Insolation values in PVoutput not accurate?

[sensij]

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.

There are multiple sources of error. Even PVWatts this time of year tends to underestimate the peak power during the day. If PVOuput is using something like the POA calculation used by PVWatts, compounded with temperature errors, stacking up to 20% doesn't seem impossible.

The same sort of error occurs in the southern hemisphere in May:

http://pvoutput.org/intraday.jsp?id=1279&sid=984&dt=20150520

[J.P.M.]

NOMB or concern with respect to PVO, but has anyone asked them and received a reply as to the algorithm used ?

As for PVWatts, my understanding is that the algorithm is similar to SAM's for array temp. est. Both use the same TMY, or SolarAnywhere data, and I suspect, but don't know for certain that PVWatts uses the HDKR model or some form of it to get an est. of P.O.A. irradiance. Sam uses that, or optionally some form of the Perez model. I do know my model and SAM are quite close with SAM using the HDKR model with not much seasonal difference. Thinking further back in my memory, SAM and something called TRNSYS get pretty close as well.

In any case, the PVWatts or SAM or any models #'s will always be off as the model's cautions and caveats clearly state.

Keeping in mind that all this is an estimate, I'm still surprised the estimates compare as favorably to one another as they seem to.

Part of what's happening for some locations might be that the actual irradiance in So. CA is more for the last 6 mo./yr. or so than the TMY models estimate. Or, the years chosen for the winter month's TMY weather modeling may have less irradiance than this recent past.

One of the things I've been putting off is comparing my Davis GHI collected at 1 min. time step to TMY data for Miramar and see what the variation might be. I've got all the Davis data for the last 25 months on 25 spreadsheets of ~ about 43,000+ rows each.

Going way out on a limb here, but something I've been thinking about for some time: Two things among many I've noticed in my data collection that may be worth mentioning: 1st, I believe my panels are rated about 4-5% conservatively. 2d, looking at well over 100 measure dates and, among other things, comparing est. (measured) array temp. and gross system efficiency (as inverter power output over P.O.A. irradiance), the coeff. of power seems to run greater than the S.P. published value of -.0038, and closer to .005 or so, with a fair degree of spread. More work needs to be done here, but, if the panels are overproducing in cooler weather = more output than the spec sheet data would predict, perhaps it might be due to the coeff. of power being greater than spec'd, with conversely warmer weather (poorer) performance being masked by the conservative power rating.

One of my goals in getting the Davis was to help eliminate the world's shortage of unprocessed data.

[sensij]

NOMB or concern with respect to PVO, but has anyone asked them and received a reply as to the algorithm used ?

I did in fact ask, but in response I got a non-commital "algorithm is in the public domain" or something like that. I'll have to see if I can find that email...

With regard to the rest... @ J.P.M. thank you for taking a pass at the suitability of using panel voltage as a proxy for cell temp. During periods of dynamic irradiance it does not work as the mppt algorithm chases the changing conditions, but under normal conditions the curves are pretty smooth. I'm getting closer to being able to extract this information through automation... another forum member recently pointed us towards github code for sniffing the inverter's communication to the portal, which contains this panel specific information. Once I've got that logging in place, a weather station is going to be even harder to resist.

[J.P.M.]

I did in fact ask, but in response I got a non-commital "algorithm is in the public domain" or something like that. I'll have to see if I can find that email...

With regard to the rest... @ J.P.M. thank you for taking a pass at the suitability of using panel voltage as a proxy for cell temp. During periods of dynamic irradiance it does not work as the mppt algorithm chases the changing conditions, but under normal conditions the curves are pretty smooth. I'm getting closer to being able to extract this information through automation... another forum member recently pointed us towards github code for sniffing the inverter's communication to the portal, which contains this panel specific information. Once I've got that logging in place, a weather station is going to be even harder to resist.

Sounds like you got blown off.

Agreed about needing near/quasi steady state irradiance. low or zero wind velocity also helps. That's why I only measure stuff on very clear days at min. incidence angle. As best as I can SWAG it, voltages and temps. under low wind and GHI irradiance that doesn't change by more than ~ 1 W/m^2 over several min. might allow me to guess an array ave. temp. +/- ~ 0.75 deg. C. by either reading the voltage off the inverter screen or measuring temps. There also seems to be some reasonable agreement between the two, with both usually coming within 1-2 deg. C. of that empirical correlation from a prior post. All of which may give some appreciation of the problems in trying to get a handle on array fouling and rates of fouling which was, and is, for me, the end goal of measuring array temps. in the first place.

[sensij]

Yeah, blown off, but I had bigger fish to fry at the time. Anyway, here's some data over the past couple days, as the weather has gotten substantially warmer.

First, the panel voltage, all 12 panels overlaid. You can see how screwy the numbers get when shade hits in the evening. Voltage for each panel is reported in 0.125 V increments.

Mid-November.GIF

Actual power is from a revenue grade meter. Everything else is calibration unknown. Temp compensated PVO estimates are made with -0.41%/deg C coefficient of power.

On 11/16, mid-day
Voltage: 28.25
Actual Power: 2400 W
PVO Estimate (Temp compensated): 2152 W
PVO Estimate: 2071 W
Wunderground Temp: 60.6 deg F = 15.9 deg C
Inverter temp: 94.6 deg F

On 11/19, mid-day
Voltage: 26.375
Actual Power: 2205 W
PVO Estimate (Temp compensated): 2049 W
PVO Estimate: 2042 W
Wunderground Temp: 75.6 deg F = 24.3 deg C
Inverter temp: 116.4 deg F

Panels are Canadian Solar CS6P-260P
NOCT = 45 deg C
Vmp @ NOCT : 27.7 V
Temp coefficient of Voc: -0.31%/deg C = -0.154 V / deg C

Even though the data sheet gives the Voc above, when I do the math on the difference in Vmp between STC (25 deg) and NOCT (45 deg), I come up with a temp coefficient of Vmp of -0.5%/deg C, or -0.136 V/deg C, which I'll use below.

A voltage of 28.5 V suggests
Tcell = (28.5 - 27.7) / (-0.136) + 45 = 39.1 deg C = 102.6 deg F

A voltage of 26.375 V suggests
Tcell = (26.375 - 27.7) / (-0.136) + 45 = 54.7 deg C = 130.2 deg F

A difference of 15.6 deg C in cell temp between those two days, at -0.41% / deg C thermal coefficient of power in the linked data sheet.
(-0.0041 * 260 * 12) * 15.6 = -200 W

Actual peak power difference was -195 W.... pretty good agreement, probably too good, given the uncertainty in some of this and the factors I've ignored. I'll need to look at more pairs of days like this to have more confidence in this approach going forward.

What does all this have to do with PVOutput's estimate? If we look at the difference in Wunderground temp between the two days, it is 8.4 deg C. 8.4 deg C * (-0.0041*260*12) = -112 W. The difference in the PVO insolation estimate between the two days was -103 W, -29 W of which is attributed to irradiance difference (from the estimates with temp compensation off), leaving only -74 W due to temperature. -74 / (-0.0041 * 8.4) = 2149 W.

From this, it looks like PVOutput is applying the temperature compensation as a continuous % of power, even as power drops due to lower irradiance. In other words, if the non-temp corrected model estimates 2150 W, the temp correction is applying the user entered coefficient to that. As far as I can tell, that is incorrect. If I'm reading correctly, the Sandia and NREL models both convert the %/deg C value to W/deg C, using the STC power as the conversion factor. This creates a fixed power loss per deg change, not a proportional loss.

So, yes, there are problems with temp compensation (no attempt to model cell temp, and misapplying the coefficient), but it is clear to me now that there are also problems with how the non-temp compensated estimate is being made. I guess I'll try asking again.

[J.P.M.]

@ Sensij:

Thank you. Just got first cup and sat down. Your stuff is interesting and has a familiar looK to it. Will be back after further perusal/thought ad some GHI/P.O.A, FWIW.

Later:

FWIW: Raw GHI at my location : 11/16/2015, 3,760 W/m^2 per day. 11/19/2015, 3757 W/m^2/day. My clear day model shows 3,927W/m^2 per day and 3,832 /m^2 per day respectively. The diff. in modeled insolation day/day is due to the higher zenith angle on 11/19.

The diff. between modeled and actual is mostly due to late afternoon shading. The rest is probably due to a difference in the published pyranometer temp. compensation and actual (my SWAG is it's a lot less/deg. than the (-)1/1500 per deg. f. away from 77 deg. F. but I use that until I get a better handle on it).

More FWIW, and not much more than anecdotal because it doesn't include all the other necessary stuff: Day long output: 11/16 = 25.39 kWh, 11/19 = 23.88 kWh. 5.232 kW array. 18.75 deg. tilt, 195.75 deg. az.

[sensij]

@J.P.M. thanks for posting your data. Just saw it now... I had been looking for a new post in the thread and didn't realize you had made an edit.

It looks like the raw GHI numbers support the idea the temperature is the major driver in the change in production between 11/16 and 11/19.

Also, if I understand correctly, your clear day model shows that the insolation change due to sun angle is 3832/3927 = 97.5%. I'd need to model that against my array orientation to estimate P.O.A. and see how close it comes to PVO's model, which shows the change in P.O.A. insolation between those days of 2042/2071 = 98.6%. Maybe there is something there that will give a clue as to what PVO is doing... my latest request for more information has been met with silence; I guess I don't donate enough.

Thinking you had chosen not to respond, I went looking for nearby weather station data that might show GHI. Found this one, not too far away. With 10 min sampling, it shows 3881 Wh/m2 on the 16th, and 3899 Wh/m2 on the 19th. The 16th had just a touch of mid-morning clouds where I'm at (and at the weather station) that impaired the daily insolation relative to a truly clear sky.

Also, and of particular interest to the data I shared, the 11:30 am irradiance on the 19th was 681 W/m2, which the peak for the day, matching the time of the peak in my data as well. On the 16th, the 11:30 am peak was 686 W/m2, less than 1% difference and for purposes of analysis, can probably be treated as equivalent. I think the equivalence of GHI in this set gives a little more support to using the voltage as cell temp proxy... still lots of uncertainty in that method, but that is one less major factor being ignored.

Another factor to consider... according to that weather station, wind was out of the NNW on both days, but much calmer on the 19th (15-20 mph gusts on the 16th, 5-10 mph gusts on the 19th). That may help contribute to the bigger difference in cell temp relative to ambient temp that was estimated/observed on those two days.

I could put this all together and see if the empirical correlation comes close to what you shared, but probably not worth it until I'm using a weather station that is reporting conditions that are actually at my array.

[J.P.M.]

@J.P.M. thanks for posting your data. Just saw it now... I had been looking for a new post in the thread and didn't realize you had made an edit.

It looks like the raw GHI numbers support the idea the temperature is the major driver in the change in production between 11/16 and 11/19.

Also, if I understand correctly, your clear day model shows that the insolation change due to sun angle is 3832/3927 = 97.5%. I'd need to model that against my array orientation to estimate P.O.A. and see how close it comes to PVO's model, which shows the change in P.O.A. insolation between those days of 2042/2071 = 98.6%. Maybe there is something there that will give a clue as to what PVO is doing... my latest request for more information has been met with silence; I guess I don't donate enough.

Thinking you had chosen not to respond, I went looking for nearby weather station data that might show GHI. Found this one, not too far away. With 10 min sampling, it shows 3881 Wh/m2 on the 16th, and 3899 Wh/m2 on the 19th. The 16th had just a touch of mid-morning clouds where I'm at (and at the weather station) that impaired the daily insolation relative to a truly clear sky.

Also, and of particular interest to the data I shared, the 11:30 am irradiance on the 19th was 681 W/m2, which the peak for the day, matching the time of the peak in my data as well. On the 16th, the 11:30 am peak was 686 W/m2, less than 1% difference and for purposes of analysis, can probably be treated as equivalent. I think the equivalence of GHI in this set gives a little more support to using the voltage as cell temp proxy... still lots of uncertainty in that method, but that is one less major factor being ignored.

Another factor to consider... according to that weather station, wind was out of the NNW on both days, but much calmer on the 19th (15-20 mph gusts on the 16th, 5-10 mph gusts on the 19th). That may help contribute to the bigger difference in cell temp relative to ambient temp that was estimated/observed on those two days.

I could put this all together and see if the empirical correlation comes close to what you shared, but probably not worth it until I'm using a weather station that is reporting conditions that are actually at my array.

You're welcome. I agree that my array output diff. 11/16/2015 - 11/19/2015 is probably/mostly due to amb. temp. diff. Winds around here were mostly light during daylight hours for those dates. Without a lot of data from measurements done as similarly as possible to one another, I'd not say anything more than of a qualitative nature about it.

You're understanding of my modeled GHI is correct. Modeled GHI is changing @ ~ 31 W/m^2 per day for the dates under consideration.

The actual data from Meso West you cite seems to show pretty good agreement with my actual data (except as noted below). In one way I'm a bit surprised they're that close. In another way, I'm not surprised given the clearness of the day which tends to make regional data a bit more uniform.

FWIW: Solar noon was ~ 11:34 hrs. P.S.T.depending on long. and actual date (and time of day if separating fly crap from pepper is important).

However, my GHI (irradiance) for 11/16/2015 @ 1130 hrs. P.S.T.was 608 W/m^2 and pretty steady at 608 - 612 W/m^2 from 1120 hrs. P.S.T to 1145 hrs., and rock steady at 612 W/m^2 from 1123 to 1143 P.S.T.on 11/19/2015., all data recorded @ 1 min. intervals.

There were no clouds at my location for either day.

The Weatherunderground site about 1 1/4 mi. from me (KCAESCON10, Boulder Knolls) recorded 612 W/m^2 @1120 hrs. and 617 W/m^2 @1135 hrs. on 11/16, and on 11/19 it was 617 W/m^2. at 1130 hrs. P.S.T.

Bottom line: All that makes me wonder what the Meso West data was measuring at 1130 hrs, P.S.T. on those dates. (unless it's from a station in central Mexico). Is that a typo maybe ? You're P.O.A. irradiance ought to be around 800-810 W/m^2 or so at 1130 hrs., P.S.T. for those dates.

My gut guess after comparing my data w/ Boulder Knolls from 2 + years is that their sensor is off level a bit in sort of a SSW direction by a degree or so, but I've not confirmed that, nor am I inclined to. Their instrument and mine are the same Davis model. Those data make me think the Meso West instrument (and your array) saw some cloud reflection @ 1130 hrs. You/I being ~ 20+ or so miles apart and you being a bit closer to the coast and closer to airports may give you a higher probability of some cloud/reflection activity.

What you may be seeing sort of first hand is some of the fun I have on a regular basis. Given how much the atmosphere can change over short distance and time, and how much greater published instrument accuracy can make readings be different between even identical instruments than what I see between my measured stuff and the Meso West data, notwithstanding what looks like those ~ 681 W/m^2 discrepanccies, I'm surprised the #'s are as close as they seem to be.

Get the Davis. Just make sure the location is good for representative data for the array and is as easily accessible as possible for solar sensor cleaning. Not much sense in measuring stuff for array fouling SWAGS and other purposes unless the sensor is cleaned before every measurement exercise.