Comparing Production Estimates to Actual Production

[pleppik]

The winter before had way more snowfalls than this one. Maybe the installer's estimate
was based on that sort of thing. Was he just as far off in summer? Bruce Roe

I have data going back to last September, and for both September and October we overproduced by around 30%. We don't normally have snow that sticks in either month, though light snow is possible in September and common in October. With only half a year under my belt so far it's hard to draw firm conclusions, but the fact that this was a dry winter for us doesn't seem to adequately explain the overproduction.

[pleppik]

In April, we produced 773 kWh, as compared to the installer's estimated production of 631 kWh. Insolation this past month was 106% of the six-year average, so with the extra sun I would have expected 672 kWh.

Solar-Analysis-1.gif

That puts our production in April at 115% of the adjusted estimate. So with eight months of data, we have produced more power than expected every single month after adjusting for the weather and snow cover. The odds of this being a fluke are less than 0.5%. That equates to 736 kWh more power generated in the past eight months than expected.

Solar-Analysis-2.gif

I'm less clear on what it all means. It's possible that my installer intentionally underestimated my power production, and I can think of a couple good reasons why he might have wanted to.

It's also possible that there's just a lot of general conservatism in making solar estimates. Underestimating production by 5% in four different places gives an overall estimate that's 20% to 25% too low.

[J.P.M.]

In April, we produced 773 kWh, as compared to the installer's estimated production of 631 kWh. Insolation this past month was 106% of the six-year average, so with the extra sun I would have expected 672 kWh.

[ATTACH=CONFIG]6561[/ATTACH]

That puts our production in April at 115% of the adjusted estimate. So with eight months of data, we have produced more power than expected every single month after adjusting for the weather and snow cover. The odds of this being a fluke are less than 0.5%. That equates to 736 kWh more power generated in the past eight months than expected.

[ATTACH=CONFIG]6562[/ATTACH]

I'm less clear on what it all means. It's possible that my installer intentionally underestimated my power production, and I can think of a couple good reasons why he might have wanted to.

It's also possible that there's just a lot of general conservatism in making solar estimates. Underestimating production by 5% in four different places gives an overall estimate that's 20% to 25% too low.

My money would be on a combination of things with the vendor underestimate at the top of the list for a lot of reasons, starting with more money in their pocket.

Where did you get the insolation #'s ?

[pleppik]

My money would be on a combination of things with the vendor underestimate at the top of the list for a lot of reasons, starting with more money in their pocket.

I don't know, though in my particular situation the size of the array was limited by the available roof area, so there was no way to sell me a larger system than the one I bought.

Where did you get the insolation #'s ?

There's a weather station on Mesowest a few miles away which has insolation data going back to 2007. I'm using the 2007-2013 data to establish the baseline, and comparing monthly average insolation against the average of the six years of historical data. Ideally I'd like to have 20+ years of historical baseline, but I haven't found a good (free) source for both current and historical data at that level.

[J.P.M.]

I don't know, though in my particular situation the size of the array was limited by the available roof area, so there was no way to sell me a larger system than the one I bought.


In that case, looks like they sold you as large a system as would fit, not necessarily one that was not oversized for the duty. Point is, looks like they could have sized the system smaller.


There's a weather station on Mesowest a few miles away which has insolation data going back to 2007. I'm using the 2007-2013 data to establish the baseline, and comparing monthly average insolation against the average of the six years of historical data. Ideally I'd like to have 20+ years of historical baseline, but I haven't found a good (free) source for both current and historical data at that level.

Have you converted that data to P.O.A.?

[pleppik]

have you converted that data to p.o.a.?

p.o.a.?

[J.P.M.]

p.o.a.?

Plane Of Array.

Most data is reported as global horizontal = solar irradiance measured and reported as total measured insolation on a horizontal plane, direct, diffuse and reflected over a period of time. That data must them be adjusted/converted/estimated to how much solar radiation is received in the plane of the receiving device, commonly a flat surface at some non horizontal orientation.

[pleppik]

Plane Of Array.

Most data is reported as global horizontal = solar irradiance measured and reported as total measured insolation on a horizontal plane, direct, diffuse and reflected over a period of time. That data must them be adjusted/converted/estimated to how much solar radiation is received in the plane of the receiving device, commonly a flat surface at some non horizontal orientation.

Got it, thanks. The Internet Acronym Dictionary came up with something like 154 different acronyms for POA, and none seemed to make sense in this context.

No, I didn't attempt to make any adjustments for the orientation of my array vs. the insolation data. All I did was take the average insolation over the month, and compare it to the average insolation from prior years (as reported by the same station) to determine whether this April was more or less sunny than prior Aprils on average. So for example, April 2015 had 106% of the average insolation of the past several Aprils.

That said, since the production estimate for the month does take into account array orientation (as well as seasonal shading in the winter months), I think this is a reasonable first-order approximation.

[J.P.M.]

Got it, thanks. The Internet Acronym Dictionary came up with something like 154 different acronyms for POA, and none seemed to make sense in this context.

No, I didn't attempt to make any adjustments for the orientation of my array vs. the insolation data. All I did was take the average insolation over the month, and compare it to the average insolation from prior years (as reported by the same station) to determine whether this April was more or less sunny than prior Aprils on average. So for example, April 2015 had 106% of the average insolation of the past several Aprils.

That said, since the production estimate for the month does take into account array orientation (as well as seasonal shading in the winter months), I think this is a reasonable first-order approximation.

You're welcome.

Without some consideration of The P.O.A., direct vs. diffuse. vs. reflected (especially it seems for your panel specifics and surface reflectivity), I'm not sure I'd be able to justify or figure even to a first approx. But, opinions vary.

[pleppik]

You're welcome.

Without some consideration of The P.O.A., direct vs. diffuse. vs. reflected (especially it seems for your panel specifics and surface reflectivity), I'm not sure I'd be able to justify or figure even to a first approx. But, opinions vary.

My general goal here is to try to estimate whether my installer's production estimate was optimistic or pessimistic.

I recognize that I'm working with limited data, and I need to have several years of production to make a true comparison, and so forth.

But given all that, is there an alternative approach you would recommend?

What I have to work with is the monthly production estimate from the installer, actual production data from the eGauge, and weather data including insolation data from a station a few miles away going back to 2007.

[J.P.M.]

My general goal here is to try to estimate whether my installer's production estimate was optimistic or pessimistic.

I recognize that I'm working with limited data, and I need to have several years of production to make a true comparison, and so forth.

But given all that, is there an alternative approach you would recommend?

What I have to work with is the monthly production estimate from the installer, actual production data from the eGauge, and weather data including insolation data from a station a few miles away going back to 2007.

Depending on your goals, which sound more like verifying the installer's estimate than estimating equipment performance, unless you're prepared to get more involved with solar resource assessment and also how solar electric works, I'd stay with what you have and not ascribe much hope to verification of estimated performance beyond a first approx.

[sensij]

One other technique you might want to try (suggested by J.P.M. some time ago and which I have personally found useful), since you have a bunch of data, are clear sky comparisons between PVWatts (or SAM) and real life. If you can pick a couple days each month with a clear sky generation curve, and find in the PVWatts hourly output a modeled clear sky day within a couple days of the real life one, you might be able to "calibrate" the PVWatts loss factor so that the generation curves come close to matching. With a few data points a few months apart, hopefully temperature and wind effects would tend to get averaged, although fouling or degradation over time might still create some structural drift.

I would probably try to avoid clear sky days when there was a lot of snow on the ground, since the ground reflection factor will make the comparison to PVWatts more complicated.

[J.P.M.]

One other technique you might want to try (suggested by J.P.M. some time ago and which I have personally found useful), since you have a bunch of data, are clear sky comparisons between PVWatts (or SAM) and real life. If you can pick a couple days each month with a clear sky generation curve, and find in the PVWatts hourly output a modeled clear sky day within a couple days of the real life one, you might be able to "calibrate" the PVWatts loss factor so that the generation curves come close to matching. With a few data points a few months apart, hopefully temperature and wind effects would tend to get averaged, although fouling or degradation over time might still create some structural drift.

I would probably try to avoid clear sky days when there was a lot of snow on the ground, since the ground reflection factor will make the comparison to PVWatts more complicated.

Of the reasons I didn't suggest that method are the possible use of reflectors (unknown to me) sometimes used with panels of the type in Pleppik's system, and the effects of higher surrounding reflectivity due to snow, neither of which PVWatts handles well, if at all. Hence, my suggestion about solar resource research. PVWatts may not be suitable to estimate output in this situation.

[pleppik]

Of the reasons I didn't suggest that method are the possible use of reflectors (unknown to me) sometimes used with panels of the type in Pleppik's system, and the effects of higher surrounding reflectivity due to snow, neither of which PVWatts handles well, if at all. Hence, my suggestion about solar resource research. PVWatts may not be suitable to estimate output in this situation.

No reflectors in my installation, but snow does happen in my area.

Interesting thought on doing an hourly clear-day-to-clear-day comparison. I'll have to noodle on that a bit. My goal really was to just get a sense for how close the system production estimates are to reality, but the clear day comparison might be another direction to approach the question.

[J.P.M.]

No reflectors in my installation, but snow does happen in my area.

Interesting thought on doing an hourly clear-day-to-clear-day comparison. I'll have to noodle on that a bit. My goal really was to just get a sense for how close the system production estimates are to reality, but the clear day comparison might be another direction to approach the question.

From what you've reported thus far, I'd bet some form of clear day comparison will get a lot closer to reality than those vendor's estimates.