You are on page 1of 10

Energy Efficiency Analysis of Grid-Connected Photovoltaic Inverter with

Storage Battery
Chee Lim Nge1,2, Ole-Morten Midtgrd1, Lars Norum2
1
UNIVERSITY OF AGDER
4876 Grimstad, Norway
2
NORWEGIAN UNIVERSITY OF SCIENCE AND TECHNOLOGY
7491 Trondheim, Norway
E-Mail: chee.l.nge@uia.no

Acknowledgements
The research underlying this paper was supported by Research Council of Norway and Elkem Solar.
The authors are grateful for the PV outdoor measurement data shared by Georgi Hristov Yordanov.

Keywords
Photovoltaic, Energy Storage, Efficiency.

Abstract
PV is an intermittent renewable power source. Grid-connected PV inverters conventionally use grid as
the energy reservoir. As the level of PV penetration increases substantially in recent years, battery is
proposed as the energy storage in grid-connected PV system. This is to increase on-site consumption,
which in turn avoids grid voltage rise and reduces the loss in transmission line. Additionally, it allows
PV inverter to dispatch power similar to the conventional power generations. This paper analyzes the
overall efficiency of grid-connected PV inverter with storage battery using system level regression
models. The analysis postulates nonlinear relation between the total power loss and the battery
charging. Voltage droop battery charge controller that simplifies the efficiency optimization is
developed based on the regression models. The analysis and the controller are verified by the
component level power loss calculations. The controller is compared in simulation to the conventional
constant-current/constant-voltage battery charging using PV outdoor measurement data.

Introduction
Global installation of photovoltaic (PV) system has grown substantially in recent years [1]. PV is
however an intermittent power source. The conventional grid-connected system feeds power into the
grid when PV generates and the local load draws electricity from the grid when PV ceases generation.
The grid as energy reservoir is inefficient due to the loss in transmission line. This becomes an
economic drawback as the level of PV penetration increases in countries that implement PV feed-in
tariffs. The recent revision of Renewable Energy Act in Germany introduces different tariffs in order
to encourage on-site consumption [2], [3]. Battery as energy storage is proposed to increase local
consumption [4]-[6]. It also mitigates the problem of grid voltage rise due to clustered PV
installation [7]-[9]. This paper aims at analyzing the overall system efficiency of grid-connected PV
inverter with storage battery. Battery charging power that maximizes the overall efficiency is derived
based on the system level regression models. Voltage droop battery charge controller is proposed as a
simple yet effective way to optimize the efficiency. The method to calculate the component level
power losses is presented. It verifies the proposed controller and the regression models. Finally, the
droop controller is compared in simulation to the conventional constant-current/constant-voltage
charging method using PV outdoor measurement data.

Optimization of Overall System Efficiency


The topology investigated in this paper consists of three conversion stages as shown in Fig. 1. The
boost converter consists of an inductor, Lbs, a diode, Dbs and a switch, which shown here is a power
MOSFET, Mbs. The front-end boost converter is connected to the conventional 36-cell or 72-cell
crystalline silicon modules, which typically have a maximum power voltage at 18V or 35V. The
battery pack consisting of ten lithium-ion polymer cells in series is located between the two boost
converters. The first boost converter steps up the PV voltage, Vpv to the battery voltage, Vb. The second
boost converter further steps up Vb to the dc-link voltage, Vdcl, nominally 400Vdc. The power
MOSFETs, M1-M4 are the switches for the full-bridge inverter. The LCL filter, which consists of two
inductors Li, Lg and a capacitor Cf, attenuates harmonic current from the inverter. The full-bridge
inverter converts Vdcl to the 50Hz grid voltage, Vg, nominally 230Vac.
PV+

Lbs1

Ipv
+
Vpv
-

Dbs1
Ib

C1

Mbs1

Lbs2

If
+
Vb
Rb

Dbs2
Li

C2

Voc

PV-

+
Vdc C
dcl
Mbs2

M1

M3
Li

M4

Lg
Cf Lg

Ig
+
Vg
-

M2

Fig. 1: Power circuit topology


The battery is modeled in Fig. 1 as series connected internal resistance, Rb and open-circuit voltage
source, Voc. Equation (1) describes the system power flow. The output current from the front-end boost
converter branches into the battery, Ib and the feed-in path, If. The terms Ppv and Lpv denote the power
and loss functions in PV power path. The variable, Lf is the aggregate conversion losses of second
boost converter and full-bridge inverter. The variable, Lb is loss function of the battery power path.
The power in the functions of the voltages and currents depicted in Fig. 1 are given by (2).

Ppv Lpv = Pf + Lf + Pb + Lb

(1)

Pf = Vg Ig

(2a)

Pb = Voc Ib

(2b)

Ppv = Vpv I pv

(2c)

System Level Regression Models


Linear regression models at the system level are used to establish the solution of efficiency
maximization. Equation (3) is obtained by substituting the loss functions in (1) with the regression
models as given by (4). The loss coefficients, , and are functions of battery state-of-charge
(SOC). Equation (4b) is plotted in nominal values in Fig. 2 and compared to component level power
loss calculation results of different SOC.

Pp ( + 1) Pf Pf 2 Pb Pb2 = 0

(3)

Pp = Ppv Lpv

(4a)

Lf = Pf + Pf 2 +

(4b)

Lb = Pb 2

(4c)

It is assumed that the battery has negligible faradic losses and such ideal property is observed in
lithium-ion batteries [10]. Equation (5a) is the battery loss in the function of Rb and Voc, which yields
in (5b). Both Rb and Voc are also functions of SOC [11].

Rb I b 2 = Voc 2 I b 2 = Pb 2

(5a)

= Rb / Voc 2

(5b)

Efficiency Optimization
Fig. 3 depicts component level power loss calculation results at SOC = 20%, Ipv = 7A and Vpv = 30A.
The total system power loss, Ltot is the sum of Lpv, Lb and Lf. The power loss of PV power path, Lpv is
not shown here as Ppv is constant. Fig. 3 shows that minimum total power loss can be obtained by
adjusting Vb. Increasing the level of Vb effectively increases Pb and reduces Pf proportionally.
Obviously, Ltot is not linearly proportional to Pb or Pf. The Lf curve in Fig. 3 is consistent with those
shown in Fig. 2 and the regression model given by (4b). The Lb curve is on the other hand correspond
to the regression model given by (4c).
SoC=20%

SoC=100%

Eq. (4b)

14

Lf

38

39

Ltot

25

Power losses (W)

12

Feed-in power loss, Lf (W)

Lb
30

10

20

15

10

4
0

50

100

150

200

Feed-in output power, Pf (W)

Fig. 2: Feed-in power loss as a function of Pf

0
37

40

41

Battery voltage, Vb (V)

Fig. 3: Power losses at SOC = 20% and Ipv = 7A

The objective function given by (6a) is subjected to the constraint (7). Using the method of Lagrange
multiplier, , the Lagrange function for efficiency maximization is written in (8).

F = Pf + Pb

(6)

= Pp ( + 1) Pf Pf 2 Pb Pb 2 = 0

(7)

= Pf + Pb +

(8)

For a given Pp, the independent variables in (6)-(8) are Pf and Pb. Solve the partial derivatives (9)-(11)
and the solution is given by (12). One of the solutions of (12) is Pb that yields minimum Ltot.

d
= 1 ( 2 Pf + + 1) = 0
dPf

(9)

d
= 1 ( 2 Pb + 1) = 0
dPb

(10)

d
= Pp ( + 1) Pf Pf 2 Pb Pb 2 = 0
d

(11)

Pb +

Pb

4 ( Pp ) + 2 + 2
4 ( + )

=0

(12)

To verify the results in Fig. 3, the equation given by (13) is obtained by substituting for Pb in (12). It
shows that Vb that yields minimum loss is proportional to Pp. However, to implement (13) in hardware,
the currents, Ib and If need to be simultaneously measured.

Vb I b =

( + )

Pp +

2 + 2 4
4 ( + )

(13)

By assuming that the voltage drop across Rb is negligible, Vb is substituted with Voc


2 + 2 4
Voc I b
I
I
V
+
+
(
)
b f oc
4 ( + )
+

(14)

By substituting for Ib as a function of Vb in (14), the battery terminal voltage, Vb as a function of If is


obtained

Vb = Rf I f + Vf

(15)

where


Rf Rb

2 + 2 4
Vf
Rb + Voc
4Voc

(16a)
(16b)

Equation (15) gives the load-line of the second boost converter that yields minimum Ltot. It defines the
voltage droop controller where Vb is controlled to vary directly proportional to the converter current, If.
The configuration of the controller is presented in [12]. The emulated internal resistance and opencircuit voltage are Rf and Vf respectively.
Fig.s 4 and 5 verify droop controller at different levels of Ipv and SOC. The PV voltage, Vpv is fixed at
30V. The Ltot curves are plotted from the results of component level power loss calculations. To
calculate the markers of droop control in the figures, the loss coefficients are nominal values but Voc is

SOC dependent in (15). The necessity to use SOC dependent Voc is obvious by comparing Fig.s 4 and
5. The trajectory of minimum power loss Vb at a given SOC is within 1V. The operating range of Vb is
however approximately 4V.

Ipv=1
Ipv=4
Ipv=7

Ipv=2
Ipv=5
Min

Ipv=3
Ipv=6
Eq. (15)

Ipv=1
Ipv=4
Ipv=7

Ipv=2
Ipv=5
Min

Ipv=3
Ipv=6
Eq. (15)

20

31

Total power loss, Ltot (W)

Total power loss, Ltot (W)

18
26

21

16

11

16
14
12
10
8

6
37

38

39

40

6
39.5

41

Battery voltage, Vb (V)

Fig. 4: Total system power loss at SOC = 20%

40.0

40.5

41.0

41.5

Battery voltage, Vb (V)

Fig. 5: Total system power loss at SOC = 85%

Component Level Power Loss Calculation


Switching power pole is the building block of the circuits in Fig. 1 [13]. The power devices switch
complimentarily and conduct the inductor instantaneous current. The average current flowing through
the boost converter input inductor Iin can be obtained directly. For the front-end boost converter, the
input current is the PV maximum power point current, Ipv. The battery terminal voltage, Vb is stepped
between Voc and the upper voltage limit in order to scan for the minimum total power loss. Each step
yields values of Ib and If, which in turn is the input current to the second boost converter. The fullbridge inverter inductor average current, Ig is iteratively calculated from the input average current until
the power losses converge. The voltages, Vdcl and Vg are assumed constant.
The converters on the Pf path utilize IPB60R099CP power MOSFETs and C3D03060A Schottky
diodes. The full-bridge inverter utilizes bidirectional sinusoidal PWM. In order to reduce the
conduction losses, two-phase interleaved boost converter topology is implemented. Each phase of the
front-end boost converter consists of one MBRS3200 Schottky diode and two STP75N20 MOSFETs
placed in parallel. The boost converters operate at 25kHz switching frequency while the inverter
switches at 49.75kHz. The inductance of Lbs and Li is 680H while the series resistance is
approximately 200m. The battery pack is assembled from twelve parallel branches of ten lithium-ion
polymer cells in series (10S12P configuration). The empirical models for the 850mAh TCL PL383562 lithium-ion polymer battery from [11] are used in this calculation.
A common set of equations is derived for both boost converters, where Vin is the input voltage and Vout
is the output voltage. Equation (17) gives the inductor peak-to-peak current, I for boost converter,
where fsbs is the switching frequency. The MOSFET duty-cycle, dM is given by (18). The boost
converter operates in continuous conduction mode (CCM) when Iin is larger than half of I but goes
into discontinuous conduction mode (DCM) when Iin is smaller. The first element on the right-hand
side of (18) is dM in CCM while the second element is that in DCM. Similar convention is used
henceforth for boost converter equations that include both operating modes.

I =

Vin d M
Lbs f sbs

(17)

V
2 Lbs f sbs I in (Vout Vin )
d M = min 1 in ,

VoutVin
Vout

(18)

Equation (19) gives the inductor peak-to-peak current for full-bridge inverter at the kth pulse, where fsi
is the switching frequency, ma is the modulation index and k is the phase angle.

I k =

Vdc
2
1 ( ma sin k )

2 Li f si

(19)

Conduction Losses
Equation (20) gives the generic expression for the conduction loss, where Irms,X is the rms current
flowing through device X and rX is the ohmic resistance. The p-n junction voltage drop in diode is
calculated by multiplying the threshold voltage with the diode current.

Pcond , X = rX I rms , X 2

(20)

Equation (21) gives the rms inductor current for the boost converter.
I rms , Lbs

I 2 2
2
= max I in +
I in I
,
12
3

(21)

The rms currents of MOSFET and diode can be calculated by multiplying the inductor current with the
respective duty-cycle. The diode duty-cycle,

V
2 Lbs f sbs I inVin
d D = min in ,

Vout Vout (Vout Vin )

(22)

The rms currents of the full-bridge inverter inductor and MOSFETs are given by (23) [14]. The
inverter upper and lower MOSFETs switch complimentarily. It is important to note that MOSFET onresistance is operating temperature dependent and it in turn is dependent on the device power losses
and ambient temperature. Iterative numerical method is used to solve for the MOSFET on-resistance.
2

I rms , Li = 2 I rms , Mx =

1 Vdc 3 4

2
2

ma ma + 1 + I rms , g
3 4 f si Li 8

(23)

Power Device Switching Losses


For both boost converter and full-bridge inverter, MOSFETs switch on at the valley and switch off at
the peak of the inductor current. The peak inductor currents of boost converter and inverter are given
by (24) and (25) respectively. The valley of inductor current can be similarly obtained by subtracting
I/2 from the average current.
I

(24)
I p = max I in +
, I
2

I pk = 2 I g sin k +

I k
2

(25)

Analytical switching loss models for power MOSFET that includes the effect of the parasitic
interconnect inductances and nonlinear device capacitances are proposed [15], [16]. A simpler
approach is to use empirical model from the experimental results [17], [18]. For this paper, the
switching energy is expressed in a quadratic function of inductor current. The switching loss of the
MOSFETs in full-bridge inverter can be expressed in finite series as given by (26). The coefficients,
k1, k2, k3 and k4 in (26) are obtained by solving the analytical models in [15]. Equation (27) gives the
phase angle where the valley of the inductor current crosses zero in the first-half cycle.

PM , swi

4f
= si
mf

m f (1/2 + / 4 )
m f (1/2 / 4 )
2
k1I vk + k2 I vk +
k3 I pk + k4 I pk 2

k = m f / 4
k = m f /4

(26)

Vdc
,
2
8 Li f si I g + Vdc ma

= min

(27)

Other Losses
Steinmetz equation is used to calculate the hysteresis and eddy current losses. The gate switching
losses is dependent only on the gate voltage and switching frequency.

Energy Efficiency Analysis


The outdoor measurement results of a 1.71m2 poly-silicon PV module in Grimstad, Norway on July 2,
2010 (020710, summer) and January 19, 2011 (190111, winter) are shown in Fig.s 6 and 7. Results
from days with clear sky are chosen to facilitate the discussions.
Ipv

Vpv

Ipv

8.0

30

Vpv
35

5.0
4.5

7.0

4.0

25

3.0

3.5
3.0
30

2.5
2.0
1.5

PV voltage (Vdc)

4.0

PV current (Adc)

5.0

PV voltage (Vdc)

PV current (Adc)

6.0

2.0
1.0
1.0

0.5

0.0

20
7

12

17

Time (hr)

Fig. 6: PV power on July 2, 2010

0.0

25
9

11

13

15

Time (hr)

Fig. 7: PV power on January 19, 2011

MATLAB is used to calculate the power losses at the component level. Voltage droop controller is
compared to the constant-current charging (CC mode). The upper current limit for CC mode is set to
C/5, C/10 and C/20 respectively, where C denotes current that would discharge the battery in an hour.

For safety reasons, both voltage droop control and CC mode switch to constant-voltage charging (CV
mode) when Vb reaches the upper voltage limit. Similarly, to calculate Vb of droop controller, loss
coefficients are nominal values but Voc is SOC dependent in (15). SOC needs to be estimated in
hardware implementation using batter fuel gauge. The estimation error is simulated, which results in
the ripple in droop controller curves.
The extra loss, Le given by (28) is introduced as a mean of performance comparison. It defines the
deviation of total power loss of the corresponding charging method from the ideal charging given by
(13).

Le, x = Ltot, x Ltot,eq13


C/5

(28)

C/10

C/20

Eq. (15)

C/5

C/10

C/20

Eq. (15)

2.5

Battery charging current, Ib (Adc)

Battery voltage, Vb (V)

41.00

40.00

39.00

38.00

37.00

2.0

1.5

1.0

0.5

0.0
7.5

9.5

11.5

13.5

15.5

17.5

7.5

19.5

9.5

11.5

Time (hr)

Fig. 7: Battery voltage on July 2, 2010


C/5

C/10

C/20

13.5

15.5

17.5

19.5

Time (hr)

Fig. 8: Battery current on July 2, 2010

Eq. (15)

C/5

C/10

C/20

Eq. (15)

0.8

1.60

Grid feed-in current, Ig (Arms)

1.40
1.20

Extra loss, Le (W)

1.00
0.80
0.60
0.40
0.20

0.6

0.4

0.2

0.00
7.5

9.5

11.5

13.5

15.5

17.5

-0.20
-0.40

19.5
0.0
7.5

Time (hr)

Fig. 9: Extra power losses on July 2, 2010

9.5

11.5

13.5

15.5

17.5

19.5

Time (hr)

Fig. 10: Grid feed-in current on July 2, 2010

On 020710, the Le curves of CC mode at C/10 and the droop control are close to the zero. Both curves
overlap and the losses increase in the CV region. The Ib of C/10 is approximately equal to the average
value of the droop control Ib curve. As comparing to CC mode at C/10, Le curves of C/5 and C/20
deviate noticeably from zero. This verifies nonlinear relation between the total power loss and the
battery charging as postulated in Fig. 3. Fig. 10 shows droop control that scales the grid feed-in power
proportionally to PV power and results in low peak Ig.
C/5

C/10

C/20

Eq. (15)

C/5

41.50

C/10

C/20

Eq. (15)

2.5

Battery charging current, Ib (Adc)

41.00

Battery voltage, Vb (V)

40.50
40.00
39.50
39.00
38.50
38.00

2.0

1.5

1.0

0.5

37.50
37.00

0.0
9

10

11

12

13

14

15

10

11

Time (hr)

Fig. 11: Battery voltage on January 19, 2011


C/5

C/10

12

13

14

15

Time (hr)

C/20

Fig. 12: Battery current on January 19, 2011

Eq. (15)

C/5

1.60

C/10

C/20

Eq. (15)

0.5

Grid feed-in current, Ig (Arms)

1.40

Extra loss, Le (W)

1.20
1.00
0.80
0.60
0.40
0.20
0.00
9
-0.20

10

11

12

13

14

15

Time (hr)

Fig. 13: Extra power losses on January 19, 2011

0.4

0.3

0.2

0.1

0.0
9

10

11

12

13

14

15

Time (hr)

Fig. 14: Grid feed-in current on January 19, 2011

On 190111, the Le curve of CC mode at C/10 deviates from that of droop control. Contrary to voltage
droop control, CC mode is incapable of maintaining low Le as PV average power varies. The average
value of the droop control Ib curve lies approximately between those of C/10 and C/20. Comparison
between Fig.s 8 and 12, shows that the average Ib of droop control varies with the PV average power.

The battery is charged to approximately 100% SOC at C/5 but the total loss is noticeably high. The
size of the battery needs to be reduced if the objective is to fully charge the battery in winter using
droop control.

Conclusion
This paper analyzes the overall efficiency of grid-connected PV inverter with storage battery at the
system level using regression models. Voltage droop battery charge controller that simplifies the
efficiency optimization is further developed. The controller and the models are verified with the
component level power loss calculations. The calculation results using PV outdoor measurement data
show that droop control is capable of maintaining low total power loss at various levels of PV average
power. The battery charging of droop control is however directly proportional to the PV power. In
order to fully charge the battery under low irradiance conditions, the values of Rf and Vf need to be
adjusted accordingly [12]. The conventional CC-CV battery charging yields low Ltot if the upper
current limit is set close to the average charging current of droop control.

References
[1] Trends in photovoltaic applications: Survey report of selected IEA countries between 1992 and 2009,
International Energy Agency, Rep. IEA-PVPS T1- 19:2010, Aug. 2010
[2] BMU: Solar radiation installations attached to or on top of buildings, Renewable Energy Sources Act of
25 October 2008 as last amended by the Act of 11 August 2010, Germany, 2010, ch.2, sec.33
[3] C. Podewils: Still Attractive, Photon International, Jun. 2010
[4] H. Hata et al.: Study of energy management for decreasing reverse power flow from photovoltaic power
systems, IEEE-PES/IAS Conf. Sustainable Alternative Energy, Sept. 2009, Valencia, Spain.
[5] M. Braun et al.: Photovoltaic self-consumption in Germany using lithium-ion storage to increase selfconsumed photovoltaic energy, 24th European Photovoltaic Solar Energy Conf., Hamburg, Germany,
pp.3121-3127, Sept. 2009
[6] A. U. Schmiegel et al: The Sol-ion system, an integrated PV-system with lithium-ion batteries: system
performance, 25th European Photovoltaic Solar Energy Conf., Valencia, Spain, pp.3803-3805, Sept. 2010
[7] S. Wakao and N. Ueda: Investigation of the Storage Battery Station for Effective Utilization of Electric
Power Generated by PV Clusters, Conf. Rec. IEEE WCPEC 2006, vol.2, pp.2419-2422
[8] R. Hara et al., Testing the technologies, IEEE Power and Energy Mag., vol.7, no.3, pp.77-85, May-June
2009
[9] Y. Ueda et al., Analysis Results of Output Power Loss Due to the Grid Voltage Rise in Grid-Connected
Photovoltaic Power Generation Systems, IEEE Trans. Ind. Electron., vol.55, pp.2744-2751, July 2008
[10] G. M. Ehrlich: Lithium-ion batteries, Handbook of Batteries, 3rd ed. New York: McGraw-Hill, 2002, ch.35,
sec.5, pp.67-70
[11] M. Chen and G. A. Rincon-Mora: Accurate electrical battery model capable of predicting runtime and I-V
performance, IEEE Trans. Energy Convers., vol.21, no.2, pp.504-511, June 2006
[12] C. L. Nge et al.: Power management of grid-connected photovoltaic inverter with storage battery, IEEE PES
PowerTech 2011, Trondheim, Norway
[13] W. Robbins et al.: A building-block-based power electronics instructional laboratory, 33rd Annu. IEEE
PESC, vol.2, pp.467-472, 2002
[14] C. L. Nge et al.: Power loss analysis for single phase grid-connected PV inverters, IEEE INTELEC 2009,
Incheon, Korea
[15] Y. Xiao et al.: Analytical modeling and experimental evaluation of interconnect parasitic inductance on
MOSFET switching characteristics, 19th Annu. IEEE APEC, vol.1, pp.516-521, 2004
[16] Yuancheng Ren et al.: Analytical loss model of power MOSFET, IEEE Trans. Power Electron., vol.21, no.2,
pp.310-319, Mar. 2006
[17] Y. Ikeda et al.: The power loss of the PWM voltage-fed inverter, 19th Annu. IEEE PESC, pp.277-283, Apr.
1988
[18] F. Blaabjerg et al., Power losses in PWM-VSI inverter using NPT or PT IGBT devices, IEEE Trans. Power
Electron., vol.10, no.3, pp.358-367, May 1995

You might also like