Insolation values in PVoutput not accurate?

[wwu123]

As a number of us with realtime panel monitoring have our data being uploaded to pvoutput.org, has anyone noticed that the "insolation" values you can overlay on your actual production seem to be a bit off? Basically, that in the summer the insolation estimates are somewhat higher than you actually achieve on a bright sunny day, while in the winter the insolation estimates are somewhat lower?

For the longest time, I thought that it was because of temperature and wind effects, since the insolation can't account for decreased or increased efficiency due to panel temperature. So summer when the panels are hot, the output will be less than rated, and the opposite in winter. But now I'm looking at this week's output values, and my actual production at mid-dday is killing the insolation by better than 20%; that can't be all temperature.

Then I thought my tilt is off, but I played with a range of tilts and it doesn't account for the delta.

Finally, I'm thinking that since it's an Australian site, that maybe they don't have the correct latitudes for US zipcodes, and that they are not translating our zipcodes into the proper latitude for the insolation estimate?

[UkiwiS]

As a number of us with realtime panel monitoring have our data being uploaded to pvoutput.org, has anyone noticed that the "insolation" values you can overlay on your actual production seem to be a bit off? Basically, that in the summer the insolation estimates are somewhat higher than you actually achieve on a bright sunny day, while in the winter the insolation estimates are somewhat lower?

For the longest time, I thought that it was because of temperature and wind effects, since the insolation can't account for decreased or increased efficiency due to panel temperature. So summer when the panels are hot, the output will be less than rated, and the opposite in winter. But now I'm looking at this week's output values, and my actual production at mid-dday is killing the insolation by better than 20%; that can't be all temperature.

Then I thought my tilt is off, but I played with a range of tilts and it doesn't account for the delta.

Finally, I'm thinking that since it's an Australian site, that maybe they don't have the correct latitudes for US zipcodes, and that they are not translating our zipcodes into the proper latitude for the insolation estimate?

Perhaps ask the question here?

https://getsatisfaction.com/pvoutput/topics/insolation_data

[sensij]

I see the same thing, and as you've suggested, attribute most of it to temperature effects not accounted for in the clear sky model. This time of year, I actually see some of the same sort of divergence in the PVwatts model output as well, which generally does a better job than PVO's "insolation" of producing realistic clear sky production curves throughout the rest of the year. Actual midday production peaks are higher than the PVWatts model shows, and I haven't tracked down what the difference is. Without a weather station installed next to the array on my roof, I'm not confident I'll be able to sort out how much of the mismatch comes from POA irradiance, and how much is temperature. My *guess* is that for roof mount arrays, the low overnight temperatures and thermal mass of the house are contributing to lower than expected cell temps throughout the day, but I have done very little to test that idea yet.

[sensij]

Perhaps ask the question here?
https://getsatisfaction.com/pvoutput...nsolation_data

I tried asking BB a couple times about what clear sky model he is using, but haven't gotten a very good answer yet. You can see a post of mine in the followup thread specifically discussing temperature, although some conversation was through email as well:

https://getsatisfaction.com/pvoutput/topics/insolation_modifications

[sensij]

One clue to support that thermal effects are responsible is comparing "insolation" with and without entering a panel thermal coefficient. For example, with zero coefficient, my mid-day peak "insolation" on 11/17 is 2061 W, against 2295 W actual. With a -0.43% thermal coefficient, the "insolation" peak jumps to 2125 W, reducing the error by 27%. I think PVO is just using ambient temperature, and not making any attempt at predicting the effect on the *cell* temperature, which can amplify differences in ambient temp, especially if wind is involved.

Also, there is evidence out there that inverter temperature can affect its efficiency as well, by enough to contribute to some portion of the mismatch.

[J.P.M.]

As a number of us with realtime panel monitoring have our data being uploaded to pvoutput.org, has anyone noticed that the "insolation" values you can overlay on your actual production seem to be a bit off? Basically, that in the summer the insolation estimates are somewhat higher than you actually achieve on a bright sunny day, while in the winter the insolation estimates are somewhat lower?

For the longest time, I thought that it was because of temperature and wind effects, since the insolation can't account for decreased or increased efficiency due to panel temperature. So summer when the panels are hot, the output will be less than rated, and the opposite in winter. But now I'm looking at this week's output values, and my actual production at mid-dday is killing the insolation by better than 20%; that can't be all temperature.

Then I thought my tilt is off, but I played with a range of tilts and it doesn't account for the delta.

Finally, I'm thinking that since it's an Australian site, that maybe they don't have the correct latitudes for US zipcodes, and that they are not translating our zipcodes into the proper latitude for the insolation estimate?

Since I don't have access to the models PVO uses to estimate P.O.A irradiance, or array temps., I'd say those #'s are estimates, and I'd take an educated SWAG and say the're not much better than maybe 5 % +/- a lot in terms of accuracy on consistently clear and clam days, and, while not terrible, not worthy of the confidence folks may want to ascribe to them.

That's not a knock, just reality. The P.O.A. estimate itself is subject to a fair amount of different methodology - some methods/models being easier/harder and some being more appropriate than others for some situations and orientations. In spite of what folks think, there is not "one best" method/way to model solar irradiance, with different models often giving somewhat disparate results for different days/conditions at the same location.

I am however, pretty sure they have the correct coordinates and some information about "local" conditions from things like the Weatherunderground and NOAA sites. How that data is handled and conditioned down to a site level is often where things begin to get very uncertain. One example of many: On a sunny day, the ambient air temp. on my roof is ~ 5-15 deg. F. warmer than the air at ground level as a function of wind vector, time of day, season, etc. I'm not sure how any remote site estimate would handle that.

[Sunking]

Real Simple. In engineering we:

Measure with a Micrometer.
Mark the spot with Spray Paint
Cut it with a Fire Ax.

Good engineers as they gain experience realize it is no joke and put in some justifiable overkill.

Any questions?

[J.P.M.]

One clue to support that thermal effects are responsible is comparing "insolation" with and without entering a panel thermal coefficient. For example, with zero coefficient, my mid-day peak "insolation" on 11/17 is 2061 W, against 2295 W actual. With a -0.43% thermal coefficient, the "insolation" peak jumps to 2125 W, reducing the error by 27%. I think PVO is just using ambient temperature, and not making any attempt at predicting the effect on the *cell* temperature, which can amplify differences in ambient temp, especially if wind is involved.

Also, there is evidence out there that inverter temperature can affect its efficiency as well, by enough to contribute to some portion of the mismatch.

Lots to discuss here. More later.

[wwu123]

But again the thing is right now my actual production is 20% over the pvoutput estimate. Ambient temps are around 60 to 65F, panel temps are probably a bit higher but really not far from the design point. For this to be purely temp coefficient, panel temp would be like 40 C below the design point.

So others are seeing a Delta, but not as pronounced?

[J.P.M.]

But again the thing is right now my actual production is 20% over the pvoutput estimate. Ambient temps are around 60 to 65F, panel temps are probably a bit higher but really not far from the design point. For this to be purely temp coefficient, panel temp would be like 40 C below the design point.

So others are seeing a Delta, but not as pronounced?

Sounds like something is amiss.

BTW, under a lot of mostly common daylight conditions of sun where the incidence angle is < ~ 35-45 deg., and the array is functioning normally, a PV panel will probably be between 20 and 30 deg. C. warmer than the surrounding ambient air, which air, if that array is on a roof, is probably somewhat warmer than the air at ground level, especially with a darker roof color.

At least that's the case for my array and roof as > 100 measurements of each of 16 panel temps. have shown me. Those measured temps., averaged over the 16 panels are usually within 1-2 deg. C of a correlation that uses P.O.A irradiance and wind speed, and not too far off what an energy balance on the entire array will est. for an average array temp.

[sensij]

But again the thing is right now my actual production is 20% over the pvoutput estimate. Ambient temps are around 60 to 65F, panel temps are probably a bit higher but really not far from the design point. For this to be purely temp coefficient, panel temp would be like 40 C below the design point.

So others are seeing a Delta, but not as pronounced?

20% is high, but not ridiculously so. Look at the systems on Team San Diego. Here is one that is 15% higher than the PVO estimate today, and was 20% a couple days ago. That system has generally published believable data throughout the year.

PVOutput's "insolation" model has some fundamental flaws, which show up more severely right now than they do other times of the year. Without any documentation to go on, all I can suggest is to accept it for what it is and take some time to learn more about modeling if you want to work on generating a better estimate.

[J.P.M.]

PVOutput's "insolation" model has some fundamental flaws, which show up more severely right now than they do other times of the year. Without any documentation to go on, all I can suggest is to accept it for what it is and take some time to learn more about modeling if you want to work on generating a better estimate.

+1. I'm not familiar w/ PVO's modeling, but while the whole idea of irradiance modeling is subject to a lot of variables and interpretation, most/many models regularly and consistently get to +/- ~5-10% of actual recorded data. I suspect if PVO's est. are off more this time of year, and are consistently higher rather than lower, it may have something to do with how that model treats (or doesn't treat) the conversion of GHI to P.O.A. values.

One example: If, for example, the values used are GHI and not P.O.A values, that would (for most array orientations) lower the estimated irradiance this time of year and thus the output estimate. That would make actual output higher than the estimate.

One other thought comes to my mind, and not a knock: Is everyone's orientation correctly stated ? And/or how does the PVO model use that info ?

[sensij]

One example: If, for example, the values used are GHI and not P.O.A values, that would (for most array orientations) lower the estimated irradiance this time of year and thus the output estimate. That would make actual output higher than the estimate.

One other thought comes to my mind, and not a knock: Is everyone's orientation correctly stated ? And/or how does the PVO model use that info ?

The PVO model does a decent job of using P.O.A., and with some experience, it is pretty clear when comparing actual output to modeled "insolation" when someone has mis-entered the orientation of their array. I think its biggest weakness is in temperature compensation. Most of the year, for properly configured systems, the output tracks the "insolation" pretty well. That doesn't mean all PVO data is accurate... harder to detect system size and orientation errors certainly exist, and calibration of the actual output reported will typically have a few percent drift too. Despite that, I'm glad PVO includes the "insolation" view, because it does provide useful information as long as one is aware of its potential deficiencies. Some folks have taken the time to enter sub-arrays with different orientations, and the "insolation" view graphically separates out the potential contribution of each face throughout the day, kind of a neat feature, in my opinion. I wish that it was better documented.

[J.P.M.]

The PVO model does a decent job of using P.O.A., and with some experience, it is pretty clear when comparing actual output to modeled "insolation" when someone has mis-entered the orientation of their array. I think its biggest weakness is in temperature compensation. Most of the year, for properly configured systems, the output tracks the "insolation" pretty well. That doesn't mean all PVO data is accurate... harder to detect system size and orientation errors certainly exist, and calibration of the actual output reported will typically have a few percent drift too. Despite that, I'm glad PVO includes the "insolation" view, because it does provide useful information as long as one is aware of its potential deficiencies. Some folks have taken the time to enter sub-arrays with different orientations, and the "insolation" view graphically separates out the potential contribution of each face throughout the day, kind of a neat feature, in my opinion. I wish that it was better documented.

Perhaps the PVO model uses amb. local air temp vs. a calc'd array temp. to est. array instantaneous eff. While I suppose it's possible, I'd have some question as to whether doing so would cause a 15-20% diff. in actual to estimated output values.

I suspect being aware of the potential deficiencies is widespread enough that it can cause problems in somewhat direct proportion to the potential ubiquity of the ignorance.

FWIW, one empirical correlation I've compared to actual data on est. array temp.:

array ave. cell temp. ~~ {EXP [a+ (b*c)]}*P.O.A. + ((P.O.A/1000) *3.0) + T, amb.

Where:

a == -3.56
b == -0.075
c == wind velocity in m/sec

P.O.A == Plane Of Array irradiance, Watts/m^2.
T., amb. = amb. temp. , deg. C.

Usually what I measure as an average "instantaneous" array temp. is actually more quasi instantaneous - 16 back panel temps which are each an ave. of 4 readings/panel at somewhat random locations over the backside of each panel, taken w/ a hand held IR thermometer - twice - once over 8 min. to 2 min., and again 2 min. to 8 min. on each side of the time of min. solar incidence angle for that day. I then add the P.O.A/1000 *3.0 term to get the cell temp. The 4 minutes around min. incid. angle time is a trip to the inverter to hand record voltages, currents and in/out power from the inverter display, and a return to the roof for the second round of temp. measurements.

@ Sensij: I've used your suggestion of a year or so ago to est. array temps from voltages. FWIW, it seems to be about as good as actual temp. measurements, but tells me nothing about temp. distribution over the array. Still, using voltages seems to produce no more uncertainty than actual temps. You'll still need GHI and a way to convert to P.O.A. however.

Anyway, the empirical correlation is usually within a degree or two (C.) of what I measure.

Take what you want of the above. Scrap the rest.

[wwu123]

Perhaps the PVO model uses amb. local air temp vs. a calc'd array temp. to est. array instantaneous eff. While I suppose it's possible, I'd have some question as to whether doing so would cause a 15-20% diff. in actual to estimated output values.

In my case I feed PVO temperature data from my local Wunderground weather station (not quite so nice as your Davis Vantage Pro) at 15-min intervals - those stepwise jumps in the temp, when combined with the panel coefficient (-0.4% per deg C) caused a ripple effect in my insolation estimate. I think they're just using ambient air temp rather than estimating panel temp, in which case there's little difference from the panel design temp, and if actual panel temps are 20-30 deg C more, my output should be lower than insolation estimates that just use the 25 deg C design temp.

It's the same for all PVO systems in my area, all the actual outputs right now are quite a bit higher than the insolation values, and so far it seems for SoCal and other ones I've sampled acros the US as well. That's why I think there's a systemic error in their formulas.

My azimuth is slightly east of due south, but even if I'm off by 5-10 degrees, I've played with errors in both azimuth and tilt, they don't cause anything near this level of difference.