatmeans that the DC/DC-converter and the storage capacitor canbe combined to a module which behaves like a capacitor withmuch higher capacitance than actually installed.

In all simulations the PV-generator is operating close to itsMPP with a power of 2.7 kW. The short circuit current is 8 Aand the open circuit voltage 400 V. The generator is connectedto a DC-link capacitor from which a typical load currentof a following inverter stage (Iinverter = Ipeak| sin(ωt)| issimulated by a controlled current source.

A. Conventional DC-link

For comparison an inverter with a conventional DC-link issimulated first. The capacitance of the DC-link is 1.34 mFwhich corresponds to 0.5 mF/kW. A block diagram is shownin figure 2. The load current Iload, the current delivered fromthe PV-generator IPV and the capacitor current Icap are shownin figure 3. Also the DC-link-voltage is plotted; it has a rippleof 12 Vpp at MPP-operation of the inverter.

From the results it can be seen that Icap = IPV − Iload.Later on the active decoupling circuit will deliver this currentto relieve the DC-link capacitor from the low frequency load-current.

~

I I

U

I

PV load

cap

DClink

DC

AC

Fig. 2. Block-diagram for simulation with conventional DC-link

B. Current-feed-forward

Now the DC-link capacitance is reduced to 50 µF. Inaddition the decoupling circuit is connected with a storagecapacitor which is also 50 µF. That means the overall installedcapacitance is reduced from 1.34 mF to 100 µF.

Fig. 3. Results for conventional DC-link

The decoupling circuit has a PI-controller for currentcontrol. The reference value for the decoupling current iscalculated from the average PV-current and load-current(Idecouple = IPV − Iload).

The result is shown in figure 4. The voltage ripple at the DC-link is very low, approximately 3 V. The ripple at the storagecapacitor is much higher, around 200 V. That means that thedecoupling is working effectively. A problem is also visible:Since the average voltage of the storage capacitor is not con-trolled, it changes depending on the working conditions of theinverter. In the simulated case it is slowly decreasing. When itdrops below the DC-link-voltage, a further stabilization of theDC-link will become impossible. Also the voltage rating ofthe installed capacitor should not be exceeded. Therefore anadditional control loop for the voltage of the storage capacitoris needed.

Fig. 4. Results active decoupling with current feed-forward

C. Current-feed-forward with additional recharge circuit[5] proposes a concept for keeping the voltage at the

storage capacitor above the DC-link voltage: A combinationof a transformer and a diode rectifier recharges the storagecapacitor from the grid, when its voltage drops below a definedlimit. The circuit is shown in figure 5.

~

Igrid

I PV I load

DC

AC

I inverter

Idecouple

UDClinkCDCl

1S

2S

Cstore

recharging circuit

Fig. 5. Decoupling with recharging circuit

This circuit is simulated with the current-controller fromthe previous simulation. The turns-ratio of the transformeris chosen to deliver a peak voltage of 450 V. When thevoltage of the storage capacitor drops below that value, therecharging will start. The results in figure 6 show that thevoltage stabilizes above the DC-link-voltage, allowing properoperation of the decoupling circuit. But since the rechargingcircuit has a passive diode rectifier it genaretes a current peakwhen the grid voltage reaches its peak values. So the gridcurrent as the sum of the inverter current and rechargingcurrent contains harmonics whenever the recharging takesplace. Actually, the recharging circuit seems to move theproblem of power decoupling from the decoupling circuitto the grid. The distortions in the grid current are a directconsequence of missing energy in the inverter. The inverter isfeeding a reduced power to the grid whenever it has to deliverthe peak voltage, thus delivering a non sinusoidal current.

D. Virtual capacitance methodTo solve the problem of the uncontrolled voltage at the

storage capacitor, the control structure shown in figure 7 wasdeveloped. The idea is to make the DC/DC-converter and thestorage capacitor behave similarly to a conventional capacitorwith a much larger capacitance. The current controller isunchanged and used as the inner control loop. The referencecurrent is generated by an outer control loop which controlsthe voltage in the DC-link. Using these two loops it is possibleto keep the DC-link voltage at a given reference. To get controlof the voltage at the storage capacitor, a characteristic is usedwhich couples the DC-link voltage to the storage voltage.When the storage capacitor is fully charged the reference forthe DC-link voltage is set to a high value, when the storage isnearly empty (voltage close to DC-link-voltage) the referenceis set to a low value.

Fig. 6. Results with recharging circuit

PI PIDC/DC

+PWM

DC−link

I dec DClU

storeU DCl,refU

dec,refI

Fig. 7. Block-diagram: Virtual capacitance controller

An example for the storage-voltage to DC-link-voltagecharacteristic is shown in figure 8. In the normal operatingregion of the storage capacitor (e.g. 450..800 V it is very flat,i.e. the DC-link voltage does not change much. But whenthese limits are exceeded, the DC-link voltage is increasedor decreased more drasticly. This forces a balancing of thestorage voltage.

The proposed solution has several advantages: 1.) No com-munication is needed between the control of the inverter andthe decoupling circuit. That means the the active decouplingcan be added to existing concepts without changes in thecontrol. 2.) The storage voltage is balanced in a soft way,meaning that there is no sudden marginal PWM operation. 3.)The recharging circuit is not necessary.

To show the advantage of the control structure, first asimulation without the voltage characteristic was done forcomparison. The reference for the DC-link-voltage is at a fixedsetpoint. Figure 9 shows the result. The storage voltage isdrifting down slowly, at some point dropping to the DC-linkvoltage. Marginal PWM occurs and the controller gets instable.

Figure 10 shows the same scenario with additional voltagecharacteristic. The slow drop of the storage voltage is stoppedby reducing the decoupling current. The voltage at the DC-linkis constant in the beginning, but is reduced during the timeinstances when Ustore is leaving the aimed working range.Normally, the control of the inverter would then reduce thecurrent fed to the grid in order to stabilize the DC-link-voltage.(This was not included in the simulation.)

In the simulated example the characteristic gets very steepwhen the storage voltage drops below 450 V. If a smoother

Fig. 8. Voltage characteristic

Fig. 9. Controlled DC-link-voltage

behavior of the decoupling circuit is wanted, this could beadjusted.

Fig. 10. Controlled DC-link-voltage with characteristic

III. EXPERIMENTS

A. ”Dumb” replacement of electrolytics by film-capacitorsAs a first experiment, a commercial PV-inverter is used

(SMA Sunny Boy 4000TL) and the electrolyitc capacitorsare replaced by a film capacitor of approximately the samecapacitance. This experiment is purely academic, because therequired film capacitors have a high volume and price. Any-way, it will deliver valuable information about the losses in theelectrolytic capacitors. It is expected that the overall efficiencywill be higher when film capacitors are used, because theyhave a lower ESR and smaller leakage currents compared toelectrolytic caps.

A picture of the setup is shown in figure 11. The originalcapacitors are removed except for two which are required togenerate a middle point potential. The film-capacitor (2380 µF,800 V) is placed next to the inverter and connected by copperbusbars and wires. The efficiencies of the modified and theoriginal inverter are compared in figure 12. The thick graphscorrespond to the original inverter as a reference and the thinones to the modified setup. It is visible that the efficiency is,as expected, slightly higher when filmcapacitors are used. Thedifference varies between 0.1 and 0.2 percent.

Fig. 11. Photo of Sunny Boy inverter with film capacitor

B. Prototype of active decoupling circuitAs a next stage a prototype of the active decoupling circuit

is buildt. It is controlled by a DSP (TI 320F28027).Figure 13 shows a block diagram of the the prototype.

The inductor current and the voltages of the storage- andDC-link-capacitors are measured and fed back to the DSP.CoolMOS-FETs from Infineon (SPW47N60C3) are used aspower switches. The power and control parts are electricallyisolated for safety and for easy connection to a PC via a USB-interface. Two film-capacitors can be placed directly on theboard and additional capacitors can be connected externally.The used inductance is 500 µH with a saturation current of20 A. A photo of the circuit board is shown in figure 14.

The described prototype will be integrated into an invertersystem for studying the simulated control structures. Also theoverall efficiency will be determined.

Fig. 12. Result of efficiency comparison

Cstore

DCLC

Gatedriverscaling

Analog

PC320F28027

DSP

shutdownOV−

interlockGate

supplyPower

USB

analog measurements

gate signals

Isolation

Fig. 13. Blockdiagram of prototype

IV. CONCLUSION

Parallel power decoupling can be used for decreasing thenecessary DC-link capacitance drastically. It is not sufficientto control the circuit as a current source. An additional controlof the voltage at the storage capacitor is important for stableoperation. One possibility for such a voltage control is the useof a voltage characteristic, leading to a behavior of the circuitlike a virtual capacitance.

The losses generated by the decoupling circuit will partlybe compensated by lower leakage and ESR in the capacitorsand (in single stage inverters) by lower voltage ripple at theinput terminals which leads to better MPP-adaption.

V. FUTURE WORK

The proposed contol concepts have to be ported to the DSPand the mentioned experiments with the prototype have to bedone.

Fig. 14. Photo of prototype circuit board

For controlling the circuit one-cycle-control like describedin [6] could be very helpful, because it can react very quicklyon changes of the voltages on both sides of the converter.Therefore it could help to increase the performance of thecontrollers. This will also be tested in the future.

REFERENCES

[1] T. Shimizu, K. Wada, and N. Nakamura, “Flyback-type single-phaseutility interactive inverter with power pulsation decoupling on the dcinput for an ac photovoltaic module system,” Power Electronics, IEEETransactions on, vol. 21, no. 5, pp. 1264 –1272, sept. 2006.

[2] F. Shinjo, K. Wada, and T. Shimizu, “A single-phase grid-connected in-verter with a power decoupling function,” in Power Electronics SpecialistsConference, 2007. PESC 2007. IEEE, june 2007, pp. 1245 –1249.

[3] J. Schonberger, “A single phase multi-string pv inverter with minimalbus capacitance,” in Power Electronics and Applications, 2009. EPE ’09.13th European Conference on, sept. 2009, pp. 1 –10.

[4] A. Kyritsis, N. Papanicolaou, and E. Tatakis, “A novel parallel activefilter for current pulsation smoothing on single stage grid-connected ac-pv modules,” in Power Electronics and Applications, 2007 EuropeanConference on, sept. 2007, pp. 1 –10.

[5] A. Kyritsis, N. Papanikolaou, and E. Tatakis, “Enhanced current pulsationsmoothing parallel active filter for single stage grid-connected ac-pvmodules,” in Power Electronics and Motion Control Conference, 2008.EPE-PEMC 2008. 13th, sept. 2008, pp. 1287 –1292.

[6] K. Smedley, “One-cycle controller for renewable energy conversionsystems,” in Industrial Electronics, 2008. IECON 2008. 34th AnnualConference of IEEE, nov. 2008, pp. 13 –16.