Proceedings Chapter
Reference
Evaluation of the Different Losses Involved in Two Photovoltaics Systems
SCHAUB, Patrick, MERMOUD, André, GUISAN, Olivier
Abstract
The nominal efficiency of PV-modules is around 12%. The overall efficiency of PV systems lies typically around 8-9%. The purpose of this work is to understand and explain such a difference and, consequently, identify actions which can be significant in the design and the optimization of PV systems. We have monitored in great detail, two PV systems located in the city of Geneva. One of 7.5 kW is connected through three inverters to the three phases of the grid. The other one of 2.2 kW is directly connected to trolleybus lines at a DC voltage of 600 V. All losses involved in such systems depend on local conditions and are closely correlated.
It may be difficult to evaluate each one separately. Nevertheless, by appropriate selections of data and by testing components in separate experiments, we have evaluated the different losses affecting the performances of the systems under study. These are distributed in several contributions of comparable significance; for example, in the former system:
PV-modules characteristics lower than expected, MPPT efficiency decrease with solar radiation, ohmic losses in wiring, temperature [...]
SCHAUB, Patrick, MERMOUD, André, GUISAN, Olivier. Evaluation of the Different Losses Involved in Two Photovoltaics Systems. In: Hill, Robert ; Palz, Wolfgang & Helm, P. Twelfth European Photovoltaic Solar Energy Conference . Bedford : H.S. Stephens, 1994. p.
859-862
Available at:
http://archive-ouverte.unige.ch/unige:120905
Disclaimer: layout of this document may differ from the published version.
1 / 1
12th European Photovoltaic Solar Energy Conference, Amsterdam, 11-15 April 1994
. I
Evaluation of the Different Losses Inv~lved in Two Photovoltaics Systems
. Schaub, A Mcrmoud, 0. Guisan '.
C?roup of Applied Physics, University of Geneva, , ' '
4, chcnun de Couches, CH 1231 Couches-Geneva,
Switzerla.ud~v
Tel (22) - 789 13 11, FAX (22) - 347 86 49
I
"i " '-'• l'.
. ABSTRACT; The nominal
efficien~y
of PV:-modulcs is around 12%. The overallcIBci~b~ji'~V
systems lies~yp1c~lly ar~und 8-~ Yo. The puiyo~e of tl~s work 1~ to widerstaud aud e1q1lain such a difference and, couscqueutly, 1dc11t1fy actrons which cau be significant m the design and the optimization of PV systems. We have monitored iu great detail, two PV ~stems located in the city of Geneva. One of7.5 kW is co1111ected through three inverters to 'the three phases ofth~ gnd. ~e other one of2.2 kW is directly connected to trolleybus lines at a DC voltage of600V.
All losses mvolved ill such systems depend on local conditions and arc closely correlated. It may be difficult to cvalu~tc each one separately. Nevertheless, by appropriate selections of data and by tcstiug components in separate e~e.runeuts? we have evalu~ted .the diffcrcut losses affecting the performances of the systems wider study. These are d1stnbute~ _m several contnbut1ous of comparable siguificaucc; for example, in the former system: PY-modules charactcnsllcs lowe~ t~an expected, MPPT efficiency decrease with solar radiation, ohmic losses iu wiring, temperature clfc~ts, mctdenc~ angle effects thr~ugh glass, poor inverter efficiency during operation, inverter startup near threshold, little shado~tg due. to a security fe~tcc. Spectral effects arc significant, but uot quantifiable in this frame. Altogether, the ll01ll11lal efficiency of 12.4% 1s reduced to 7.8%, i.e. an overall loss factor of 37% over the rated energy.
1. -
Introduction
Two photovoltaic systems, located in the city of Geneva, were monitored iu great dclllil (n qunrtcr of hour step time) duriug more thau one year. Ou both syStcms, the PV-pallels nrc arr:iugcd in sheds on flat roofs, with free nlr circulatiou alJ nrouud.
The Ii.est one is a pilot )ustallatiou, equipped with 42 modules in series (2.2 kW nominal, 17.9 sqm), which arc dil'cctly com1coled to the overhead lines of a trolleybus uctwork (TPG, public transports of Geueva), at a nominal voltage of 600V DC. Half of the pnuels aro of the model M55 from Arco/Sicnious (ruonocrystalline cells), al)(! half are an equivalent product, consisting of Arco cells encapsulated by tbc Swiss firm Atlantis Eucrgie AG. We measured there cssc1llinlly the global irradiation in tltc collector plnuc, voltage aud current nl the tcnni1Jnls oftl1c system, LJ1c voltage al LJ1e middle poiut between the L.wo kinds of modules, the ambient temperature and tile temperature of two irradiated cells at the midpoint of panels.
This system is extremely simplified since there is no necessity of power conditiouiug. 11tc number of panels was adjusted to fit the voltage rcquircmeut by higher temperatures in SUllllllcr.
111c scco1td system is owucd by the SIG (Services lndustricls de Gcncvc, L11c public utility of Gcucvn), and measured by our tcnm nt tbc m1ivcrsiLy of Geucva. l11e field, consisting of .126 panels Arco M55 and 16 of Atlantis, has a nominal power of7.5 kW (at standard conditious) for au aren of 6 I .2 sqm. Lt is divided into three subficlds, wit!J. 16 modules of tlircc p1111cls iu sclics each, lllld coru1ectcd to the three 1>hascs of LJ1e AC grid through th.rcc SI3000 iuvcrtcrs. These inverters operate at the maximum power 1>oi111, at a relatively low nominal voltage of 50-60V, resulting iu high currents and t11orcforc high ohmic losses.
Meteorological measurements al LJ1is second system were more compJcte: we recorded irradiatiou in boll1 the borizoutal and the panel plaucs, the diffuse horizontal compoueut (pyranomctcrs CMll), ambient temperature and wind SJ>Ccd. Eleclrilfl!I measurements included voltage and current on both the De; and AC sides of cncb phase; PC power was calculnlcd and accumulated al each mcasurcmeut sample (every 2 secouds), whcu AC powers were given by specialised devices performing true
u•1
calculation iu real-time. 111c total,
.energy was cross-checked by usual energy-co1mtcrs; overall clccttical measurement accuracy's are estimated lo be better than 0.5%.
Cells tc1111>ernturc were measured continuously al Ll1c back side of two panels, but one cannot cXJlCCt accuracies better than n few degrees Celsius (3 to 6°) from such measurements;
moreover, their reprcseutntivity over the whole field, with dill'crcnt c:1.11osurcs to wind, is very difficult to estimate.
Mensurcments aud rcs1Llts arc detailed iu the fiual reports of each 11rojccL ([I] aud [2]).
111c main objective of our stud.ies is to understand why, stnrting from panels of 12.4% uollliual efficiency specifications nt standard conditions, we obtain only average nwtual efficiencies of 9.3% for the TPG, :urd 8% for the SlG installations.
This work was petfonncd thanks to Lhc-financial support of Lhc Swiss Federal Office of Energy (OFEN/BEW), as well as SIG aud TPG conttibutious.
2. - SIG System Behaviour Analysis
2.1- Experimental results
111rough the whole analysis, we will consider the global i.11solatio11 in the panel plane as the significant and iudependcnt variable. We will then try to isolate each perturbing effect as a fimction of this variable.
We conceutrate first on the a1111ual results of one phase of the SIG (AC) system, eliminating the data of about 10 days because of the inverter breakdown.
As a raw result of the measurements, fig l shows the anuunl soJar incident energy, and LJ1c measured DC and AC energy distributions by bins of 12 W/m2• Then we plot the annual efficiency distributiou by cnlculnti.11g, for each bin, the ratio between AC energy nnd i11cidc11l energy, nonnalised to the total pauels area.
These distributions depend 011 local climate and system conditions. Allllual diffuse fraction goes up to 48% of the global irradiation in tbe collector plaue. We notice a cul of the produced AC energy up to arow1d 70 W/m2, corrcspoudiug to the sclf-consum1>tio11 of tile iuvc1tcr; and near to this region, a thresho.ld. effect which affects progressively the DC production as well as the AC one. This
will
be explained later on.2500
~ 2000 "e
"
1500 !.'.!!
"
~
1000 ~ w"
"
5000
o.s
sooo
DC Energy 4500
4000 AC Energy 3500
3000 2500 2000
\SOO 1000 SOO
o.s
~
·;;
f
ffi
0 <
10
System AC efficiency 8%
0,5
Global radiation lkW /m1) (Jtobtd fndlatlort (kW/m'J Global radia1ion {kW /m1]
Fig 1. Measurements of Incident radiation, PY DC and AC productions and global efficiency.
2.2. - Methodology
Multiple pe1turbating elfects arc closely correlated, extracting them directly from these data is not possible.
Thc1eforc we had to evaluate each of them independently by separate experiments, specific selections or physical obvious models. '!lien we established correction procedures and applied them sequentially to the annnal distributions.
TI1c final result is an ciliciency "calculated" distribution, which should correspond to the measured one. TI1is is an iterative process, where the fiual comparison of global AC efficiencies indicate the validity of our correction assumptions.
fig 2 shows the relative loss of each studied phenomenon, as a function of the incident radiation. At each step we plot a partial efficiency factor ~, bin to bin related to the previous remaining energy. Sta1tiug energy is the nominal rated energy given by the manufacturer at standard conditions
(i.c 48 panels·53W for lOOOW/m2), multiplied by the total
incident energy in each bin.
2.3.- Losses analysis
Charnctcristics default : the first contribution taken into accow1t is the quality of the PV-pauels, put together with the mismatch of the characteiistics. Characteristics of all used panels have been carefully measured in-sitn [3], and are about 8.9% lower (weighted average between Arco and Atlantis used modules) than the manufacturer specifications, with a dispersion standard deviation of 2.5%. Following PY-characteristic combination simulation results of PVSYST [4], we estimated that the mean power of the field lies at a third of the standard deviation below the average. This first loss is assumed to be indcpcndaut of the solar radiation.
MPP-loss: according to the one-diode modclisation of the cells [5], the nominal efficiency at MPP drops at lower insolatious (it behaves roughly like the log of the incident radiation). Parametrization of the global field is calculated for the average parameters of the panels, assumed to be representative of the whole field; a parallel resistance of 200 olun per panel is added to the original model. Calculated for the reference temperature of 25°, this f2 loss (by respect to 1000W/m2 operation) contributes by 5% to our data.
Shading: A security fence is shadowing some panels during pa1t of the afternoon. The corresponding loss has been estimated by comparing efficiency distributions for data selected in the time intervals when no shadowing occurs, to the global ammal ones. Titis comparison was made after ohmic losses co1Tectio11s, and with energy data normalized to a panel temperature of25°C (through the one-diode model), since this induces strong anisotropy between morning and afternoon. TI1e f3 partial efficiency has no effects ou high insolations, since shadowing docs not occur at noon; it shows a global annual contribution of2%.
Ohmic losses: the resistance is calculated from the copper resistivity of the wiring as a function of the temperature
(p = 1.68. · I
o-s
Om · (I + 6.8 Io - J
T). Viewed from the invc1tcr terminals, it is of 46.2 rn.O for each phase. TI1c (R · i') loss follows a linear coJTcction as a fw1ction of the global incident, wltich amounts to 2.5% aimually (4% at maximum power). We nrny noti~c that this value is strongly rclnted to the i11vc11cr voltage opcratiou: when doubling the voltage (six panels in sciics), the ohmic loss would be divided by four!Cells temper:1turc: TI1is contribution is estimated by a thermal conclation recorded on-site between the field efficiency and measured panel temperature, after ohmic corrections and for severe selections of data. TI1is conclatiou gives a thcnnal power factor of -0.4 % /°C, against about -0.5 %/°C according to the one-diode model. Such a difference may be partially c:1.1>laincd by the doubtful rcpresentativity of our two measurement points as compared to the real average cells temperature. TI1c annual field mean temperature, weighted by the incident insolatiou, is 35°C when weighted ambient temperature is l 9°C. Winter operation brings a gain , but annual thcnnal loss is 3.4% by respect to the standard operating conditions of25°C.
TI1c cu1ious behaviour of the £5 distribution at high insolatious is c>q1laincd by the fact that very !ugh irradiations arc observed mainly during transitory conditions, when sun rays are reflected on sun·owtding clouds, occuring under wtstable weather conditions when temperature is not so high.
Incidence angle: Fresnel reflections and transmission through the glass cover are well approximated by the so-called
"ASHRAE" model ( cf. [5], p 309), describing the attenuation as a fw1ctiou of the incidence angle from one only parameter b0.
Titough for one-glazing flat plate collectors, b0 is about 0.1, analysis of the field data are more compatible with b0=0.05, value co111irmcd by investigations with sun simulator on a single panel. TI1is low effect should be due to the very different rcflexion conditions 011 the back side of the glass, which is
"optically" connected through the EVA encapsulation to the auti-rcflcction layer of the cell. Losses are calculated separately for the direct and the diffuse components; the diffuse attenuation results of integration over the visible hemisphere, staying constant during tlic year. The f6 factor looks rather linear, and becomes close to the unity at high insolatious.
Inverter threshold: below au insolation threshold of L50 W/sqm (i.e. 300W nt the inverter input), the inverter cuts off and the f.icllf opcrnting poiut stays near to the open circuit volrnge. 11ms the DC measured energy drops, and is uo more rc1>rcsontntivc of the MJ>J> capability. ·n1c missing energy was evaluated by pcrfonning a linear extension of the MPP values between 200 mld 400 W/sqm. TI1is DC-thrcsbold loss of 3%
should be included to the overall losses involved by the inverter.
Inverter AC efficiency: detcnnined by the bin to bin ratio of AC to the remaining DC energy over the threshold, this coutril)Ution is vc1y high (17.8%), but closely related to the inverter type. Recent models nrc expected to show f.1r better performances.
. ,..
2.4. - Global balances
Finally, the overall computed cfficicucy distribution is the product ofnll above
fi,
cnlculatcd for each bin. The general shnpc is quite similar to tbe measured 011e, nud the annual result (the weighted sum of bills) gives n resulting Jossar
36.9% (ofthe 1'atcd energy), i.e. a system global e1Jicic11cy of 7.8% (of
incident energy).
Global annual losses appear on the fig. 3. ·n1c overall losses arc shared into two maiu contributions (modules pcrfonnauccs and iuvertcr operation efficiency), and 6 miuor ones of equivalent sigojficancc, iudicating that none of them could be prcfcrcncially neglected in a proper treatment.
2.5. -S11cctral effects
l11c little difference between these data , to be compared Lo the ciq1crimcntal 8% clliciCllcy, may be attributed pai1ly to the difficulties of the analysis, mainly due to the correlations between factors. Nevertheless, we did not account for spectral effects up to now.
Fig 4 shows the DC cilicicncy nftcr corrections 3,4,5 and 6, for brackets of 20% in dill'usc/global ratio. We notice that the efficiency with strong diffi1se is better thnn the one for strong direct. We have tried to extrapolate 2 cnvclopc-cu1vcs for 100% diffi1sc and 100% direct. At JOO W/ml, we observe a difference of 13 %, but no more visible after 700 W /m2•
l11ercforc, we may conclude that spectral effects seem significative at low iusolatious, and tllat spectral composition may be related to the diffuse to global components.
Nevc11hcless, tile spectral coutc1it of the diffuse is not suilicicntly dctem1incd to allow systcnu11ical conclusions. A proper cvnluntion of these cffocts imply detailed SJ!Cctrnl 111c.1surcmcnts of the incident rndiatio11, which is out of the scope of this experiment.
1 1.1
3. - Analysis of the TPG 2.2 kW DC-system
111.is system is much more s.impler tllan the previous one, since the field is directly coupled to the DC load, without power conditiowng. ldcutical methodology is ap1>lied.
Clrnrnctcristics dcfnults is worst because, on one hand there arc a hnlf of Atlantis panels, and ou the otbcr band, as all pnucls arc connected in scric, the overall mismatch effect i more 11ronounccd. Nevertheless, 1hc maxinmm 11owcr poim operation i. much less affected by the mismatch· limn the current sum ~t fixed voltage, nnd specific characteristics simulations should be performed to dctcn11i11c exactly where to allo..:atc these respective contributions.
MPP-loss is slightly better, depending on the panel parnmctri.wtion with the one-diode model: this sample of panels fits better with Rµ= I OOO
n
tl1:111 with R11=200n.
Fixed vollagc opcrtttion induces loss by respect to the MPI' cnpnbilily. 13y optimizing the number of panels, it could be reduced to about 2.5 %.
Shading of neighbour lighting sheds occurs in summer, early in the morning and late in the evc1iing, and for trees net on the direct component during some dnys in dcccmbcr and jm111ary. Ncvc1thclc s corrcspondiug losses arc only 1.3%.
Cells tcm11cralurc cll'ccl is very low siucc the field works al fixed voltngc, relatively fnr from l11c MPP towards the current part of the charnctcristics.
We notice that the kind of these losses is very different from the previous ones, mainly due to the fixed voltage operation. The non-linear nspccts of the fixed voltage complicate the analysis.
On tile basis of these measurements, a posterior optimization indicates that using 38 ARCO panels only, the global aununl system efficiency could be raised up Lo 10.6 %.
1
.
I
llil~~--WIWWWWii
Ohmic wiring loss 2.5 %
0.8 0.8
0.6 PV Characteristic default 1 0.6
v·
Inverter threshold~ and mismatch 9.8 %
t -
r- DC loss 3%...
0.4...
0.40.9
0.2 0.2
0 0.8 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 . 0.8 1 1.2 Global lrradiance G lk W /m'I Global lrradiance G lkW/m'J Global lrradiance G [kW/m'I
·~ -
1.1 Cells Temperature 3.4 % 10.8 / " ,.,_ (with resp. to 25 °C)
0.8
r
0.6
1
MPP loss 5 % 1 ~~ 0.6
~ 0.4 !£? 0.9
~~ •
OJ...
0.4• . •
Inverter operating efficiency 17.8 %0.2 0.2
•
0 0.8
·- -
0 -$'0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
Global lrradiance G lkW/m'J Global lrradiance G lkW/m'J Global lrradiance G [kW/m'I
,
1.1 Shadowing 2% 1.1
'* '° l~-
Incidence angle 3.3 %
> 8 '
1
~~
1·-
~ 0 c 6 ( Efficiency after all!2
CD...
~ '(3 <1J0.9 0.9
:::
4/
above corrections
<1J
2 7.8%
u
0.8 0.8 <( 0
0 0.2 0.4 0.6 O.B 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 0 0.2 0.4 0.6 0.8 1 1.2
Global lrradiance G lkW/m'J Global lrradiance G [kW/m'I Global lrradiance G lkW/m'I
Fig 2. Partial efficiencies decomposition as a fW1ction of the irradiation.
Loss ratios are the average annual losses referred to the previous analysis stage .
Solar inch.Jent energy at SIG
Solar incident energy at TPG
Nominal orlicioncy 12.4 %
&hared inlo:
Nominal efficiency 12.4 % shared into:
Lossoe 23~3 %
U&olul onorgy 76,7 %
flJliW vollauo operation 7.2 %
Detailed losses at SIG
Detailed lo&&es ot TPG
MPP 2,8 %
Fig 3. Annual partial loss balances for the SIG and the TPG systems.
(expressed as fractions of the panel rated nominal efficiency).
r - - - -- -- - - - -- - - - -- - - - - -
l l.5
11
10.5
~
....
10...
Cl<II
·.:;
IS <II
9.S ..
u ~ 9
8.5 0
• "' a. ... ...
I
~ J'a ll '
..
.6 .... ,,. • , • I
• I
0.2 0.4 0.6
O< D/G <0.2
• D/G = 1 -D/G=O
0.8 1.2
Global Irradiancc G lkW/11111
Fig4. Spectral analysis effects according to different Diffuse/Global ratio
-;-4. - Conclusions
We have identified and quantified a set of losses involved in two PV systems, which seem to explain their main oper~tiug characteristics. In the grid-connected· system, panels and mverter performances are the main contributors, the other ones sharing equivalent moderate significances. In the DC system, panel pe1fonnances are the dominant default, and the number of panels in sedes should be optimized. Moreover, as iu the AC system we encountered several inverter breakdowns the DC-system is much more reliable due to the absence
ofpo~er
conditioning device. 111e identification of the different loss sources and their modelization were used in the elaboration of the PY-simulation tool PVSYST
f4l.
References
[l] P. Schaub, A. Mennoud, 0. Guisan. Etude de l'installation photovoltai'que de 2.2 kW des TPG, rnpport final, GAP/OFEN, 1992.
[2] P. Schaub, A. Mermoud, 0. Guisau. Etude de l'iustallation photovoltai'quc de 7.5 kW des SIG, rap11011 final, GAP/OFEN, 1993.
[3] P. Schaub, A. Mcnnoud, 0. Guisan. PV Module Characteiistics in Real Conditions, Proc. of the 1 ltb European PV Solar Energy Conference, Montrcux, Oct.
1992, PI> 1348-1350.
[4] A. Mermoud. PVSYST : A user-frienclly software for PV- systems simulation. 12th European PV Solar Energy Conference, Amsterdam, April 1994.
[5] J.A. Dufiie and W.A. Dcckmnn, Solar Engenccriug of Tlil!mml Processes, Wiley and Sous, Second Edition, 1991.
