Announcement

Collapse
No announcement yet.

Underestimated Results of Cleaning Panels

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #46
    Originally posted by sensij View Post

    Well, my results are nothing nearly as impressive as the OP's.

    Peak power looks to be 246.3 W vs 237.5 W, which is 3.7% improvement.

    It shows up about the same way in the raw electrical data. The data are a little bit noisier, but average current from 12 pm - 1 pm was 8.95 A vs 8.64 A, about 3.5% increase.

    Voltage is not affected enough to be detectable here, about 27.3 V average for both over that hour but the individual measurements have only .125 V resolution and are noisy.

    We'll see what the end of the day brings. If I'm looking at the current data correctly, the daily energy performance will end up being something less than the performance difference at peak output, maybe 2.5% - 3% ish. Cleaning.JPG
    Short response: After measuring my array's fouling for over 3 1/2years, your results seem completely reasonable to me. More information/opinion below.

    Yesterday's standard type measurements as I do on my array resulted in an estimate of my array's fouling to cause a 4.9 % performance penalty compared to what my array for that day compared to what it would output if it were completely clean.

    The prior day's result was a 5.5 % penalty.

    The day before that was 3.94 % penalty.

    Keeping in mind that results of my fouling measurements are probably good to +/- ~ 0.75 % to 1.0%, that is, if I measure fouling at, say, 3.0%, the actual fouling will probably be something like 2.00 and 4%. or thereabouts, maybe a bit tighter than for a clean array, with a confidence level somewhere between 90% and 99 % or so. More on that variation in a minute.

    Once/yr., when the incidence angle on my array at the daily time of minimum incid. angle (when the cos(incid. angle) is >0.99 or so, from early April to June as it turns out), I clean my array every morning, rain or shine. The first cleaning in April being as thorough as possible w/ lots of water, soap, soft cloth brushing, rinsing, squeegeeing, wiping and windexing. That "defines" clean for me ( "As clean as is reasonably possible"). The subsequent every day cleanings at ~ 0630 hrs.are not as severe, consisting of a rinse, a brush with a soft cloth and another rinse, followed by a drip dry. As previously describes, I believe that method gets an array as clean as other more thorough methods to the limits of my method's ability to measure, which is why I believe that gets most all of the prior 24 hrs. accumulation off the array ( as I hope and define). Then, as with every other day with cloudless skies when I'm home and available, I measure and record my array's output as previously described. If I do that every year for enough days to get 35 or so sets of instantaneous results, and I keep everything in calibration as best as I can, I'll also get a dart throw on other things like annual performance degradation, but that's another topic. More slightly off topic:To be clear, for the rest of the year, no cleaning except rain is done. That's how I estimate the effects that rain has on array output.

    Among the many things I measure and record, for the last set of clean measurements between 04/04 and 06/21/17 I got following results:
    For array cleaned on 85 consecutive days, rain, clouds or shine.
    Cloudless skies at least 2 hours (+/- 1 hr.) around time of minimum incidence angle on 40 of those 85 days. Most of those measurement days being completely cloudless.
    Average array fouling based on measured output vs calculated clean output under the measured conditions : 0.0006, min.: -0.0079, max. 0.0084, population std. dev.:0.0038

    That is how I define a clean array, and what the subsequent year's performance is measured against..

    That negative min. fouling data is one indication of the variation or precision limits in my methods, as is the relatively high max. value of 0.0084. I also appreciate that 3 sig. figures is about all I can claim, so I carry 4 or 5 in the calcs and on the spreadsheet and truncate to 3 in the results. I show more here for information purposes only. I also report honestly, warts, embarrassments, good stuff and all, and I don't cook data - the only person I'd fool doing so is me.

    Anyway, that's a roundabout way of explaining why I believe my data for fouling penalty as a % of clean performance is no better than ~ +/-0.75% to 1.0%%. Some of that variation (And I think it's a small part), may well be day/day variation caused by dry dust that comes & goes and other things I'm not measuring, or that I am measuring but not wringing/calcing as much out of all the data as I might.

    And, overall, I believe Sensij's result so far at least, seems reasonable.

    Now, Sensij and I are about 20-25 miles apart. I'm about 20 miles from the coast. I believe he's closer, but don't know and don't necessarily want/need to. Point is, while day/day weather at either site will vary some, one to the other, (but probably not a whole lot, at least most of the time), the climates will be, and are, believe me, quite similar, at least similar enough that long term effects of climate, long term being a couple of months or so for this discussion, will be essentially the same.

    Sensij's panels have been on his roof since mid June. I'd assume they went up there new and clean. My last cleaning as was described above at was about the same time, 06/21/17. Given the closeness of the dates when arrays were known to be, or defined as clean, and given the similarity in climates and recent weather, I'd suggest that both array's were are about equally clean on 06/21. If so, and given the results I've noticed - that fouling, even on arrays very close to one another - a few hundred meters or so - can vary, and also from the variations in measurement I've observed in my method, and the possibly smaller but still likely variation in Sensij's method, I'd suggest that a difference between his latest measurement and my recent measurements of (~~0.05 - .035) = ~ 0.015 is within measurement limits for accuracy and precision. It also seems to jibe, perhaps coincidentally, with my variation of ~ +/- 0.75% to 1.0% for my method.

    While I'd respectfully suggest his results will show some of the same day/day variation mine seem to show, I'd expect his fouling estimates may be tighter in terms of day/day variation because he can get results by measuring fewer variables, that is 2, and I measure and record about 6 directly, and others indirectly, adding data scatter, and also because his method measures actual values, while my method compares actual output to output calculated as clean under the same conditions using published values that I then modify 1X/yr. during my cleaning period readings. Although my calculated values are adjusted to measured, "clean" conditions once /yr., I can't and don't expect the same precision I could get if I was able to use single panel data as would be possible if I had, for example, a SolarEdge type system. I get/measure more data to estimate more array parameters that probably have reasonable/acceptable accuracy and precision. His method gets what's probably more accuracy and precision for a lot less work by measuring 1 variable for 2 different panels, and that's not a knock or praise for either method, just a description of a the difference of 2 fit-for-purpose methods.

    I look forward to more results from Sensij. Maybe other PVOput users who have individual panel monitoring capability and access to easy cleaning of one panel will pick up the gauntlet I throw down here.

    Comment


    • #47
      Originally posted by ButchDeal View Post

      I seem to remeber others performing a test like this and the cleaning improvement was less than a week . Though that would Garry with dirt and weather, certainly a few months is enough time to bring them to their steady state.

      we will know in a few days when sensij
      FWIW, without rain, my array seems to foul at a rate such that performance decreases ~~ 0.6 -.0.8 % per week until the performance penalty is ~~ 5-6 %. After very roughly 8 -10 weeks, that rate of fouling seems to become asymptotic and steady. Local conditions will produce a lot of differences from what I think I can measure, as will weather conditions. IMO, array fouling is a non steady state process both in location and day/day variation for the same location.

      Comment


      • #48
        Originally posted by sensij View Post

        Yeah, this is dirt from the new system, panels have been on the roof since mid-June.

        Final numbers were 1.77 / 1.71 = 3.5% improvement.

        Looks like mid-day instantaneous power fouling numbers are a decent indication of overall energy impact.

        Maybe I just suck at cleaning.
        Given the small differences in results I believe I measured between various cleaning methods, as long as you know how to point a hose, I'm not sure it's possible to suck at cleaning.

        12 fouling measurements each (36 in all) on consecutive days in 2016 of 3 variations of cleaning from
        - Soap/scrub/rinse/squeegee/wipe/dist H2O rinse/windex/repeat
        - Soap/rise/wipe dry to spotless
        - Hose down and let drip and that's it

        Produced no differential performance improvement between methods that I could measure using average performance for each method. Granted that 12 trials is not a statistically valid sample size to calc a distribution, but I threw out the high and low from each one and averaged the remaining 10. Those results were part of the basis of my contention that commercial cleaning is a cost ineffective ripoff if done more than 1X/2 yrs. or so, or for extraordinary conditions or events.

        Given all the sources of possible variation that will add uncertainty to production estimates, a reasonable cleaning, especially if it involves wiping as insurance, is probably as good as necessary.

        Comment


        • #49
          Today's mid-day power data looked similar to yesterdays, but I'll probably wait to download the panel current data for a better look until I've got a few more days collected. In the meantime, here is the daily relationship between the panels' energy production, up through yesterday's result. DailyComparison.JPG
          CS6P-260P/SE3000 - http://tiny.cc/ed5ozx

          Comment


          • #50
            Originally posted by sensij View Post
            Today's mid-day power data looked similar to yesterdays, but I'll probably wait to download the panel current data for a better look until I've got a few more days collected. In the meantime, here is the daily relationship between the panels' energy production, up through yesterday's result. DailyComparison.JPG
            Thank you. Can any info on least Squares fit or mean/std. dev. for that data be easily added ?

            Also, and this is no more than a look through, but do you think 1.1.17 doesn't perform quite as well under less daylong irradiance, or perhaps it's operating at higher temps. under less irradiance for some reason(s) ?. I'm looking at the 08/15 and 08/24 ratios and the lower production on those days. FWIW, my production on those days was ~ 75% and 61 % or so, respectively of what a clear day output with the same amb. temps. would have been.

            Comment


            • #51
              Originally posted by J.P.M. View Post

              Thank you. Can any info on least Squares fit or mean/std. dev. for that data be easily added ?

              Also, and this is no more than a look through, but do you think 1.1.17 doesn't perform quite as well under less daylong irradiance, or perhaps it's operating at higher temps. under less irradiance for some reason(s) ?. I'm looking at the 08/15 and 08/24 ratios and the lower production on those days. FWIW, my production on those days was ~ 75% and 61 % or so, respectively of what a clear day output with the same amb. temps. would have been.
              I'll work on overlaying some more statistical information as we go forward. I'm about 9 mi from the coast, and suffering from clouds longer on those days than you did. I bought the mounting hardware for my Davis today, so I'll have better data coming in soon.

              Looking more closely at the 8/24/17 data, it looks like I'm running into the limits of this particular data collection technique. Energy is reported with hourly granularity, and I can't recreate the energy numbers from the instantaneous power data. Instantaneous power agrees the product of voltage and current, shown in the second chart.

              The energy performance difference on this day mostly occurred in the four points 12:00 to 15:00, with three out of the four strongly favoring 1.1.10. However, the none of the power, voltage, or current data over that period supports the same difference.

              The biggest difference in the electrical data occurred at around 15:30, with panel 1.1.7 showing a spike, but 1.1.10 didn't report at all at that time, so its lines are flat there simply from lack of data.

              More on the electrical data... from 6:46 to 18:57, there were 136 reports from 1.1.7 and 138 reports from 1.1.10.

              For 1.1.7 - Average time between the reports was 5.4 min, with a minimum of 1 min and a maximum of 15 min. (clustered around the mean, long tail)
              For 1.1.10 - Average time between the reports was 5.3 min, with a minimum of 1 min and a maximum of 13 min. (clustered around the mean, long tail)

              The timing of the reports is not predictable, and knowing when one optimizer reports doesn't tell you anything about when any other particular optimizer reports.

              I think it is likely that under partly cloudy / dynamic skies, the energy report is not capturing everything that occurred over that hour. If I can figure out a way to estimate energy production by integrating the power data (accounting for the random timing), I may be able to provide a more accurate daily generation number that smooths out the apparent outliers in the panel to panel correlation on lower generation days.

              2017-08-24-ep.JPG
              2017-08-24.JPG

              Last edited by sensij; 09-29-2017, 12:22 AM.
              CS6P-260P/SE3000 - http://tiny.cc/ed5ozx

              Comment


              • #52
                Originally posted by sensij View Post

                I may be able to provide a more accurate daily generation number that smooths out the apparent outliers in the panel to panel correlation on lower generation days.
                Ok, my first stab at recalculating the relationship between 1.1.7 and 1.1.10 using integrated power data results in 99.5% performance of 1.1.7 on 8/24, instead of the 98.4% originally charted. This is going to take some work, but probably worth it for a higher quality time series.

                Also, in light of what I'm seeing here, I'm slowly backing away from the assertion I made in an earlier thread that cloudy days are capable of inducing mismatch between panels. Although these panels are close together and not the best example to use, it looks difficult to distinguish measurement error (due to timing variation) from true panel to panel correlation breakdown on cloudy days.
                CS6P-260P/SE3000 - http://tiny.cc/ed5ozx

                Comment


                • #53
                  Originally posted by sensij View Post

                  I'll work on overlaying some more statistical information as we go forward. It sounds like I'm somewhat west of you, suffering from clouds longer on those days than you did. I bought the mounting hardware for my Davis today, so I'll have better data coming in soon.

                  Looking more closely at the 8/24/17 data, it looks like I'm running into the limits of this particular data collection technique. Energy is reported with hourly granularity, and I can't recreate the energy numbers from the instantaneous power data. Instantaneous power agrees the product of voltage and current, shown in the second chart.

                  The energy performance difference on this day mostly occurred in the four points 12:00 to 15:00, with three out of the four strongly favoring 1.1.10. However, the none of the power, voltage, or current data over that period supports the same difference.

                  The biggest difference in the electrical data occurred at around 15:30, with panel 1.1.7 showing a spike, but 1.1.10 didn't report at all at that time, so its lines are flat there simply from lack of data.

                  More on the electrical data... from 6:46 to 18:57, there were 136 reports from 1.1.7 and 138 reports from 1.1.10.

                  For 1.1.7 - Average time between the reports was 5.4 min, with a minimum of 1 min and a maximum of 15 min. (clustered around the mean, long tail)
                  For 1.1.10 - Average time between the reports was 5.3 min, with a minimum of 1 min and a maximum of 13 min. (clustered around the mean, long tail)

                  The timing of the reports is not predictable, and knowing when one optimizer reports doesn't tell you anything about when any other particular optimizer reports.

                  I think it is likely that under partly cloudy / dynamic skies, the energy report is not capturing everything that occurred over that hour. If I can figure out a way to estimate energy production by integrating the power data (accounting for the random timing), I may be able to provide a more accurate daily generation number that smooths out the apparent outliers in the panel to panel correlation on lower generation days.


                  2017-08-24-ep.JPG

                  2017-08-24.JPG
                  Thank you for the consideration and information. I (think) I understand. Lots of good info. I forgot that data capture is not completely regular. Still, a lot easier than what I do, but predictability on reporting times would be nice.

                  Yea, I'm @ ~ +1400 ft. above mean sea level and stuff burns off here about 1 - 2 hrs. sooner in the A.M. than east of the 15 around the 56.

                  On pyranometer data, or more accurately irradiance measurement, my monitor outputs V and I data to 2 decimal places every 5 minutes. After a bunch of eyeballing I believe I can use the array string current readings as a pretty good approx. for P.O.A irradiance. Long story, but if I correct the string currents for panel temp., and compare the currents ( which come off the monitor at .00 resolution and usually within 0.01 of one another, (or you) can calc. P.O.A from currents. You may want to use an average of the panels or whatever. FWIW, I'm still working on this one, but it looks like string current, after temp. correction matches up pretty well with what my method of calcing P.O.A. from GHI via modified HDKR gets, and that seems to fit all the other stuff I'm measuring along the way to an estimate of array fouling. Maybe something for your spare time. Seems that GHI to P.O.A via HDKR gets real close to what string current suggests.

                  BTW: Have you made peace w/ your HOA over weather station placement, or are they in for a surprise ? NOMB, just curious.

                  Comment


                  • #54
                    Originally posted by J.P.M. View Post

                    BTW: Have you made peace w/ your HOA over weather station placement, or are they in for a surprise ? NOMB, just curious.
                    I got a letter last week complaining that I didn't mow my lawn often enough (I've got one HOA board member across the street, and another a couple houses down), after a letter last year that I didn't water it often enough. I'm not feeling overly cooperative at the moment, so a surprise is coming. I'll try not to be obnoxious with its placement, though.
                    CS6P-260P/SE3000 - http://tiny.cc/ed5ozx

                    Comment


                    • #55
                      Originally posted by sensij View Post

                      I got a letter last week complaining that I didn't mow my lawn often enough (I've got one HOA board member across the street, and another a couple houses down), after a letter last year that I didn't water it often enough. I'm not feeling overly cooperative at the moment, so a surprise is coming. I'll try not to be obnoxious with its placement, though.
                      Get inside the tent. You'll get pissed on less. FWIW, I got on the board, actually the Arch. Rev. Comm. I initially did it to get out of the cart and help push a bit. Good neighbor thing, etc. Got to review/monitor all the PV installs and rewrite HOA PV guidelines. Good/bad fortune: Good: Got to very see up close/personal what prices/contracts/vendor interface/job quality (or lack of it) look like. Way over 100 installs and counting. Priceless experience. I learned a lot. Bottom line big take away: For most bang/$$, hands down, use established electrical contractors. Bads: Sometimes I learned more than I wanted to know; If done right, a drain on time; I've pissed off a few neighbors; Worst part: I've watched more than a few neighbors get completely screwed price and/or quality wise by vendors and be in denial about it. Bottom line for PV customers: Knowledge is power. Get some and help avoid getting screwed quite so badly.

                      Comment


                      • #56
                        Originally posted by sensij View Post

                        I got a letter last week complaining that I didn't mow my lawn often enough (I've got one HOA board member across the street, and another a couple houses down), after a letter last year that I didn't water it often enough. I'm not feeling overly cooperative at the moment, so a surprise is coming. I'll try not to be obnoxious with its placement, though.
                        just lower your blades way down.... they will stop complaining about mowing infrequently enough and less need for watering...

                        I am pretty sure I have never watered my lawn
                        OutBack FP1 w/ CS6P-250P http://bit.ly/1Sg5VNH

                        Comment


                        • #57
                          Originally posted by steveholtam View Post
                          Last week I cleaned my 22 panels after almost a year on my roof. I've read that the power difference, if any would be negligible after cleaning under normal circumstances. Here in Northern California we had epic rainfall this spring, and I figured that would get me through to this winter. I noticed though that they were looking quite dusty. After getting on the roof and looking closer, they looked even worse, and I decided that they needed cleaning.

                          Went to Lowes and got the biggest professional window squeegee they sold, an 18 inch Ettore model and the corresponding 15 foot expandable pole. I also bought a cloth window washing scrubby tool. After dragging the hose up, the normal water pressure from the hose nozzle was plenty good to wash off the dirt. And there was a great deal of dirt. More than it even appeared just by looking at them. I have hardish water in my area, so as soon as I washed the panels down a few times, I squeegeed them to remove the water and dirt. I worked from the bottom, pulling the dirt and water towards me. The whole process only took about 15 minutes to clean the entire array. Each panel was basically three swipes with the squeegee. Never needed the cloth scrubbing tool, nothing was stuck on that didn't come off with the normal water flow.

                          This was all done last Wed, the 20th. Today I took a look and my readings and the increased power after cleaning is quite noticeable, as seen in the graph below. At least 4 KWh's more per day. Worth the time for sure. So I'm guessing I'm going to start doing quarterly cleanings. Lastly, I don't live near any construction or busy roads, just the normal suburbs.

                          Steve



                          ScreenHunter_615 Sep. 25 11.56.jpg
                          Not sure how close you are to me but here is my output this month. I am 95687 and have 18 270W panels. To clean the panels I use a homemade spotless rinse set-up I built to wash my cars that makes deionized water through a DI resin filter. If you search DIY spotless car was there are lots of examples.



                          Attached Files

                          Comment


                          • #58
                            Originally posted by jmargo View Post

                            Not sure how close you are to me but here is my output this month. I am 95687 and have 18 270W panels. To clean the panels I use a homemade spotless rinse set-up I built to wash my cars that makes deionized water through a DI resin filter. If you search DIY spotless car was there are lots of examples.


                            I suppose D.I. can't hurt, but from what I've done, I couldn't measure a difference in performance after adjusting for irradiance, angle of incidence, wind and panel temps. between using D.I. or distilled H2O vs. plain old hose water. including the visible, random, and somewhat unsightly to some, hard water spots left by the hose water. Save your money.

                            Comment


                            • #59
                              I want to play along....

                              I've found two panels that are right next to each other and their Power charts seem to be identical. 1.2.7 and 1.2.8

                              https://monitoringpublic.solaredge.c...lic#/dashboard

                              At 3:45 eastern, I cleaned one of these two with a dry mop.

                              We haven't had any rain in about 2 weeks, and there's a very light coating of dust, it appears, on the panels. I'm way away from roads, down a dirt and gravel road, so the dust from that has probably collected on my panels.


                              See if you can figure out which one I cleaned!

                              I'm attaching a photo of these two panels before the cleaning.

                              EDIT: Crap, the public view site won't allow charting. I'll post the chart from these two after they get shaded this evening.
                              Attached Files
                              Last edited by NukeEngineer; 09-29-2017, 04:10 PM.
                              https://pvoutput.org/list.jsp?sid=54099

                              Comment


                              • #60
                                After adding another four days with higher temps, I can still see an improvement post cleaning. I would guess that really dirty panels like mine were will generate an additional 7 to 9 percent of energy. That last rain here was .10 inch back on June 8th. Not sure if that is enough rain to clean the panels or foul them. Otherwise, it would have been throughout April where we had 2.93 inches, and I'm assuming that cleaned them. So either they collected enough dust in five months since the rains to reduce production 7-9%, or maybe we overestimate the power of the rain cleaning and I had a full eleven months of dirt up there.

                                I'll monitor them better this winter and see how well the the rain does versus me hosing them off with a spray nozzle. And I'll do some clean / non-clean panel comparisons as well.


                                ScreenHunter_618 Sep. 29 11.38.jpg

                                Comment

                                Working...
                                X